An Intelligent Multi-Agent System using XML for Adaptive
Employment Agency Management
VINCENZO BARRILE1,*, PIERO FRANCESCO SPANO’1, EMANUELA GENOVESE1,
GABRIELE BARRILE2, GIUSEPPE MARIA MEDURI1
1Department of Civil Engineering, Energy, Environment and Materials (DICEAM),
Mediterranea University of Reggio Calabria,
Via Zehender (loc. Feo di Vito), 89124, Reggio Calabria (RC),
ITALY
2LUISS - Libera Università Internazionale degli Studi Sociali Guido Carli,
Viale Pola 12, 00198 Roma,
ITALY
*Corresponding Author
Abstract: - In recent years, e-commerce has become a significant social and cultural phenomenon. Many
organizations now offer their services online. Within this context, online recruitment services play an important
role in supporting individuals in job searches and assisting companies in finding personnel. Conversely, the
potential that AI forecasting systems offer in distinct application fields is well known using increasingly high-
performance algorithms that are particularly promising and certainly used successfully in job searches. In this
field, in general, companies insert their job offers (job proposals) into a database; individuals are supported in
their search for a job offer using a search engine based on classic Information Retrieval (IR) techniques.
Regarding this, it is presumable that the Information Retrieval techniques currently used by recruitment
services, which do not involve the use of rich user profiles, can provide a single individual with a large number
of job offers, many of which are of little interest for him. This result could cause strong dissatisfaction for the
user, and his renunciation of the use of such services. The Geomatics Laboratory, in the context of forecasting
studies in the territorial and environmental fields, is experimenting with forecasting systems also in the business
and economics fields to create a customized search engine, proposing a Recommender System developed using
intelligent agents and based on XML. This system leverages detailed user profiles to assist users in
personalized job offer searches, which combines classic Information Retrieval techniques with User Modeling
techniques and artificial intelligence techniques. As a result, in generic domains, our system quickly achieves
satisfactory results; however, it struggles to maintain this performance after many sessions. Conversely, in
specialized domains, while our system requires many sessions to reach satisfactory performance initially, it
consistently delivers outstanding performance once this phase is complete.
Key-Words: - Recommender-System, Information-Retrieval, User-Modeling, Multi-Agent, XML, Artificial
Intelligence, Forecasting Systems.
Received: March 5, 2024. Revised: August 2, 2024. Accepted: August 27, 2024. Published: September 30, 2024.
1 Introduction
Some approaches based on Recommender Systems
have been proposed in the scientific literature to
build personalized search engines capable of acting
in different application contexts, such as web
browsing, e-commerce, e-learning, and so on, [1],
[2], [3], [4], [5], [6], [7]. Furthermore, in some of
these proposals, Recommender Systems have been
combined with intelligent agent technology to make
them autonomous, reactive, and proactive. The
combination of artificial intelligence (AI)
technologies with organizational practices has seen
a significant rise, encompassing various domains
from job recruitment to social network interactions.
The rapid advancement and adoption of AI,
particularly generative AI tools, necessitate a
comprehensive understanding of their implications
on workplace dynamics, employee experiences, and
job designs. While some view AI as a catalyst for
enhancing efficiency and productivity, others
express concerns regarding its impact on human
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1945
Volume 21, 2024
workers. In this field, it is important to focus on
three different areas of interest: the impact that
artificial intelligence is having today on job search
and job experience, the role of AI within social
networks to search for work, and the integration of
this technology within staff recruitment processes.
The use of AI in organizations has been studied on
multiple levels, highlighting issues such as
collaboration between humans and artificial
intelligence, the perception of the capabilities of
algorithms compared to human ones, the attitude of
workers towards artificial intelligence, the use of AI
as a control tool, and the broader market
implications of the use of AI in the workplace.
These themes highlight the complex interaction
between AI and humans, offering insights into both
the potential benefits, advantages, and challenges of
integrating AI into various level contexts.
In the last years, we assisted to the emergence
of social networks as tools for employment and
business-related user interactions leading to a
significant influence of artificial intelligence in this
field. As users increasingly depend on instant
messaging applications, and specialized social
networks for job searching, recommendation
systems based on AI have become fundamental.
Evaluations of such systems demonstrate high
acceptance rates, underscoring the effectiveness of
AI in managing large volumes of content and
facilitating user engagement, [8], [9]. Moreover, the
adoption of AI in recruitment processes has
revolutionized talent acquisition, reducing
complexities in candidate sourcing, screening, and
evaluation. AI technologies not only streamline
administrative tasks but also promise to improve
hiring quality and mitigate human biases. The
incorporation of AI into e-recruitment processes
fosters a more efficient, cost-effective, and user-
friendly candidate experience, ultimately providing
organizations with a sustainable competitive
advantage, [10]. The only Recommender System
existing in the literature and capable of supporting
online recruitment services is CASPER. This system
operates in the following way: Once a user has been
assigned, it classifies the job offers in relation to his
needs and suggests job proposals based on his past
behavior (in particular, it suggests proposals like
those that he/she/it she has accepted in the past and
avoids suggesting proposals similar to those that
he/she has rejected previously), [11], [12].
The objective of this article is to advance the
state of the art by updating a previously
implemented initial work and improving its results;
the aim is to propose a Recommender System
created through intelligent agents and based on
XML; the proposed system uses quite detailed user
profiles to support the user in the personalized
search for job offers. The system operates as
follows: each user is associated with a User Agent
(UA) that manages her profile. The system also
features a Recruitment Agent (RA) which supports
the various User Agents in selecting job offers
relating to the corresponding users. This activity is
carried out considering the user's past behaviors
(suitably stored in her profile) and those interests
not considered by him/her in the past, but which
appear to be potentially interesting for him/her in
the future. The selection of job offers is performed
by means of an item-based recommendation
algorithm. There are three elements that
differentiate our approach from CASPER. First, our
system is based on intelligent agents. The use of
agents allows for improved autonomy, reactivity,
and proactivity; furthermore, it allows workloads to
be easily partitioned between the different entities
involved, which significantly improves its
scalability. Finally, intelligent agents simplify the
conduction of existing legacy systems thanks to the
definition of appropriate wrappers; consequently,
our system can easily interact with existing online
recruitment sites, and this helps to improve its
accuracy, [13], [14], [15], [16], [17].
Secondly, the work proposal selection
algorithm, which is the basis of our system, allows
us to select the proposals considering not only the
user's past behavior but also some proposals that he
has not taken into consideration in the past, but
which appear to be potentially interesting for him in
the future, [18], [19], [20], [21], [22], [23].
Finally, our system is based on XML. XML is a
standard for representing and exchanging data.
XML combines the capability to represent data
(typical of HTML) and the capability to manage
data (typical of databases). An XML information
source consists of a simple textual document;
consequently, it is light and versatile, can be easily
exchanged, and can be stored on various hardware
and software platforms. Despite the simplicity of the
language, the information representation rules
present in XML are powerful enough to allow the
conduction of sophisticated queries, [24], [25], [26],
[27], [28], [29], [30], [31].
So, the objective of this paper is to demonstrate
how this system can improve accuracy, scalability,
and interaction with existing recruitment sites,
providing advanced and personalized support to
users in their job search. The research employs a
combination of intelligent agent technology and
XML-based data management to develop a robust
and scalable recommender system. The proposed
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1946
Volume 21, 2024
methods have been identified to make the most of
the autonomy and reactivity of intelligent agents and
the versatility and interoperability of the XML
system. The selection of them is justified by the
necessity for a highly scalable, efficient, and easy-
to-integrate system that can communicate with
online recruitment platforms already present online.
The main scientific hypotheses are: the integration
of intelligent agents in the recommender system will
significantly improve the system's scalability and
accuracy compared to existing systems like
CASPER; using XML for data representation and
exchange will enhance the system's ability to
manage complex queries and improve
interoperability with existing platforms; the
inclusion of proposals not previously considered by
the user but potentially interesting will increase user
engagement and satisfaction with the job search
process, [32], [33], [34].
2 Theory and Methodology
2.1 Intelligent Agent
Oriented Agents or Intelligent Agents represent one
of the major innovations in the field of software
engineering and Artificial Intelligence. In this
article, Intelligent Agents and the agent-oriented
programming paradigm are introduced by
examining the distinct types of agents and the main
purposes for which they were designed. An agent,
software or robotic, is characterized by the ability to
perceive its environment through sensors and act in
the environment through actuators; it incorporates
learning, reasoning, and planning capabilities and
performs tasks for the benefit of a user or another
agent. The presence of multiple agents, involved in
a common mission, within the same physical or
virtual environment, requires addressing
coordination and collaboration issues. In particular,
in highly dynamic environments, each agent can
benefit from the use of models that describe the
evolution, in space and time, of the environment and
the state of the other agents (context). Maintaining
these models can be further complicated by the
incompleteness and inaccuracy of sensory data. An
agent is uniquely defined by a PAGE, an acronym
for Precept, Action, Goals, and Environment.
Percept: the agent must be able to collect data
from the environment around it, know where it
is, and how to move within the environment
itself
Action: what an agent can perform.
Goals: objectives that the agent intends to
pursue
Environment: environment from which it draws
data and in which it acts.
We can define an agent as a set of software and
hardware components that act to achieve objectives
established by the user. In reality, the term agent is
used to indicate a broad class of research and
development projects; the same researchers, even in
the field of Artificial Intelligence, have not yet
defined general guidelines to which the design of a
system of agents must comply.
2.2 Agent’s Characteristics
An agent has crucial capabilities contributing to its
effectiveness across various tasks. Communication
stands as a fundamental attribute, allowing the agent
to interact with other agents within the system and
users alike. This proficiency facilitates information
exchange, instruction reception, and the provision of
guidance or assistance, as necessary. An agent is
also distinguished by its capacity for autonomous
action. Unlike passive advisors, it can independently
execute a diverse array of operations, ranging from
basic tasks to complex actions.
The versatility of an intelligent agent allows it
to anticipate and satisfy users' needs and wants
without the need for direct intervention. A
fundamental aspect of these agents is their
autonomy, which allows them to operate alone and
independently, make decisions, and carry out tasks
without the need for continuous supervision by the
user. This autonomy translates into quick and
efficient responses, allowing the agent to quickly
adapt to changing circumstances and improve
outcomes. Furthermore, an intelligent agent uses the
accumulated experience to improve decision-
making processes, thanks to the analysis of data and
feedback from past interactions with which it can
obtain valuable information on user preferences,
behaviors, and patterns. This allows it to guarantee
personalized advice, anticipate needs, and offer
tailor-made solutions that meet user expectations.
The main characteristics of an intelligent agent
include adaptability and the ability to learn in a fast
and personalized way. Despite the various
definitions proposed by experts, there is a
unanimous consensus on some essential properties
to classify a software entity as an agent:
- Autonomy: This is an important property,
rooted in the very concept of the agent since its
birth. Both the user and the machine can initiate
actions and monitor events. Software that
provides this type of human-computer
interaction is called autonomous. Autonomous
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1947
Volume 21, 2024
agents use their knowledge of users' needs and
interests to perform repetitive tasks. This quality
is a difficult property to obtain, especially
because it is strongly dependent on the
characteristics of the content in which the
program is operating. For human agents, the
level of autonomy is clearly defined by laws,
customs, or conventions. Indeed, some kind of
agents vary their degree of autonomy based on a
series of factors. Related to the concept of
autonomy is the concept of proactivity.
- Adaptivity: adaptive agents must be able to
learn through interaction with the external
environment. This can include the physical
world, human users, other agents, or the
Internet. For this reason, adaptive agents are
often called learning agents. The qualities
necessary for an adaptive agent are:
- Reactivity: the agent perceives its environment
and responds promptly to the changes that have
occurred.
- Social ability: the agent interacts with other
agents through distinct types of communication
languages. The purpose of an adaptive agent is
to adapt to the user, i.e., to learn the user's
preferences. Adaptive behavior can be obtained
in numerous ways, the simplest of which is to
consider a set of users who perform the same
operations and deduce their own behavior.
- Collaborative behavior: it is a typical
characteristic of a set of agents, each of whom is
assigned a specific task, but who collaborate in
the pursuit of a common goal; they are often
performed in parallel. Common problems in this
context are establishing which agent must
perform a certain task, integrating the
information collected, and presenting it to the
user. Agents must be able to communicate with
other agents via appropriate communication
languages, and being able to share a knowledge
base is a crucial point in building cooperative
agent systems. Since knowledge is shared, it
must be represented in such a way that all
agents can access it. The solution adopted is a
representation conceptualized through
ontologies.
- Mobility: refers to the ability of agents to
migrate in a “self-directed” manner from one
host to another in a network, to fulfill their as-
signed tasks. This concerns finding information
about each host, balancing traffic in the
network, and so on. Mobility can be considered
as an extension of the original concept of
autonomy.
Combining all these definitions we can define
intelligent agents of robots or software programs
that are autonomous, adaptive, collaborative, and
mobile.
2.3 An Ideal Internet Agent
To be considered optimal, an agent needs a set of
specific capabilities. It must exhibit autonomy and
adaptability, alongside the ability to interact
effectively with users. This interaction necessitates a
shared language with common vocabulary and
syntactic structures. The primary challenge with
existing search engines lies in their linguistic nature
without a deep understanding of specific domains.
This issue can be addressed by employing the
concept of 'ontology' in Artificial Intelligence.
Ontology formalizes knowledge by structuring it
through class and subclass relationships. An ideal
agent should demonstrate competence, enabling it to
make informed decisions and actively provide
services rather than merely offering advice.
However, the imperative of user privacy often
clashes with the conventional structure of the
Internet, which typically requires access to personal
data for specific services.
2.4 Programming of Intelligent Agent
An agent, from a legal point of view, is someone
authorized to act in someone else's interest; an
intelligent agent can learn the behavior of a user by
watching the actions and transactions performed by
him, in particular by observing his most repetitive
behaviors. Intelligent agents are programs that
function as a filter between the network user and the
information arriving from them. In particular, they
attempt to pick out the data that their respective user
needs and, after having memorized the choices
made by him over time, they will select, based on
his profile, the information that best suits them.
Each agent has a history, called a perception
sequence, which keeps track of everything the agent
has perceived through its sensors. For every possible
sequence of perceptions, an ideal rational agent will
perform the action that maximizes its performance.
Once we have established that an agent's behavior
relies only on its sequence of perceptions, we can
characterize a particular agent by making a series of
expected actions in response to each sequence of
perceptions. In most cases, it is an exceedingly long
table, not to say infinite, unless a limit is imposed a
priori on the size. This list is called a mapping. We
can uniquely describe an agent through mapping by
trying out all possible sequences of perceptions and
recording the actions in response. The problem is:
how to build a program to implement perception-to-
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1948
Volume 21, 2024
action mapping? This amounts to establishing how
agents should use the data at their disposal to decide
how to act.
There are four distinct categories within agent
programs, each with its own approach to decision-
making and action execution. Simple reflex agents
operate according to a set of predefined rules. When
they find themselves in a particular situation, they
consult their repertoire of rules and, having found
the appropriate rule, perform the corresponding
action without further reflection. More complex
than reflected agents are world-tracking agents. In
addition to reacting to immediate stimuli, they
maintain an internal state that allows them to adapt
based on past experiences and the current state of
the environment. By integrating past observations
and feedback, this category of agents can make
more refined decisions and change their actions
accordingly.
Goal-based agents operate with specific goals in
mind. Unlike reflex agents, which react only to
immediate stimuli, these agents evaluate potential
actions based on how much they contribute to
achieving their goals. This approach allows them to
make both reactive and proactive decisions, working
towards achieving pre-defined goals. Utility-based
agents use a utility function to evaluate the
desirability of various states or sequences of states.
They assign a numerical value (utility) to each state,
reflecting how well it satisfies the agent's goals. By
quantifying the desirability of different outcomes,
they can navigate complex situations by resolving
conflicts between competing goals and selecting
actions that lead to the most promising outcomes.
These four types of agent programs offer distinct
strategies for decision-making and action execution.
Each approach has its strengths and limitations,
shaping the behavior of agents in diverse ways and
enabling them to navigate diverse environments
effectively.
2.5 Framework of Intelligent Agent
An Agent Program is a function that implements
agent mapping based on perceived actions. An agent
program can run on some architecture, hardware, or
software; the architecture must provide a degree of
isolation between the actual computer and the agent
program so that high-level programming is possible.
In general, the architecture makes the data perceived
by the sensors available to the agent program,
executes the agent program, and transmits the
actions decided by it to the actuators. The main
differences between programs and agents are that
the program is: static, the actions are initiated by the
user, non-interactive (the dialogue is written a
priori), non-persistent (must be called to run again,
once stopped), predictable (does what is told),
follows instruction and stay in one place; while the
agent is: dynamic, its actions can be initiated both
by user or by agent system, can interact with users
or other agents, is able to learn and to adapt, is
persistent, is able to interpret what is means instead
of what is said, can initiate action and is able to
move through other servers.
2.6 Reasoning Techniques for Agents
It has been seen that the agents must be able to act
autonomously and provide the user with the
requested response. For this purpose, there are
different reasoning techniques that are used.
Rule-based reasoning techniques: serve as the
foundation for inference engines, allowing users to
define rules or agent systems, often termed
production rules in Artificial Intelligence. However,
two significant challenges arise: firstly, users need
to manually maintain data since these systems lack
autonomous adaptation; secondly, the expansion or
modification of the rule set may lead to conflicts,
rendering the agent incapable of resolving them.
Knowledge-based reasoning techniques: These
techniques, tailored to specific domains, underpin
inference mechanisms, especially those utilizing
rule-based approaches. Developers furnish agents
with domain-specific task information, allowing
them to deduce suitable responses based on their
knowledge when confronted with particular
situations. The main problem with these systems is
that they require a large amount of work required.
The best example in this application area is Cyc, a
formalized representation of a vast amount of
common human knowledge. This knowledge base
can be used to build agents, each of which shares a
common core of knowledge.
Simple statistical analysis: the simplest learning
technique that an intelligent agent can use is
statistical analysis. It can be used to determine the
temporal correlation, if any, between events of
interest.
Fuzzy agents: when an agent must work with
partial and perhaps imprecise information, a useful
tool is fuzzy logic; this is a form of logic used in
various ex-pert systems in which variables can be
classified based on truth or falsity values
represented by values between 0(true) and 1(false).
Evolutionary Neural Networks: neural networks
consist of a series of interconnected nodes, each
with its own weight. The most popular type of
neural networks are feed-forward networks; these
consist of an input layer, an output layer, and a
hidden layer; each of these layers consists of one or
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1949
Volume 21, 2024
more processing units (neurons); each unit in a
given layer connects all units in neighboring layers
but not those in the same layer; each unit receives
input from the units of the lower layer and sends
output to the units of the upper layer. Weights are
assigned by neural networks to connections between
units, signifying the strength of the link. Agents that
simply use neural networks, however, are limited to
learning locally from certain documents or web
pages. An agent's ability to explore is nevertheless
restricted by the vast amount of information
available online.
2.7 Environments
Agents operate in different environments, whose
particular characteristics significantly influence their
behavior and the decision-making processes that
involve them. These environments can be identified
and classified in distinct ways, highlighting the
particularly complex dynamics that come into play
in agent interactions with the surrounding
environment. A fundamental aspect of
environmental classification is the concept of
accessibility, which refers to the presence of the
agent's sensors that can identify, receive, and collect
all aspects inherent to the choice of a certain action.
In contrast, an inaccessible environment presents
complications that influence the action of the agent's
sensors, making it necessary to maintain the internal
state for informed decision-making. Furthermore, it
is important to underline how the quality of the
action taken by the agent within each episode
depends entirely on the characteristics of that
particular episode, while subsequent episodes are
completely independent of past actions. Non-
deterministic environments, in contrast, introduce an
element of randomness or uncertainty into the
outcome of actions, making decision-making a more
complex effort. The quality of the action within an
episodic setting is determined solely by the unique
characteristics of each episode, unprovided of any
influence from previous episodes. In contrast, non-
episodic environments have actions potentially
having long-term consequences that are independent
of individual episodes. Environments can then be
classified as a function of dynamics: dynamic
environments are characterized by their ability to
submit alterations or transformations, introducing
additional levels of complexity for agents as they
navigate evolving circumstances. In contrast, static
environments remain constant throughout the
decision-making process, offering a more stable and
predictable scenario. Semi-dynamic environments,
on the other hand, remain overall unchanged but
highlight variations in the agent's performance over
time. Finally, environments can be classified as
discrete or continuous, based on the nature of the
data and actions available within them. Discrete
environments present an established set of
perceptible data and actions. Instead, continuous
environments offer infinite possibilities. Therefore,
the classification of environments along these
various dimensions provides valuable insights into
the challenges and opportunities that agents
encounter in their respective domains. By
understanding the shaded characteristics of different
environments, researchers and practitioners can
develop more effective agent architectures and
algorithms tailored to navigate and thrive within
specific environmental contexts.
2.8 Types of Use of Agents
Limiting ourselves to considering the case in which
the environments are made up of computer
networks, distinct types of use of the agents are
observed.
•A first model was born as an evolution of the
client-server architecture.
It can be “injecting” an agent into the network to
perform a certain task, regardless of whether the
connection with the client is interrupted. Injected
agents navigate the network directly to service
nodes, transforming client-server communication
into local interactions. This approach minimizes
network traffic and reduces the user's need for
extensive service knowledge. Agents adopt a
metaphor akin to market operations: visiting, using a
service, and departing. Equipped with intelligence,
these agents handle demanding tasks for users,
configuring communication environments based on
individual preferences. Users train agents to perform
searches on their behalf by providing sample texts.
The trained intelligent agent autonomously selects
documents and web pages based on learned
information. In the complex network landscape,
using background agents to retrieve information
aligns with a qualitative evolution. Unlike client-
server relationships requiring numerous data
transfers, agents, when injected, locally connect to
the server, streamlining the retrieval of client-
relevant data.
• Another agent model is used to automate
repetitive tasks and learn user habits, acting as an
intermediary to facilitate the execution of tasks
within work processes. The greatest advantage of
these agents is represented by the possibility of
operating on behalf of the user even when he is not
connected.
• An agent can function as a front-end for
accessing a particular service. The provider must
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1950
Volume 21, 2024
consider how potential users will use the new
service when defining it. To interface with the
service, the user needs to install a particular client
on his node or use a generic browser. In the latter
case, the server provider does not have the
possibility of providing any added value to enrich
and personalize the offer. To overcome this
inconvenience, the service access front-end is
defined by the service provider as an intelligent
agent that migrates to the client node following a
service request and which contains the logic
necessary for interaction with the service itself. As
we have just seen, the term "intelligent agent" is
used with an overly broad spectrum of meanings:
from adaptable user interfaces to communities of
processes cooperating to achieve a common goal.
For this reason, it becomes difficult to give a precise
and exhaustive definition of the term.
As regards data collection and procedures, the
data was collected through user interactions with the
recommendation system. This includes accessing
the system, searching for job offers, viewing
recommended proposals, and interacting with
graphical interfaces. Controlled experiments were
conducted to evaluate the effectiveness of the
recommendation system under different conditions
or algorithm variations. Experimental conditions
have included variables such as system usage
frequency and user preferences.
As regards the conducted experiments, a
representative sample of voluntary users with
diverse professional backgrounds was selected.
Subsequently, the database was populated with a
significant number of job offers, selecting 1000
offers from www.jobpilot.it. The participants then
took part in various sessions: 30-minute training
sessions to explain the system and answer questions;
15-20 minutes profile collection sessions to allow
participants to enter their data; 1-hour query
execution sessions where participants could execute
their queries, evaluate the offers, and provide
feedback; 30-minute feedback and discussion
sessions to gather qualitative feedback and discuss
any issues encountered. Selection and interaction
biases were managed to ensure that the sample of
participants was representative and to standardize
the training session for all participants so that
everyone had the same level of understanding of the
system.
3 General Architecture
The system consists of two main types of agents: a
User Agent (UA), which manages the user's profile
and its interaction with the system, and a
Recruitment Agent (RA), which manages job offers.
Finally, the system uses a Job Proposal Database
(JPD) which stores data relating to job proposals. In
the JPD, it can insertable inserted, remove or change
the job offers with a precise Interface Agent. In
addition, JPD can be automatically enriched by
appropriate Wrapper Agents (like the one described
which continuously `monitor' existing online
recruitment sites, ex-tract new job proposals, and
store them in JPD (if such proposals are not already
present). JPD can be described by means of an XML
document whose:
• JIDl is an identification code;
• JTopicSetl = {JTopicl1, JTopicl2, . . .,
JTopiclm} is a set of `topics' describing JPl
• JCharacteristicSetl = {JCharacteristicl1,
JCharacteristicl2, . . ., JCharacteristiclp} is a set of
characteristics associated with JPl. Each feature is
described by a pair feature, shown in Table 1.
Table 1. The possible characteristics associated
with a job offer
Characteristic
Description
Characteristic
Description
Salary
The
associate
salary to job
offers.
Mobility
Indicates
whether the
job offer
requires the
candidate to
move to
another
location
Residence
The city of
job offers.
Experiences
years of
experience
requested
Foreign
languages
Foreign
language
required of
candidates.
Academy
academic
qualification
s required of
the candidate
Ability
Skills
required of
candidates
Type of job offer
Full-Time /
Part Time
job
In the recommendation system, the various
agents play distinct roles, interacting to facilitate the
personalized job recommendation process
seamlessly. The User Agent (UA) serves as the
intermediary between users and the system,
managing user profiles and interactions. It collects
user preferences and past behavior, communicating
them to other system components. The Recruitment
Agent (RA) is tasked with managing job offers,
evaluating postings, and recommending relevant
options based on user profiles. Using algorithms, it
needs job ads from the Job Proposal Database (JPD)
and gradually improves its recommendations. The
Interface Agent, connecting the External subjects
and the system, simplifies the process of addition,
removal, and adaptation of jobs’ offers inside the
JPD; its incorporation with external origins and the
consequent database update guarantees the integrity
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1951
Volume 21, 2024
and the relevance of the data. The connections that
occur between the agents are essential for the good
working of the entire system and the process is the
following: the users’ request is sent to the RA, this
one enters inside the JPD looking for relevant posts.
At the same time, interaction occurs between the
Interface Agent and the external authorities updating
JPD. At the end of the process, through the User
feedback collection, operated by UA, there are
refines in profiles and better recommendations. In
other words, these agents work together to provide a
clear, customized job suggestion experience,
increasing user engagement and happiness.
More in detail (Figure 1), at the beginning of the
process, the operator (called “ui user”) exposes its
will to explore the eventual job’s offers, sending an
inquiry to their assigned User Agent, denoted as
"Uai". Once received the user’s request, the User
Agent (Uai) starts to communicate with the system's
Recommendation Agent (RA). In this phase, the
User Agent, Uai transfers the user’s question,
correlated with the pertinent details, collected from
the user’s profile, allowing the foundation for the
recommendation process. During the process, one of
the most important functions of the system is the
Recommendation Agent (RA), which is responsible
for sorting through the job proposals stored in the
Job Proposal Database (JPD). Using an advanced
classification algorithm, RA determines if
employment offers are appropriate by determining
how well they match the user's profile. This
judgment is based on a huge number of factors, such
as the user's past actions and projected areas of
interest, giving an accurate selection of job
proposals related to the user’s inclination and
professional goals. These selected suggestions are
then retransmitted back to the user agent (Uai) for
submission to the user (ui), aiding informed
decision-making. Once the proposals have been
obtained, the Uai interacts with the user, showing
him the personalized selection of work proposals to
be evaluated and considered. Having a holistic view,
the user can carefully evaluate each work proposal
selecting the options that best fit his preferences and
goals. Once the choice is complete, Uai merges the
user’s selection into their profile, keeping the
system updated with the latest preferences and
decisions. The system's recommendation
capabilities are improved through this iterative
profile refining process, which can lead to more
customized and individualized suggestions. In the
described process, the Naive Bayes classifier is the
algorithm used for classification. When it is
necessary to classify words or documents, such as
user profiles and job advertisements like those that
we have used in the case study, this kind of
algorithm works especially well. Based on Bayes'
theorem, the Naive Bayes classifier starts from the
hypothesis that characteristics are conditionally
independent of each other. In this case, the system
was trained using a predictive variable, that is the
characteristics of the user profiles and the job
advertising, the aim was to assess the probability
that a specific user is attracted to a specific job
offer. Naïve Bayes classifier was applied to forecast
the likelihood of a user being attracted to new job
offers, according to offers’ features and historical
conduct. In our opinion, according to the scientific
literature, the Naive Bayes classifier shows good
computationally efficient properties, and it is easy to
adapt to handle a large volume of data, making it
appropriate for systems that work with large
amounts of data and job offers are processed.
The use of XML in this system is an integral
component, performing an important role in aiding
the exchange of data within the different elements of
the system. By leveraging XML as a standardized
format for data encoding, the system adeptly
transmits and interprets queries, user profiles, job
proposals, and other pertinent information, thereby
fostering efficient communication and enabling
effective decision-making processes.
Fig. 1: General Architecture of the system
4 Results: Applicative Case Study
Using the described methodology and structure, this
system was tested to verify its effectiveness in
quickly matching users to the most suitable job
opportunities by assessing the compatibility
between the user’s profile and job offers. The
testing process consists of various steps:
1. User Profile Upload: when a user first logs
in, the system requests them to upload their
profiles, which should contain basic
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1952
Volume 21, 2024
information like job titles, experiences, jobs
preferences.
2. Job Offer Keywords: the system links each
job offer to keywords or parameters to
define their fields.
3. Combination of User Profile and Job
Proposals: Using the Bayes’ algorithm, the
system flatters the suggesting process that
occurs among user profile and job offers,
examining the degree of correspondence
among each job proposal, linking the
keywords found in the previous step and the
characteristics of the user profile.
4. Starting the search: after uploading the user
profile and indexing the job with keywords,
the research process is launched. This led
to initiate queries to identify job
opportunities that closely align with the
user's characteristics and experiences.
5. Identification of suitable job offers: based
on the alignment of the user profile with the
job proposal keyword, the system finds and
ranks the best job offers and the search goes
further. These employment offers are
prioritized and presented to the user.
6. Search storage: inside the system the
research results are stored together with
user’s feedback. Thanks to this, the system
is allowed to keep recorded research and
results, simplifying future adjustments.
7. Search Refinement based on History:
Leveraging the stored search data, the
system continually refines and improves its
search algorithms. By analyzing past search
patterns, user preferences, and successful
matches, the system adapts its
recommendations to better align with user
expectations and requirements.
The entire applicative case study is presented in
the next subparagraphs.
4.1 Graphical user Interface
In the creation of this paper, we tried to define
friendly graphical interfaces. Figure 2 shows the
"entry" interface to the system. In particular, on the
left side of the screen there is a menu divided into
sections. The first section, "Sign up", allows the
storage of a single job proposal or user in the system
database. These interfaces adhere to fundamental
design principles, such as user-centered design,
consistency, simplicity, and feedback mechanisms:
User-centered design is paramount, as it entails
understanding user needs, behaviors, and goals to
craft interfaces that align with their expectations;
Consistency across design elements, including
layout, typography, and color scheme, fosters
familiarity and coherence, enhancing usability; An
important characteristic of the system is the
simplicity. In the proposed system, the implemented
interface is based on an unassuming approach that
shows the information without doubts and in a clear
way, minimizing the cognitive burden. Users are
driven to interactive objects and tools which
guarantee that they can comprehend how to use the
interface in the more useful way, using visual
feedback mechanisms and clear pathways.
Usability of the proposed system also improves the
client experience, emphasizing accessibility; in
particular, users are helped during the research to
find the wanted services, thanks to the system’s
user-friendly navigation paths and clear content
organization. Several tools offered by the graphical
interface, such as interactive objects, and intuitive
input form, aid user engagement; Input forms are
designed with clear labels and fields, inline
validation, and autocomplete features to assist users
in providing accurate information efficiently.
Interactive elements, such as buttons and links, are
visually distinct and responsive, encouraging user
engagement.
Fig. 2: System input interface.
The next interface “Find work” is the most
important part of the system and is described in the
next paragraph. It displays the results obtained when
a user executes a query. All interactions between the
system and the user take place via Web pages. An
example of interaction is illustrated in Figure 2.
Through this page, it is possible to enter all the
information that describes a user who accesses the
system. Initially, the user is invited to fill in the
form in the figure which is the interface of the XML
document in Figure 3.
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1953
Volume 21, 2024
Fig. 3: Graphical interface for registering a user
Fig. 4: Empty XML knowledge base for a generic
user
At the end of filling out the form, the system
will provide a Username which will be used by the
user to identify themselves whenever they wish to
access the service. The information entered by the
user is stored using appropriate XML documents. A
generic organization that wants to insert one or more
job proposals into the system has a graphical
support interface at its disposal. This interface is
shown in Figure 4. It graphically represents the
XML document in Figure 5.
Fig. 5: Graphical interface for inserting new job
proposals
Fig. 6: XML knowledge base for a generic job
proposal.
The information associated with the single job
proposal will be stored in an XML document,
(Figure 6). To search for job offers of interest, a user
has an appropriate graphical interface at his
disposal. It is shown in Figure 7. The results
following the execution of a query by the system are
presented to the user via the graphical interface
shown in Figure 8.
Fig. 7: Graphical interface for searching for job
offers of interest to a user.
Fig. 8: Results of a query.
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1954
Volume 21, 2024
Throughout this endeavor, the team focused on
crafting intuitive and user-friendly graphical
interfaces to facilitate seamless interaction between
users and the system. The design objective was to
provide effortless access to various functionalities
such as user registration, job proposal insertion, and
job search. For instance, the user registration
interface allows users to input their information
through a straightforward web form, structured in
alignment with an XML document layout. Upon
completion, users are assigned a username for future
access, and their provided details are stored within
XML documents. Similarly, organizations use a
dedicated graphical interface to input new job
proposals into the system, with each proposal's
information structured within an XML document to
ensure consistency and coherence. Leveraging XML
as the data storage format offers versatility and
interoperability, facilitating seamless data exchange
across various systems and platforms. All
interactions with the system occur through web
pages, ensuring accessibility via standard web
browsers and simplifying user engagement. The
overarching aim of these interfaces is to establish a
scalable and user-centric system supporting both
users and organizations in the recruitment process,
delivering a pleasant user experience and
streamlining complex operations into clear and
intuitive actions.
4.2 Experimental Analysis of the System
Behavior
To evaluate the performance of our system, we
performed several experiments. In them, the
Recruitment Agent was "launched" on a Pentium IV
with a 2.6 GHz processor and with 512 Mb of
RAM, while the User Agents were activated on PCs
equipped with a Pentium III processor with a 500
MHz frequency and with 128 MB of RAM. In our
experimental campaign we used a set of users
made up of fifty
volunteers. The available job offers were extracted
from the website www.jobpilot.it; the number of
available proposals is equal to one thousand. These
proposals are associated with different application
domains such as Information and
Telecommunications Technologies.; Health Care;
Legal and Administrative Sector; Marketing and so
on. In our experiments we used two metrics widely
used in the literature, namely Precision and Recall.
In particular, each user submitted a
query and our system built the corresponding
set. Subsequently, we asked each user
to identify a set of job
proposals considered interesting. Finally, we
activated our system to build the set The
chosen metrics are: Precision and Recall. Precision
and recall were chosen as evaluation metrics for the
job recommendation system due to their ability to
provide a comprehensive assessment of the system's
performance in retrieving relevant job offers. In
recommendation systems, there is often a trade-off
between precision and recall. Focusing solely on
one can adversely affect the other. Therefore,
evaluating both metrics provides a balanced view of
the system’s performance. Using both precision and
recall ensures a comprehensive evaluation. Precision
checks the relevancy of the recommendations, while
recall ensures that no relevant options are
overlooked. Their combined use helps in achieving
a balanced and user-centric recommendation
system. As an example, if our system retrieves 100
job offers for a user and 80 of these are relevant
(based on user feedback), the precision is 80%. If
there are 100 relevant job offers in the entire
database and our system retrieves 80 of them, the
recall is 80%. These metrics help us understand the
balance between providing a sufficient number of
relevant jobs offers and avoiding overwhelming the
user with irrelevant options. The Precision and
Recall associated with were calculated as
follows (1), (2):
(1)
(2)
The Average Precision and Average
Recall associated with the set were
calculated as follows (3), (4):
(3)
(4)
Before carrying out the experiments, we
hypothesized that Precision and Recall should
increase as the number of sessions carried out by
users increases; in fact, our system was designed to
adapt its behavior based on user feedback and their
preferences. To verify this intuition, in the first
series of experiments we calculated the values of
and for different sessions of use
of the system by the users. Figure 9 is the graphical
representation of the experimental results.
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1955
Volume 21, 2024
Fig. 9: Average Precision and Recall.
From this figure it is possible to observe that,
initially, both and present an
oscillatory behavior and take on rather low values in
some sessions; for example, the minimum values of
and are 0.3 and 0.36 respectively.
This behavior is justified by the fact that, initially,
both user profiles and their feedback are limited.
However, after 30 sessions, both uupdateda
previously implemented initial work and improving
its results. and become almost constant and
high; in particular, is always greater than
0.85 and avgrec is always greater than 0.75. These
results confirm our original intuitions and indicate
that the system performance is satisfactory. Finally,
note that is generally greater than ;
in fact, in this application context, precision has, in
general, greater importance than Recall since the
user could be "frustrated" by the presence of a high
number of interesting proposals. As a subsequent
experiment, we analyzed the impact of the initial
value of the Audacity Degree R0 on the system
performance; in particular, we considered different
values of R0 and calculated the corresponding values
of and after various sessions.
The corresponding results are shown in Table 2 and
Table 3.
Table 2. Impact of R0 on system performance.
Audaci
ty
Session 1
Session 5
Session 15
AvgP
re
AvgR
ec
AvgP
re
AvgR
ec
AvgP
re
AvgR
ec
0.05
0.5
0.3
0.42
0.65
0.36
0.46
0.15
0.34
0.44
0.46
0.52
0.33
0.45
0.30
0.38
0.41
0.53
0.5
0.36
0.34
0.40
0.3
0.36
0.65
0.4
0.34
0.4
0.60
0.5
0.42
0.55
0.61
0.32
0.43
0.75
0.43
0.32
0.51
0.5
0.32
0.41
0.90
0.52
0.45
0.42
0.46
0.35
0.35
1
0.43
0.52
0.40
0.44
0.33
0.33
Table 3. Impact of R0 on system performance.
Auda
city
Session 30
Session 50
Session 100
Avg
Pre
Avg
Rec
Avg
Pre
Avg
Rec
Avg
Pre
Avg
Rec
0.05
0.59
0.51
0.73
0.72
0.86
0.78
0.15
0.61
0.53
0.76
0.75
0.91
0.75
0.30
0.74
0.65
0.80
0.81
0.92
0.83
0.40
0.81
0.73
0.88
0.80
0.9
0.81
0.60
0.76
0.71
0.85
0.8
0.91
0.85
0.75
0.68
0.64
0.79
0.72
0.92
0.81
0.90
0.62
0.61
0.75
0.74
0.88
0.81
1
0.62
0.59
0.71
0.69
0.83
0.79
From the analysis of this table, we observe that
if R0 is too low or too high, or if it is less than 0.3 or
greater than 0.6, our system requires numerous
sessions to define the user's preferences and,
consequently, to obtain acceptable and
results. Conversely, when R0 varies
between 0.3 and 0.6, few sessions are needed
(generally around 25) to obtain convincing results:
for example, if R0=0.4, after 30 sessions we have
and
Results are also shown through plots in Figures
10a), 10b), 10c), 10d), 10e) and 10f).
Fig. 10a: Session 1
Fig. 10b: Session 5
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1956
Volume 21, 2024
Fig. 10c: Session 15
Fig. 10d: Session 30
Fig. 10e: Session 50
Fig. 10f: Session 100
Finally, to verify the performance of our system
in different contexts, we applied our algorithm to
data coming from different domains; some of them
are generic, while others are extremely specialized.
In this work we report the results obtained in two
generic domains, namely Legal/Administrative and
Marketing, and in two specialized domains, namely
IT Consultant and Medical Consultant. For each of
these domains, we measured and
after several sessions. The results obtained are
reported in Table 4 and Table 5.
Table 4. System performance in different
application domains
Domain
Session 1
Session 5
Session 15
Avg
Pre
Avg
Rec
Avg
Pre
Avg
Rec
Avg
Pre
Avg
Rec
Legal-
Adminis
trative
0.5
0.3
0.71
0.65
0.83
0.75
Marketi
ng
0.34
0.32
0.73
0.68
0.86
0.78
IT
Consult
ant
0.44
0.5
0.42
0.39
0.6
0.45
Medical
Consult
ant
0.4
0.5
0.28
0.4
0.54
0.41
Table 5. System performance in different
application domains
Domain
Session 30
Session 50
Session
100
Avg
Pre
Avg
Rec
Avg
Pre
Avg
Rec
Avg
Pre
Avg
Rec
Legal-
Administ
rative
0.87
0.81
0.78
0.75
0.75
0.73
Marketi
ng
0.89
0.84
0.79
0.75
0.77
0.74
IT
Consult
ant
0.7
0.58
0.86
0.75
0.94
0.89
Medical
Consult
ant
0.66
0.6
0.83
0.73
0.92
0.9
From this table it is possible to observe that, in
generic domains, our system obtains satisfactory
results quickly; however, it has difficulty
maintaining such performance after many sessions.
Conversely, in specialized domains, our system
requires many sessions to achieve satisfactory
performance, however, after this initial phase, it can
continuously guarantee outstanding performance.
The explanation for this behavior is the following:
during the first sessions, users of generic domains
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1957
Volume 21, 2024
do not have a precise idea of their needs and
consequently many suggestions proposed by the
system appear of interest to them. Conversely, users
of specific domains have a precise idea of their
interests already during the initial phase and,
consequently, are initially more difficult to satisfy.
As the number of sessions increases, users of
generic domains have a more precise idea of their
needs, but the application domain they are interested
in is too generic to satisfy their requests; however,
from Table 4 and Table 5 it can be seen that, even in
"hostile" conditions, the system achieves
satisfactory performance. Conversely, after a high
number of sessions, the profile of a user interested
in a specific domain is quite rich and detailed and
this allows our system to identify user needs with
high precision and significantly reduce the search
space for job offers. Finally, the future interests of
users of specific domains can be predicted more
easily than the future interests of users belonging to
generic domains.
The results indicate that in generic domains, our
system achieves high performance quickly but
struggles to maintain this as the number of sessions
increases. From this can be deduced that, although
the first advice given by the system could be
efficient, the system needs to be improved to
manage long-term user operations without
deteriorating its performance.
It is important to observe that, at the beginning
and in specific domains, more sessions are needed
by the system to achieve the best performance. But
when adapted to the particular domains or sectors,
for example, it constantly provides excellent results.
This result shows the system's ability to know and
adjust recommendations successfully in specialized
sectors.
A paired t-test comparing our system's precision
with that of CASPER revealed a statistically
significant improvement (p < 0.05). The 95%
confidence interval for the improvement in precision
was [0.05, 0.15], indicating a robust enhancement in
performance.
5 Conclusion
In this paper, an intelligent XML-based multi-agent
system was proposed to support online worker
recruitment services, updating a previously
implemented initial work and improving its results.
We have shown that our system consists of two
main types of agents, namely: (i) a User Agent,
which manages user profiles and interaction with
the system, and (ii) a Recruitment Agent, which
manages job offers. In addition, the system is
enriched by two further types of agents, namely: a
Wrapper Agent which finds new job proposals and
inserts them into the system database, and a
Company Agent, which allows companies to insert,
remove, and modify job proposals. As regards the
limitations, the recommendation system might be
constrained by the availability and quality of
historical data. If the data is incomplete or
unrepresentative, it could negatively impact the
accuracy of recommendations. The system's
recommendations might not always align with users'
true intentions or preferences, as the interpretation
of user behavioral data can be subjective or
occasionally inaccurate. Indeed, with the increase in
users and job proposals inside the system, there may
be some scalability issues regarding preserving
systems's efficiency and handling massive data
volumes. As a consequence, for the system’s
efficiency, it’s mandatory that the data, especially
that related to users’ profiles and job postings, must
be accurate, consistent, and complete. Addressing
data quality issues may require rigorous data
cleaning, preprocessing, and validation procedures.
On the other hand, as regards the future directions,
the proposed enhancements, including the
integration of intelligent agents, the utilization of
XML for data representation, and the
implementation of an item-based recommendation
algorithm, are technically feasible. Intelligent agents
have been successfully applied in various domains,
and there exist frameworks and tools to facilitate
their development and integration. XML is a widely
adopted standard for data representation and
exchange, supported by libraries and APIs in most
programming languages. Other potential future
developments could be to introduce a multi-modal
interaction between the agent and the user and to
develop an adaptive schema evolution, able to
dynamically adapt to changes in the data structure
without requiring manual modifications to the XML
schema, making the recommender system more
flexible.
By leveraging intelligent agents and
recommendation algorithms, the system can provide
more personalized and relevant job
recommendations tailored to individual user
preferences. Improved user interfaces and intuitive
interaction mechanisms can enhance the overall user
experience, increasing user engagement and
satisfaction. Automation of the job search process
and targeted recommendations can save users time
and effort, leading to improved efficiency and
productivity. Furthermore, the scalability and
adaptability of the system can be enhanced,
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1958
Volume 21, 2024
allowing it to accommodate a growing user base and
evolving user needs. As an evolution of the system,
one could think of applying this methodology in any
field of research, both in the documentary field, for
example, the search for online documentation that
reflects certain interests of the user, and in the
planning field by relating the specific requests that a
figure professional must have to be able to work
internally on a project. Yet another evolution could
include integrating the methodology with an e-
learning system to build a system capable of
suggesting to a user the knowledge that he/she
should acquire to access new job offers.
The system could produce more accurate
recommendations by understanding deeper into user
preferences and job demand, through the analysis of
textual and visual data.
In addition, improvements in speed and
reliability could be achieved through the
optimization of the system’s performance, for
example, using more performant algorithms, parallel
processing techniques, and hardware acceleration.
In the future, additional tests utilizing novel AI
algorithms might be conducted to improve the
system's performance in terms of timing and
outcome reliability. So, there is room for
improvement in the suggested approach to boost the
recommendation system's scalability, efficiency,
customization, and user experience.
References:
[1] Moscarini, G., & Postel-Vinay, F. (2024). On
the job search and business cycles. Revue
économique, 75(1), 73-112.
https://doi.org/10.3917/reco.751.0073 (Last
accessed January, 2024).
[2] Hambarde, K. A., & Proenca, H. (2023).
Information retrieval: recent advances and
beyond, 11, 76581 - 76604. IEEE Access.
DOI:10.1109/ACCESS.2023.3295776 (Last
accessed February, 2024).
[3] Ramachandran, K. K., Phatak, S. S., Akram,
S. V., Patidar, V., Raju, A. M., & Ponnusamy,
R. (2023, January). Integration of machine
learning algorithms for E-Learning System
course recommendation based on Data
Science. In 2023 International Conference on
Artificial Intelligence and Smart
Communication (AISC) (Greater Noida, India)
(pp. 634-638). IEEE. DOI:
10.1109/AISC56616.2023.10085048 (Last
accessed February, 2024).
[4] Singhal, A. (2001). Modern information
retrieval: A brief overview. IEEE Data Eng.
Bull., 24(4), 35-43. (Last accessed February,
2024)
[5] Cesta, A., & D'Aloisi, D. (1996). Building
interfaces as personal agents: a case study.
ACM SIGCHI Bulletin, 28(3), 108-113.
https://doi.org/10.1145/231132.231154 (Last
accessed February, 2024)
[6] Deshpande M., Karypis G (2004). Item-based
Top-N recommendation algorithms. ACM
Transactions on Information Systems,
22(1):143-177.
https://doi.org/10.1145/963770.963776 (Last
accessed February 2024)
[7] Bradley, K., & Smyth, B. (2003).
Personalized information ordering: a case
study in online recruitment. Knowledge-Based
Systems, 16(5-6), 269-275.
https://doi.org/10.1016/S0950-
7051(03)00028-5 (Last accessed March,
2024)
[8] Bankins, S., Ocampo, A. C., Marrone, M.,
Restubog, S. L. D., & Woo, S. E. (2024). A
multilevel review of artificial intelligence in
organizations: Implications for organizational
behavior research and practice. Journal of
Organizational Behavior, 45(2), 159-182.
https://doi.org/10.1002/job.2735 (Last
accessed March, 2024)
[9] González‐Briones, A., Chamoso, P., Pavon,
J., De La Prieta, F., & Corchado, J. M. (2024).
Job offers recommender system based on
virtual organizations. Expert Systems, 41(2),
e13152. https://doi.org/10.1111/exsy.13152
(Last accessed March, 2024)
[10] Gusain, A., Singh, T., Pandey, S., Pachourui,
V., Singh, R., & Kumar, A. (2023, March). E-
Recruitment using Artificial Intelligence as
Preventive Measures. In 2023 International
Conference on Sustainable Computing and
Data Communication Systems (ICSCDS)
(Erode, India) (pp. 516-522). IEEE. DOI:
10.1109/ICSCDS56580.2023.10105102 (Last
accessed February, 2024)
[11] Lieberman, H. (1995). Letizia: An agent that
assists Web browsing. In Proc. of the
International Joint Conference on Artificial
Intelligence (IJCAI-95), pages 924 929,
Montreal, Quebec, Canada.
https://cdn.aaai.org/Symposia/Fall/1995/FS-
95-03/FS95-03-016.pdf (Last accessed
February, 2024)
[12] Liu, F., Yu, C., Meng. W. (2004).
Personalized Web search for improving
retrieval effectiveness. IEEE Transactions on
Knowledge and Data Engineering, 16(1):28-
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1959
Volume 21, 2024
40. DOI: 10.1109/TKDE.2004.1264820 (Last
accessed February, 2024)
[13] Jennings, N.R., He M., Leung, H. (2003). On
agent-mediated electronic commerce. IEEE
Transactions on Knowledge and Data
Engineering, 15(4):985-1003. DOI:
10.1109/TKDE.2003.1209014 (Last accessed
February, 2024)
[14] Hsinchun, C., Yi-Ming, C., Ramsey, M., &
Yang, C. C. (1998). An intelligent personal
spider (agent) for dynamic Internet/Intranet
searching. Decision Support Systems, 23(1),
41-58. https://doi.org/10.1016/S0167-
9236(98)00035-9 (Last accessed February,
2024)
[15] Resnick, P., & Varian, H. R. (1997).
Recommender systems. Communications of
the ACM, 40(3), 56-58.
https://doi.org/10.1016/j.physrep.2012.02.006
(Last accessed February, 2024)
[16] Zaiane, O. R. (2002, December). Building a
recommender agent for e-learning systems. In
International Conference on Computers in
Education, 2002. Proceedings. (Auckland,
New Zealand) (pp. 55-59). IEEE. DOI:
10.1109/CIE.2002.1185862 (Last accessed
February, 2024)
[17] Barrile, V., Cotroneo, F., Genovese, E., &
Bilotta, G. (2023). Using Snn Neural
Networks Trained with High Resolution Data
2 and Applied to Copernicus SENTINEL-2
Data. The International Archives of the
Photogrammetry, Remote Sensing and Spatial
Information Sciences, 48, 27-31.
https://doi.org/10.5194/isprs-archives-
XLVIII-2-W3-2023-27-2023 (Last accessed
June, 2023)
[18] Barrile, V., Cotroneo, F., Genovese, E.,
Barrile, E., & Bilotta, G. (2023). An AI
Segmenter on Medical Imaging for Geomatics
Applications Consisting of a Two-State
Pipeline, Snns Network and Watershed
Algorithm. The International Archives of the
Photogrammetry, Remote Sensing and Spatial
Information Sciences, 48, 21-26.
https://doi.org/10.5194/isprs-archives-
XLVIII-2-W3-2023-21-2023 (Last accessed
June, 2023)
[19] Demir, M., & Günaydın, Y. (2023). A digital
job application reference: how do social
media posts affect the recruitment process?.
Employee Relations: The International
Journal, 45(2), 457-477.
https://doi.org/10.1108/ER-05-2022-0232
(Last accessed March, 2024)
[20] Castilla, E. J., & Rho, H. J. (2023). The
gendering of job postings in the online
recruitment process. Management Science,
69(11), 6912-6939.
https://doi.org/10.1287/mnsc.2023.4674 (Last
accessed March, 2024)
[21] Younis, O., Jambi, K., Eassa, F., & Elrefaei,
L. (2024). A Proposal for a Tokenized
Intelligent System: A Prediction for an AI-
Based Scheduling, Secured Using Blockchain.
Systems, 12(3), 84.
https://doi.org/10.3390/systems12030084
(Last accessed March, 2024)
[22] Woods, S. A., Ahmed, S., Nikolaou, I., Costa,
A. C., & Anderson, N. R. (2020). Personnel
selection in the digital age: A review of
validity and applicant reactions, and future
research challenges. European Journal of
work and organizational psychology, 29(1),
64-77.
https://doi.org/10.1080/1359432X.2019.16814
01 (Last accessed March, 2024)
[23] Sanchez, W., Martinez, A., Hernandez, Y.,
Estrada, H., & Gonzalez-Mendoza, M. (2023).
A predictive model for stress recognition in
desk jobs. Journal of Ambient Intelligence
and Humanized Computing, 14(1), 17-29.
https://doi.org/10.1007/s12652-018-1149-9
(Last accessed March, 2024)
[24] Frissen, R., Adebayo, K. J., & Nanda, R.
(2023). A machine learning approach to
recognize bias and discrimination in job
advertisements. AI & SOCIETY, 38(2), 1025-
1038. https://doi.org/10.1007/s00146-022-
01574-0 (Last accessed March, 2024)
[25] Naudé, M., Adebayo, K. J., & Nanda, R.
(2023). A machine learning approach to
detecting fraudulent job types. AI &
SOCIETY, 38(2), 1013-1024.
https://doi.org/10.1007/s00146-022-01469-0
(Last accessed May, 2024)
[26] Gnehm, A. S., & Clematide, S. (2020,
November). Text zoning and classification for
job advertisements in German, French and
English. In Proceedings of the Fourth
Workshop on Natural Language Processing
and Computational Social Science (Canada)
(pp. 83-93). DOI:10.18653/v1/2020.nlpcss-
1.10 (Last accessed April, 2024)
[27] Gupta, C., Singh, R. K., & Mohapatra, A. K.
(2023). Secure XML parsing pattern for
prevention of XML attacks. In Information
and Communication Technology for
Competitive Strategies (ICTCS 2022)
Intelligent Strategies for ICT (pp. 759-770).
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1960
Volume 21, 2024
Singapore. Springer Nature Singapore.
https://doi.org/10.1007/978-981-19-9304-
6_68 (Last accessed February, 2024)
[28] Ganeriwala, P., Chambers, C., Sen, C., &
Bhattacharyya, S. (2023, August). Functional
Reasoning of System Architecture in the
System Modeling Language (SysML) With
XML Representation. In International Design
Engineering Technical Conferences and
Computers and Information in Engineering
Conference (Boston, Massachusetts, USA)
(Vol. 87295, p. V002T02A044). American
Society of Mechanical Engineers.
https://doi.org/10.1115/DETC2023-117193
(Last accessed February, 2024)
[29] Zhu, L., Wang, J., & Bai, L. (2023). A general
characterization of integrating and querying
heterogeneous fuzzy spatiotemporal XML
data. Earth Science Informatics, 16(4), 3303-
3321. https://doi.org/10.1007/s12145-023-
01091-8 (Last accessed February, 2024)
[30] Bai, L., Jia, Z., & Liu, J. (2017). Formal
transformation of spatiotemporal data from
object-oriented database to XML. J. Digit. Inf.
Manage., 15(1), 1.
https://dline.info/fpaper/jdim/v15i1/jdimv15i1
_1.pdf (Last accessed February, 2024)
[31] Pananjung, M. B., & ST, T. E. W. (2023,
September). Development of SQL Interface
for Spatial Data Processing in XML
Databases. In 2023 IEEE International
Conference on Data and Software
Engineering (ICoDSE) (Toba, Indonesia) (pp.
79-84). IEEE. DOI:
10.1109/ICoDSE59534.2023.10291265 (Last
accessed February, 2024)
[32] Pahuja, V., Dubey, R., & Sharma, I. (2023,
December). Machine Learning-Based
Detection and Mitigation of XML SQL
Injection Attacks. In 2023 Global Conference
on Information Technologies and
Communications (GCITC) (Bangalore, India)
(pp. 1-6). IEEE. DOI:
10.1109/GCITC60406.2023.10426458 (Last
accessed March, 2024)
[33] Bikaki, A., Peters, M., Krozel, J., &
Kakadiaris, I. A. (2024). Building an open-
source collaborative platform for migration
research: A metadata modeling approach
using XML. Knowledge-Based Systems, 299,
111823.
https://doi.org/10.1016/j.knosys.2024.111823
(Last accessed January, 2024)
[34] Bilotta, G., Genoves e, E., Citroni, R.,
Cotroneo, F., Meduri, G. M., & Barrile, V.
(2023). Integration of an Innovative
Atmospheric Forecasting Simulator and
Remote Sensing Data into a Geographical
Information System in the Frame of
Agriculture 4.0 Concept. AgriEngineering,
5(3), 1280-1301. (Last accessed March, 2024)
Declaration of Generative AI and AI-assisted
Technologies in the Writing Process
During the preparation of this work the authors used
Gemini in order to improve the readability and
language for the introduction of the manuscript.
After using this tool/service, the authors reviewed
and edited the content as needed and take full
responsibility for the content of the publication.
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
The authors equally contributed in the present
research, at all stages from the formulation of the
problem to the final findings and solution.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
No funding was received for conducting this study.
Conflict of Interest
The authors have no conflicts of interest to declare.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
WSEAS TRANSACTIONS on BUSINESS and ECONOMICS
DOI: 10.37394/23207.2024.21.158
Vincenzo Barrile, Piero Francesco Spano’,
Emanuela Genovese, Gabriele Barrile,
Giuseppe Maria Meduri
E-ISSN: 2224-2899
1961
Volume 21, 2024
