Emotion Classification on Social Media Comments Using Categorical

Feature Extraction Along With the Bidirectional Encoder-based Recurrent

Neural Network Classification

S. SARANYA, G. USHA

Department of Computing Technology,

SRMIST, Tamil Nadu,

INDIA

Abstract: - All across the world, social media is one of the most widely used platforms for information exchange.

begins by preprocessing the input data with normalization, followed by categorizing characteristics in feature

extraction utilizing the Linguistic and word count analysis (LIWC). Finally, for the categorization stage, the classify

features might be supplied. Finally, for categorizing emotions, the Bidirectional Encoder based recurrent neural

network classification approach is used. The studies have been carried out with the use of typical social media data that

has been acquired from the kaggle data repository. The findings show that the suggested model outperforms all other

existing mechanisms in terms of overall performance.

Key-Words: - Social media, emotion, Linguistic and word count analysis, Bidirectional Encoder based recurrent neural

network.

views and ideas can be expressed, the availability of

brands, and the speed with which people can engage,

people spend the majority of their time online, as shown

in Figure 1. Meaningful and nonsensical data are

generated as a result of this activity on social media.

Work in this subject is possible because of the rapid

development of technology and the rise in data, as well

as its speed and costs. Text mining, natural language

processing, and other artificial intelligence techniques

are all employed in the field of emotional analysis to

social media and internet sources, such as Twitter,

WhatsApp, Youtube, and Facebook conversations, is

known as sentimental analysis.

machine learning. These mechanical systems are also

used to perform opinion mining. There are various

scenarios in which deep learning algorithms may be

used to create a sentimental analysis model, all of which

are discussed here. The sentimental analysis also be a

on certain films or other items, among other things,

depending on user input made on social media

platforms. Global networking may have taken the world

by surprise, and the world's aspects may have been

diminished as a result of this phenomenon. Using the

results of this study, we suggest a novel framework

Categorical LIWC and BEBRNN for the classification

of people's emotions when they receive positive social

media feedback. To narrow down the search space, a

LIWC-based category feature selection filter has been

precisely classify the emotions within a limited

period

The remaining section of the paper can be organized

as follows, Section 1 shows an overview of opinion

mining impact and the paper's goal contributions;

Section 2 shows an analysis of other existing

technologies; Section 3 illustrates the problem

implemented document.

2 Related Works

With social networking sites systems such as Twitter,

Facebook, Instagram, YouTube, and WhatsApp

sweeping the communication world, it has become

critical that the records residing all over these social

networking sites systems share information relevant to

the point of view, mood, and also conviction of people

Sensorial Model (LSTM-CNN), which combines

LSTM and CNN simultaneously. It's time to reflect on

the Twitter Task Sentiment Analysis's fourth year of

existence. [13], analyzing Twitter sentiment is a

complex process. Talk about it in depth. This is the tenth

parameters between LSTM cells. Each LSTM cell unit

in the circuit has a crossbar comprised of a single

memristor. Piecewise function models using equal

hardware activate the sigmoid and tanh activation

functions in the device under test. [14], BLSTM has

offered a feasible design that includes various channel

characteristics and self-attention mechanisms

(SAMF-BiLSTM). When it comes to transmitting one's

feelings, this strategy uses the most up-to-date

instruments and strategies. Rather than relying on

regression model. The limitation does not operate as

well in this task. [17], biomedical data hypertext mining

is being examined. To extract biological data from

hypertext pages, this study will make use of text-mining

algorithms based on "biomedical ontology" (e.g., online

information mining). The conventional method of

medical language analysis, the met thesaurus, is the

topic of this study. The document is constrained by its

lack of scientific relevance. To develop generative

adversarial networks, hierarchical attention processes

3 Problem Statement

Use an emotion-related grammatical category to

describe the speaker's approach toward a declaration of

certainty and speculative scope to find out which part of

the phrase is unsure. An uncertain segment of a phrase is

It is necessary to do text normalization, stemming, part

of speech tagging, and stop word deletion during the

pre-processing step to prepare the text for the main

processing stage. It is necessary to process the chosen

comments to improve the accuracy of the

importance. Then stop eliminating words from the

matrix, since this will help to reduce the matrix's size

and increase the degree of discrimination across texts.

The stemming of words is then used to consider a unit

for all of the words that are connected to a single root.

As a result, it has little impact on the system's

performance. Stop words contain grammatical terms

like conjunctions and prepositions, as well as other

phrases. There is no meaning or contextual information

in these words. In the case of studies that are primarily

categorical features based on LIWC for sentiment

analysis were as follows.

1. The first step is to familiarize themselves with all the

words in the text.

script)

comments)]

comments)] //

(12)

An association technique, shown in Figure 3, was

Fig. 3: Emotion association

Algorithm:2 Categorical Feature Sentiment Association

Input: Extracted categorical features

Output: Sentiments association

Matching)(match/hot)

End

using an RNN. For example, this RNN enables you to

determine the difference between an independent

variable and a shared one with ease. The proposed

approach first interprets and redistributes the data before

calculating the emotions during categorization using its

class probability. The cycle may be started by activating

a neuron with an equation.

denotes a lower focus word limit.

Fig. 4: BEBRNN layers

Finally, the comments can be classified depending

upon the emotions.

Algorithm: 3 Opinion Classification

Table 1. Classified output

Here a small set of keywords that were likely to

indicate emotional content belonging to any of the

identified keywords looks as follows:

“anger: anger, angry, bitch, fuck, furious, hate, mad”

added automatically. The words that emerged from this

process were then culled because they were deemed to

be inadequate descriptors of human emotion. As shown

in Table 1, the proposed classifier can discriminate

between the various emotions represented by the

"positive," "anger," and "fear" tags.

methodology it can be compared With the existing

mechanisms, [23]. The proposed technique's behavior is

tested using metrics such as accuracy, sensitivity, F1

designated as negative but does not include the exact

numbers is referred to as a "negative integer."

of the existing techniques concerning these parameters.

Figure 5 shows the proposed classification method,

Fig. 8: Number of Comments Vs. F score

Figure 8 shows an F1 rating for the solution. The

results demonstrate that, according to the suggested

strategy, the coefficient of F1 score was 93 percent.