To Analyse Human Sentiments on Social Media Platforms and Their Interest Inference by Using Deep Learning

Abstract

In the modern environment, human emotions are complicated and multifaceted. People are exposed to a wide range of information, ideas, and experiences as a result of the quick development of technology and globalisation, which can affect how they feel and how they view the world. from multiple social network platforms that users engaged in. User interest exhibits dual-heterogeneities: it is complementarily and multiple social networks reflects them comprehensively; interest of user are not independent of each other rather inter-correlated in a nonuniform way. Although past approaches towards these problems had great success that considers the dual heterogeneities simultaneously, consistency of sources, uses multi-sourced and multi-tasked learning scheme. To address this, a web application coupled with machine learning technique like linear regression to predict the human sentiments by inviting them to attempt the survey. The answers to the survey are per-processed to identify the relationship between different independent variables and a dependent variable that represents a particular sentiment. A database utilising MYSql is used to record the user personal information, the responses, and the quiz results of users who have registered while unknown users are first requested to sign up. Pre-trained data and Supervised Learning algorithms can be used to detect sentiments.

Introduction

I. INTRODUCTION

Social networks have changed the way we communicate, and they supply a variety of helpful data for gaining a complete view of users' interests and behaviours. They represent users' interests from a variety of perspectives. Due to the various architecture of a social network, it only delivers fragmented information about a person from a particular angle. For instance, users may follow accounts that they are interested in on Facebook, Twitter, etc., and search for the topics they are interested in on Quora. As a result, it is important to develop an effective strategy for numerous social network learning that not only captures this complexity better but also gauges consistency among several sources as well as the degree of confidence in each one.

A lot of work has been put into completing this challenging multi-source learning task. However, the existing method does not evaluate the level of confidence in any given source and treats all sources in the same way. The existing work [1] regularizes the source consistency but fails to explicitly describe the source confidence. Another approach [2] make use of linear models to combine data from a variety of sources. However, these models are unable to adequately describe the intricate structure of social networks. These deep learning models may reflect the complexities and consistency better, but they fail to explicitly describe the source confidence which would help create a model that is more accurate than the existing one. Another approach [1] uses deep learning (FARSEEING) model for confidence and consistency but is implemented only for Volunteerism Tendency Prediction, which is a multisourced mono-task learning model.

A structure-constrained multi-sourced and multi-tasked learning method used to infer user interests is what we propose as an addition to the earlier works. Particularly, our method jointly regularises two important traits. The first is source confidence followed by source consistency. The machine learning model's estimated likelihood that the extracted value is correct given the documents and labels you've provided is represented by confidence. A group of individuals is first aligned by connecting up their several social media accounts, and then all publicly available data for that group of users is retrieved. Second, to describe the given users, multifaceted aspects like demographic, linguistic, and behavioural variables are retrieved. For multiple social network learning, FARSEEING receives the extracted features as input. The benefit of the proposed system is that it combines social networks with source confidence and consistency regularisation using deep learning to accurately portray the intricate nature of the diverse social network.

The disadvantage is that it was implemented only for Volunteerism Tendency Prediction and, it solves only mono-task learning problems.

II. METHODOLOGY

Initial topic and review were done after much deliberation and discussion. Numerous IEEE Explore published papers were studied and five primary papers were selected, which are listed in the reference. The first blueprint was created and refined to give the following final flowchart for user sentiments.

B. Step-Two

The performance outcome of the SM2L-i over permutations of each question's answers is displayed in Table 2. We found that the more options we provide for each inquiry, the better the performance might be. This implies that the relationships among the questions are complementary rather than antagonistic. Additionally, we discovered that combining data from all of these questions can result in higher performance than using each source separately. The fact that SM2L will deteriorate into multi-task learning when the context problem only includes one source is not incomprehensible.

Table II?1 Evaluation Metrics

On Evaluation Metrics:

Precision is more crucial than memory for the task of inferring user interest. We therefore used two measures to validate our plan: S@K and P@K.

S@K: It displays the average likelihood that a proper interest will be found among the top K suggested interests.

P@K: The percentage of the top K recommended interests that are accurate is what it stands for.

C. Step Three

The outcome uses a deep learning technique to describe user sentiment. The programme uses a linear regression model to identify tasks that can be useful to the users based on those results. The history button on the website gives you access to the database that contains the quiz data and its outcomes. The matching model used by the system is a LR model, which shows a relationship between a dependent variable and one or more independent variables.LR are favoured for this system even if other supervised algorithm, such SVMs, can also be employed for decision making.

III. DESIGN AND DEVELOPMENT

A. Requirements Gathering

To construct the benchmark dataset, we need to first tackle the problem of “human sentiments alignment”, which aims to identify of users across different social networks who have provided their opinions through a quiz. To achieve this, we invited users to participate in the quiz, enabling us to identify and align users' sentiments across various platforms. Once we had identified the users, we focused on obtaining representative interests from the collected data. To achieve this, we filtered out interests that were liked by fewer than 15 users, as these were deemed less significant and not representative of the overall sentiment. This approach allowed us to ensure that the benchmark dataset was constructed using high-quality, representative data.

B. Data pre-processing

Data pre-processing is a crucial step in preparing raw data for further analysis. Its purpose is to refine the data by cleaning it and putting it in a suitable format for use with machine learning algorithms. Among the primary techniques used in pre-processing are data cleaning, normalization, and removal of duplicate values. These techniques were applied to a dataset of Human information to ensure that the text data was properly normalized.

C. Decision making using AI

SVM: The first baseline is a traditional single source single task learning method—support vector machine (SVM) [Cortes and Vapnik, 1995], which simply concatenates the features generated from different sources into a single feature vector and learns each task individually. We chose the learning formulation with the kernel of radial-basis function, implemented based on LIBSVM [Chang and Lin, 2011]. RLS: The second baseline is the regularized least squares (RLS) model [Kim et al., 2007], which also learns each task individually and aims to minimize the objective function of

regMVMT: The third baseline is the regularized multi-view multi-task learning model, introduced in [Zhang and Huan, 2012]. This model regulates both the source consistency and the task relatedness. However, it simply assumes the uniform relatedness among tasks. SM2L-eu: The fourth baseline is a derivation of SM2L-e. This method constructs the tree structure based on external source in the same manner as SM2L-e but assigns uniform weights to all nodes. SM2L-iu: The fifth baseline is a derivation of SM2L-i, which constructs the tree structure using internal source but weights all nodes uniformly. We adopted the grid search strategy to determine the optimal values for the regularization parameters among the values {10r : r ∈ {−12, · · · , −1}}. Experimental results reported in this work are the average values over 10-fold cross validation. Noticeably, we tuned the K in S@K and P@K from 1 to 10 and reported the optimal performance for each fold. Generally, the S@K reaches the maximum at K = 10, while K = 1 is much preferable regarding P@K.

For the task of user interest inference, precision is of more importance as compared to recall. We thus validated our scheme via two metrics: S@K and P@K.

 S@K: It represents the mean probability that a correct interest is captured within the top K recommended interests.        P@K: It stands for the proportion of the top K recommended interests that are correct.

D. Implementation of web page with ML predictive MODEL

Using JavaScript, we developed a user login system within a web page. Once logged in, the user is presented with a quiz consisting of multiple-choice questions. The user's answers to these questions are then used as input for a trained machine learning model that predicts the user's interests based on their responses. This prediction is done seamlessly in the background, without any disruption to the user's quiz-taking experience.

At the end of the quiz, the final result is displayed to the user, indicating their predicted interests based on their quiz responses. This approach offers a novel and engaging way for users to explore their interests while also providing valuable data for research and analysis. With this system in place, we can gain insights into users' interests and preferences and use this information to improve the accuracy of our machine learning models.\

VI. ACKNOWLEDGMENT

We express our heartfelt appreciation and gratitude to our Project Guide, Asst. Prof. Simran Pal R, Professor of Computer Science and Engineering, HKBK College of Engineering, Bangalore, for her intellectually-motivating support and invaluable encouragement during our Project work. Her comprehensive knowledge and understanding of the research topic as well as her uncompromising and sensible attitude towards research and insistence on quality work have profoundly influenced us and will definitely benefit our future work. We would like to convey our heartfelt thanks to her painstaking modification of this report.

We would like to profoundly thank Chairman Mr. C. M. Faiz and Management of HKBK College of Engineering for providing such a healthy environment for the successful completion of project work.

We would like to express our thanks to the Principal DR. Tabassum Ara, for her encouragement that motivated us for the successful completion of Project Work.  We wish to express our gratitude to DR. Ashok Kumar P. S., Professor and Head of Department of Computer Science & Engineering for providing a healthy environment for the successful completion of the Project work.

We are also grateful to DR. Deepak N.R., Professor, Dept of CSE, HKBKCE. We are extremely thankful and indebted to him for sharing his expertise, sincere and valuable guidance and encouragement extended to us. We would also like to thank all other teaching and technical staffs of Department of Computer Science and Engineering, who have directly or indirectly helped us in the completion of this Project Work.

Conclusion

In light of user interest inference, this work provides an enhanced structured of multi-sourced and multi-tasked learning method. Implementation of a human sentiments and interest inference system using linear regression model that detects known sentiments with high precision, predicts a task that aids in monitoring human attitudes, and stores this prediction and the aided task in a database. The System can be used to integrate with other technologies, such as natural language processing, to provide more comprehensive insights into human behavior.

References

[1] Devarshi K, Ganga G, D Swami, Churhcan R (2022, December). To Study and Analyse Human Behaviour on Social Media Platforms and their Interest Inference using Deep Learning. [2] Jia, Y., Song, X., Zhou, J., Liu, L., Nie, L., & Rosenblum, D. S. (2016, February). Fusing social networks with deep learning for volunteerism tendency prediction. In Thirtieth AAAI conference on artificial intelligence. [3] . Song, X., Nie, L., Zhang, L., Liu, M., & Chua, T. S. (2015, June). Interest inference via structure-constrained multi-source multi-task learning. In Twenty-Fourth International Joint Conference on Artificial Intelligence. [4] . Song, X., Nie, L., Zhang, L., Akbari, M., & Chua, T. S. (2015, August). Multiple social network learning and its application in volunteerism tendency prediction. In Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval (pp. 213-222). [5] Hayat, M. K., Daud, A., Alshdadi, A. A., Banjar, A., Abbasi, R. A., Bao, Y., & Dawood, H. (2019). Towards deep learning prospects: insights for social media analytics. IEEE Access, 7, 36958- 36979. [6] Amin, F., Ahmad, A., & Choi, G. S. (2018, April). To study and analyse human behaviours on social networks. In 2018 4th Annual International Conference on Network and Information Systems for Computers (ICNISC) (pp. 233-236). IEEE. [7] Bengio, Y. (2009). Learning deep architectures for AI. Foundations and trends® in Machine Learning, 2(1), 1- 127. [8] Larochelle, H., & Bengio, Y. (2008, July). Classification using discriminative restricted Boltzmann machines. In Proceedings of the 25th international conference on Machine learning (pp. 536-543).

Copyright

Copyright © 2023 Dr. D. Loganathan, Ganga Gyatso, Devarshi Kothari, Churchan Rai, Divyansh Swami. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.