E-word of mouth sentiment analysis for user behavior studies

Highlights

• Induce and extract the emotion concept in each emotion review by introducing eight factors of sociological theory.

• Opinion mining and sentiment classification analysis are based on user behavior rather than product features.

• We propose a new classification of conceptual learning and natural language processes.

• The basic idea of our approach is to combine machine learning with dictionary dictionary to make multiple classifications of emotions in user behavior.

Abstract

Nowadays, online word-of-mouth has an increasing impact on people's views and decisions, which has attracted many people's attention.The classification and sentiment analyse in online consumer reviews have attracted significant research concerns. In this thesis, we propose and implement a new method to study the extraction and classification of online dating services(ODS)’s comments. Different from traditional emotional analysis which mainly focuses on product attribution, we attempted to infer and extract the emotion concept of each emotional reviews by introducing social cognitive theory. In this study, we selected 4,300 comments with extremely negative/positive emotions published on dating websites as a sample, and used three machine learning algorithms to analyze emotions. When testing and comparing the efficiency of user's behavior research, we use various sentiment analysis, machine learning techniques and dictionary-based sentiment analysis. We found that the combination of machine learning and lexicon-based method can achieve higher accuracy than any type of sentiment analysis. This research will provide a new perspective for the task of user behavior.

Introduction

As global Internet presentations continue to increase, the number of consumers who provide online comments have increased significantly (Lu & Bai, 2021). If exploited properly, abundant data should produce useful insights. One insight that can be obtained from the statistics is the information of electronic word of mouth (EWOM). EWOM is known for its significant impact on consumer behavior (Tobon & García-Madariaga, 2021). EWOM communication framework demostrates the direct relation of adopting EWOM and consumers’ willingness to purchase. EWOM can provide objective information for more and more consumers who trust these communications (Yaniv & Shalom, 2021).Comment mining concerning sentiment analysis is considered to be a suite of proceedings for identifying sentiments, opinions and author's attitudes in texts, transforming them into meaningful information and using them to make business decisions (Siddiqui et al., 2021).

Sentiment classification identifies opinions and arguments in a given text, and it is part of opinion mining. It tries to find statements of agreement or disagreement in comments or reviews that involve positive, negative or neutral statements. Sentiment analysis has attracted widespread attention and has been widely used in many fields (Wang & Zhang, 2020). Up to now, many approaches of sentiment analysis have been proposed, which can be roughly split into document-level, sentence-level and aspect-level(Jiang, Chan, Eichelberger, Ma, & Pikkemaat, 2021).Most of the work of sentiment analysis can be achieved by assessing the document's polarity.Phrase and sentence levels have become common increasingly in recent years.Dictionary-based and machine learning approaches are two of the most common uses of emotion analysis(Ahlem & Khalil, 2020). Building emotional dictionary is the main way to realize sentiment analysis. For example,SentiWordNet (Madani, Erritali, Bengourram, & Sailhan, 2020), SenticNet (Hung, Wu, & Chou, 2020), and OpinionFinder (Wiebe et al., 2005) are famous emotional dictionaries. Sentiment lexicon plays a vital role, because it includes the emotional polarity of a word or phrase which can be used to classify sentiment of textual data.

Most of the available studies is based on sentiment analysis of product reviews. However in this thesis, we focus on discussing the domain of online dating service's (ODS) reviews,because this domain is different from the other product reviews (Wang, 2020). When a person writes down her/his feelings about the usage of the ODS, she/he probably comments not only on website elements (e.g., easy to use, perfect function) but also on the people who have used the website (e.g., dater, matchmaker, friend, parents). Therefore, the review features of the ODS are more complex than the products' reviews, and it's more challenging to analyze the ODS’s reviews. This paper describes a method used natural language processing(NLP) and emotion lexicon generation (Chang, Hwang, & Wu, 2021) to classify the ODS’s reviews (이득영, 장선우, & 전한종, 2019).

Many opinion mining studies have been conducted to analyze unstructured text data including customer reviews, blogs, tweets, news and online (Al-Mashhadany, Hamood, Al-Obaidi, & Al-Mashhsdany, 2021).However, there are still  some problems to be solved. First, there is less research on emotion analysis based on the theory of user's behaviour analysis. Different from traditional sentiment analysis which is based on the products in the customer reviews, the sentiment analysis method based on user behavior analysis is to extract the user's concept and analyze the emotion from the user's point of view.Next, there were few sentiment analysis studies based on multiple category problem. Traditional sentiment analysis is generally based on document-level (Li, H, Kang, Yang, & Zong, 2019), sentence-level (Eng, Nawab, & Shahiduzzaman, 2021) and aspect-level (Lu, Zhu, Zhang, Wu, & Guo, 2020),but this paper is based on the concept-level and multi-sentiment classification problem. The main problem is the extraction of concepts and the classification of multiple emotions (Sznycer & Lukaszewski, 2019).

Our work tries to perform sentiment analysis on users' behaviors influenced by conflict environmental factors obtained by using machine learning models.We selected 4800 controversial context of online dating service (ODS) which had extremely negative/positive rates of popularity in a online dating sites, extracted linguistic features  from them using NLP,and compared their performances within accuracy using machine learning models: Naive Bayes (NB), K-Nearest-Neighbor (KNN) and Support Vector Machines (SVM). To sum up, our contributions in this study are as follows:(1) A new method of learning training data in the form of classifier labels is proposed. By combining machine learning with dictionary methods, the automatically generated questionnaire has more value and credibility.(2)  The opinion mining of comments and the analysis of sentiment classification are based on user behavior, rather than the analysis method based on product features in traditional research. It provides a new direction for sentiment analysis in the future by exploring opinions and classifying emotions from a new perspective.(3) The automatic lexico-based classifier is used to classify user behaviors according to eight features of social cognitive theory as influencing factors. Four classifiers were constructed according to the eight features, and a new constructed dictionary was used to correct the false prediction results of the four classifiers, and vocabulary and learning methods were fused to make up for the defects of the two.

This is followed by a discussion of 4 areas: Section 2 describes the work related to EWOM, opinion mining and emotion analysis. Section 3 develops the narrative of our approach. Section 4 report on the case study of SOD. In section 5, experiment results of efficiency,coverage and discussion are provided. Conclusions and further tasks are presented in section 6.