Twitter sentiment analysis using fuzzy integral classifier fusion

Abstract

A thorough analysis of people’s sentiment about a business, an event or an individual is necessary for business development, event analysis and popularity assessment. Social networks are rich sources of obtaining user opinions about people, events and products. Sentiment analysis conducted using multiple user comments and messages on microblogs is an interesting field of data mining and natural language processing (NLP). Different techniques and algorithms have recently been developed for conducting sentiment analysis on Twitter. Different proposed classification and pure NLP-based methods have different behaviours in predicting sentiment orientation. In this study, we combined the results of the classic classifiers and NLP-based methods to propose a new approach for Twitter sentiment analysis. The proposed method uses a fuzzy measure for determining the importance of each classifier to make the final decision. Fuzzy measures are used with the Choquet fuzzy integral for fusing the classifier outputs in order to generate the final label. Our experiments with different Twitter sentiment datasets show that fuzzy integral-based classifier fusion improves the average accuracy of sentiment classification.

Sentiment analysis on twitter tweets about COVID-19 vaccines usi ng NLP and supervised KNN classification algorithm

“…To obtain the classification result, a supervised k-nearest neighbor (KNN) classification algorithm [17] is used. KNN uses feature similarity where it assigns a data point based on how close it is to its neighbor.…”

Sentiment analysis of tweets through Altmetrics: A machine learning approach

“…Different proposed classification and pure natural language processing (NLP)-based methods have different behaviours in predicting sentiment orientation [42]. The authors suggested that pure NLP-based methods have very low accuracy.…”

What makes a tweet be retweeted? A Bayesian trigram analysis of tweet propagation during the 2015 Colombian political campaign

“…The focus of the literature, however, is based on sentiment analysis in political speeches in social media [28]. According to Emadi and Rahgozar [29], sentiment analysis has been extensively studied with the following five classification techniques: (a) classification using minimum Natural Language Processing (NLP) preprocessing as explained in Jurafsky and Martin [30], (b) pure NLP (lexicon semantic), (c) classification with rich NLP preprocessing and feature extraction, (d) graph-based clustering method and (e) method with classifier fusion approach. However, only the first three have been extensively used, while the latter have been neglected.…”

“…In this sense, our specific contribution in this work is twofold. First, we propose a framework that overcomes some of the limitations reported by Emadi and Rahgozar [29] who claimed that n-gram latent Dirichlet allocation (LDA) topic modelling, extraction and entity network and topic-based sentiment scores require a considerable amount of tagged data. Second, our approach allows a thorough analysis of retweets as a better approach to topic detection [31].…”