Enhancing Social Media Sentiment Classification Through Effective Text Preprocessing and Machine Learning

This work focused on the impact of text preprocessing on the sentiment classification of social media posts using machine learning approaches. For the experimental part of this research, the subset containing the sentiment data was extracted from the TweetEval database. The research compared three popular classifiers, namely, Multinomial Naive Bayes, Logistic Regression, and Support Vector Machine. The extraction of features from the text was performed with the help of the TF-IDF algorithm. The evaluation of the three models was carried out on both minimally preprocessed and systematically preprocessed text samples. Experimental results have demonstrated that text preprocessing was beneficial for improving the performance of all three classifiers to the same degree. Only Multinomial Naive Bayes classifier showed a certain increase in performance due to the application of systematic text preprocessing, especially in terms of F1-score. On the contrary, the use of minimally preprocessed text led to better outcomes for Logistic Regression and SVM classifiers. Logistic Regression yielded the highest results with the minimally preprocessed text, showing an accuracy of 58.95% and an F1-score of 57.84%. Such results indicate that preprocessing of text is not always beneficial in terms of improved sentiment classification of social media data. There is some information in such texts that can be useful for classification, namely, information contained in punctuation, hashtags, repeating of the characters and other informal elements of the text. Thus, the selection of preprocessing should be made according to the specificities of the dataset and classification algorithm, but not as a mandatory step of text cleaning. The research has proved that traditional ML algorithms offer a rather simple solution to the problem of social media sentiment analysis. The combination of TF-IDF and Logistic Regression offered the best results in the absence of deep learning architecture. However, the difficulties in distinguishing between negative/positive and neutral tweets demonstrate the limitation of the traditional features representation. The future works may include the research of the effect of selective preprocessing on the sentiment classification of social media data.

[1]     Liu, Sentiment Analysis and Opinion Mining. San Rafael, CA, USA: Morgan & Claypool Publishers, 2012. [2]     B. Pang and L. Lee, “Opinion mining and sentiment analysis,” Foundations and Trends in Information Retrieval, vol. 2, nos. 1–2, pp. 1–135, 2008. [3]     E. Cambria, D. Das, S. Bandyopadhyay, and A. Feraco, Eds., A Practical Guide to Sentiment Analysis. Cham, Switzerland: Springer, 2017. [4]     W. Medhat, A. Hassan, and H. Korashy, “Sentiment analysis algorithms and applications: A survey,” Ain Shams Engineering Journal, vol. 5, no. 4, pp. 1093–1113, Dec. 2014. [5]     A. Giachanou and F. Crestani, “Like it or not: A survey of Twitter sentiment analysis methods,” ACM Computing Surveys, vol. 49, no. 2, Art. no. 28, Jun. 2016. [6]     Q. A. Xu, V. Chang, and C. Jayne, “A systematic review of social media-based sentiment analysis: Emerging trends and challenges,” Decision Analytics Journal, vol. 3, Art. no. 100073, Jun. 2022. [7]     S. Rosenthal, N. Farra, and P. Nakov, “SemEval-2017 Task 4: Sentiment analysis in Twitter,” in Proc. 11th Int. Workshop Semantic Evaluation (SemEval-2017), Vancouver, BC, Canada, 2017, pp. 502–518. [8]     P. Nakov, A. Ritter, S. Rosenthal, F. Sebastiani, and V. Stoyanov, “SemEval-2016 Task 4: Sentiment analysis in Twitter,” in Proc. 10th Int. Workshop Semantic Evaluation (SemEval-2016), San Diego, CA, USA, 2016, pp. 1–18. [9]     F. Barbieri, J. Camacho-Collados, L. Espinosa Anke, and L. Neves, “TweetEval: Unified benchmark and comparative evaluation for tweet classification,” in Findings of the Association for Computational Linguistics: EMNLP 2020, 2020, pp. 1644–1650. [10]   Cardiff NLP, “TweetEval: Repository for the TweetEval benchmark,” GitHub. [Online]. Available: https://github.com/cardiffnlp/tweeteval. Accessed: Jul. 10, 2026.