SOGIE Bill Discourse Analysis Using Latent Dirichlet Allocation, Nonnegative Matrix Factorization, BERTopic, and Latent Semantic Analysis

The Sexual Orientation, Gender Identity, and Expression (SOGIE) Bill is a proposed law seeking to address discrimination based on an individual’s SOGIE. The SOGIE Bill has sparked widespread debates on social media, leading to diverse public sentiments, stances, and opinions expressed through textual data. Since there are virtually no searchable papers about applying machine learning (ML) or natural language processing (NLP) techniques in SOGIE Bill-related documents in the current literature, the researchers intend to address this research gap by fulfilling their objective: using sentiment analysis and topic modeling techniques to analyze public discourse around the SOGIE Bill. Sentiment analysis identifies the overall sentiment of a text (negative, positive, or neutral), while topic modeling uncovers the themes discussed in the text. To achieve the main objective, the researchers took the following steps: (a) utilized the unsupervised ML model RoBERTa to assign sentiment scores to the entire corpus, dividing it into 4 subcorpora according to the overall sentiments of the texts (negative, positive, neutral, and other); (b) created and trained Latent Dirichlet Allocation (LDA), Non-negative Matrix Factorization (NMF), BERTopic, and Latent Semantic Analysis (LSA) topic models for each subcorpus; (c) evaluated the performance of the topic models using CV coherence, UMass coherence, and perplexity scores; (d) determined the best-performing topic models based on their scores and extracted their generated topics (each represented by the top 30 words); (e) deduced the general topics of the best-performing models based on the top words; and (f) validated the general topics deduced by the researchers with a lawyer to ensure their correctness and the absence of bias. Results showed that LDA consistently excelled in perplexity across the four subcorpora, NMF and LSA equally excelled in CV coherence, and BERTopic excelled in UMass coherence but with fewer and more repetitive topics. The outputs of this study, which are the general topics representing the sentiments and stances of Filipinos about the SOGIE Bill, will help lawmakers revise the bill and foster an understanding of the different perspectives on the bill.