14 Nov 2024
09:00 Master's Defense Room 85 of IC2
Topic
Influence of Topic Modeling on Sentiment Analysis from User-Generated Product Reviews
Student
Patrick Anderson Matias de Araújo
Advisor / Teacher
- Co-advisor: Marcelo da Silva Reis
Brief summary
In today’s data-driven economy, understanding customer feedback is crucial for companies to improve their products and services. User reviews, which represent a modern form of word-of-mouth marketing, offer valuable insights but are often vast and challenging to interpret manually. Advanced natural language processing (NLP) techniques provide a way to automatically extract meaningful insights from data, transforming the way companies leverage customer input to drive strategic decisions. In this context, addressing topic identification and sentiment analysis in texts is crucial. However, the complexity of integrating topic modeling and sentiment analysis as two core NLP techniques presents challenges, especially when applied to large-scale datasets with diverse linguistic features and informal languages such as sarcasm, slang, and multilingual comments. This master’s thesis investigates the integration of topic modeling and sentiment analysis to improve the interpretation of customer feedback. Focusing on user reviews from three major platforms — Amazon, Netflix, and Spotify — collected via the Google Play Store, our study leverages advanced natural language processing techniques to extract important insights from user comments. In particular, we evaluate two approaches: 1) using specialized and fine-tuned NLP models such as BERT, T5, and BERTopic; 2) using larger pre-trained transformer models such as Meta’s Llama 3 8B and Mixtral 8x7B. Our investigation evaluates how topic modeling impacts sentiment classification in multi-class settings by examining their metrics (3-class vs. 5-class). The research highlights the effectiveness of different NLP models in providing companies with deeper insights into customer behavior and enabling data-driven strategic decisions. This master’s thesis contributes by demonstrating the relevance of combining topic modeling with sentiment analysis, advancing the application of AI in business intelligence.
Examination Board
Headlines:
| Julio Cesar dos Reis | IC / UNICAMP |
| Helena de Almeida Maia | IC / UNICAMP |
| Thiago Henrique Silva | DAINF / UTFPR |
Substitutes:
| Sandra Eliza Fontes de Avila | IC / UNICAMP |
| Luiz Camolesi Júnior | FT / UNICAMP |