Sentiment analysis of product reviews using coupled sentence-bidirectional encode representations from transformers and topic modeling

Chao, Szumin

Permalink

https://hdl.handle.net/2142/124343 Copy

Description

Title

Sentiment analysis of product reviews using coupled sentence-bidirectional encode representations from transformers and topic modeling

Author(s)

Chao, Szumin

Issue Date

2024-05-01

Director of Research (if dissertation) or Advisor (if thesis)

Kim, Harrison

Department of Study

Industrial&Enterprise Sys Eng

Discipline

Industrial Engineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• product design
• product review
• topic modeling
• sentence-bert
• text transformer
• data mining
• data-driven design
• optimization
• latent dirichlet allocation
• non-negative matrix factorization
• latent semantic analysis
• natural language processing
• sentiment analysis

Abstract

In the research of product design, determining the product feature is the essential way to develop optimal business strategy, and the customer reviews on public websites are the valuable sources to reach the goal. The proposed research introduces three topic models and sentiment analysis using Sentence-BERT (S-BERT) technique in detail. Topic modeling stands as a pivotal technique in the domain of NLP (Natural Language Processing) and ML (Machine Learning), offering a systematic approach for uncovering hidden thematic structures within extensive collections of textual data. This unsupervised learning approach facilitates the optimization and organization of large datasets, enabling researchers and practitioners to identify patterns of topics across a corpus without large-scale manual labeling. S-BERT is a updated modification of BERT (Bidirectional Encoder Representations from Transformers) model which enhances its efficiency and effectiveness for generating sentence embeddings, where machines are allowed to understand human language in a numerical format precisely. The processes of data mining, model selection, statistical learning, and sentiment analysis are integrated and consolidated all at once in this proposed research. For the case study of well-known electronic products from online review, the proposed methodology regularizes the unstructured text data, and sets benchmarks for choosing a optimal model beforehand. Afterwards, it enables to capture the product keywords that customer pay much attention on with the help of topic modeling, and thereby the sentiment intensities of each product attribute are computed using advanced pre-trained language transformer, S-BERT. Therefore, the attributes with their satisfaction derived from customer review are automatically identified, which can facilitate the improvement of product development, and also clarify the guideline of quality control in an industry to solve the real-life problems.

Graduation Semester

2024-05

Type of Resource

Thesis

Copyright and License Information

Copyright 2024 Szumin Chao

Owning Collections