DOI: 10.18178/wcse.2025.06.010
Product Review Summarization System Extracting Evaluation Criteria and Visualizing with Supporting Sentences
Abstract— With the widespread use of review websites, consumers increasingly rely on product reviews. However, when reviews are excessively long or numerous, the useful information obtained per unit of reading time decreases, increasing the burden on users. To address this issue, this study proposes a new product review summarization system comprising the Evaluation Criteria Extraction Module, the Polarity Analysis Module, and the Supporting Sentence Extraction Module. First, morphological analysis is applied to reviews to extract evaluation criteria with AI assistance. Next, the sentiment polarity of sentences containing these criteria is determined, and the proportion of positive and negative opinions is visualized using bar graphs. Additionally, relevant sentences are displayed below the graphs to retain essential information while improving review readability. Performance evaluation demonstrated that the proposed system effectively reduces the burden of reading reviews without significantly compromising the amount of information or user interest in the product.
Index Terms— Product review, Natural language processing, LLM, BERT
Tanzan Rikuya, Eiji Kamioka, Chanh Minh Tran, Phan Xuan Tan
Graduate School of Engineering and Science, Shibaura Institute of Technology, JAPAN
Cite: Tanzan Rikuya, Eiji Kamioka, Chanh Minh Tran, Phan Xuan Tan, "Product Review Summarization System Extracting Evaluation Criteria and Visualizing with Supporting Sentences", 2025 the 15th International Workshop on Computer Science and Engineering (WCSE 2025), pp. 58-63, Jeju Island, South Korea, June 28-30, 2025.