A Novel Framework for High-category Coverage Clothing Recommendation System Based on Sentiment Analysis

Users are always accustomed to checking others' reviews to determine if the product meets their expectations before purchasing clothing online. Sentiment analysis (SA) technology can effectively identify emotional feedback from numerous reviews and help users and manufacturers accurately identify product defects. However, traditional SA techniques, such as sentiment lexicons and machine learning, have limitations when dealing with large-scale datasets, it is challenging to identify clothing defects or provide extensive category recommendations accurately on high-category coverage clothing. To address this issue, this study proposes a new framework for a fine-grained feature-level SA-based high-category coverage clothing recommendation system (HCCRS). We constructed a dataset containing 82,832 clothing reviews and extracted nine clothing features that users are concerned about from questionnaires and the BERT model. We designed a hybrid SA method combining BERT and SentiStrength and built a relationship model based on feature weights and sentiment scores. The experiment results show that our method outperforms traditional lexicon-based methods by 10-25% and improves by 3% compared to BERT alone. HCCRS introduces a personalized and more authentic approach, offering a fresh perspective for clothing recommendation researchers and practitioners.