Improved TF-IDF-Based LDA Topic Clustering Model for Specific Commodity Contexts and Different Sentiment Tendencies

Abstract

As Natural Language Processing (NLP) is increasingly used in Internet content platforms, more businesses and organizations are concentrating their research and development efforts on NLP. Effective user review data mining has the potential to greatly accelerate the transition of digital products, particularly in e-commerce. The information represented by the review texts of various shopping categories differs, but due to the 'generality' of their construction models, many traditional NLP text libraries frequently make judgments about a given context that are incorrect. This error is primarily due to a lack of a precise understanding of the 'characteristics' of various domains. This research chooses to review data for three different product categories (clothing, cosmetics, and laptops) with distinct contexts on the Jingdong e-commerce platform based on the aforementioned problems. To analyze and examine the classification of consumers' subject terms under different emotional tendencies for two different contexts of negative text and positive text, we propose the ED-LDA (Emotion discrimination - Latent Dirichlet Allocation) topic model. The findings demonstrate the significance of subject word grouping under various sentiment trends and context-specific characterized natural language analysis for social media review analysis of niche goods and subcultural circles.