<![CDATA[The rapid development of the e-commerce ecosystem necessitates an information retrieval system that not only relies on text relevance but also quantitatively considers the quality of user opinions and sentiments. Therefore, this experimental research aims to build and evaluate the performance of an information retrieval system on e-commerce product reviews by implementing a Sentiment-Aware Scoring approach. This innovative approach integrates the Term Frequency-Inverse Document Frequency (TF-IDF) weighting method, the lexicon-based sentiment analysis Valence Aware Dictionary and sEntiment Reasoner (VADER), and user rating normalization. The experiment was conducted using all 568,454 review documents sourced from the Amazon Fine Food Reviews Dataset. System performance evaluation was carried out by comparing a pure TF-IDF-based Baseline model against a Proposed model that combines TF-IDF, VADER, and ratings on the top ten search results (Top-10) using a pseudo-relevance labeling approach. The experimental results show that the integration of sentiment and ratings effectively overcomes the weaknesses of traditional methods, which often display low rated products merely due to lexical relevance. The Proposed model was proven to significantly improve search quality, indicated by an 11.54% increase in the Precision@10, Recall@10, and F1@10 metrics respectively, as well as a 4.13% surge in the NDCG@10 score, reaching a perfect value of 1.0000. In conclusion, the application of the Sentiment-Aware Scoring framework is highly effective in presenting search results that are structurally relevant while holistically representing consumer perception, thus holding great potential to enhance user satisfaction during online shopping.]]>
Copyrights © 2026
