Research on Similar Product Recommendation Based on Sentence-BERT and LDA

There are hundreds or thousands of similar products in the market, with similar product ratings. Different consumers have different preferences for the products. Therefore, presenting consumers with evaluation information on fine-grained attributes of similar products from different merchants is more beneficial for consumers to make purchasing decisions. This study combines the Sentence-BERT word embedding model with the LDA model. Firstly, convert the online comment text into a set of word vectors with richer semantic information; Secondly, the vectorized word set is fed into the LDA model for topic feature extraction; Finally, the BosonNLP sentiment dictionary is used to calculate the sentiment values of various thematic features of similar products from different merchants, in order to analyze the advantages and disadvantages of similar products from different merchants and recommend high-quality merchants to consumers. This study compared the methods of combining other word embedding models with LDA models and commonly used topic extraction methods. The experimental results of this study improved the granularity and accuracy of topic segmentation, and achieved good topic consistency.