Product Feature Extraction via Topic Model and Synonym Recognition Approach

EasyChair Preprint 1307

Abstract

As e-commerce is becoming more and more popular, sentiment analysis of online reviews has become one of the most active areas in text mining. The main task of sentiment analysis is to analyze the user's attitude towards different product features. Product feature extraction refers to extracting the product features of user evaluation from reviews, which is the first step to achieve further sentiment analysis tasks. The existing product feature extraction methods do not address flexibility and randomness of online reviews. Moreover, these methods have defects, such as relying on labor, low accuracy and recall rate. In this study, we propose a product feature extraction method based on topic model and synonym recognition. Firstly, we set a threshold that TFIDF value of a product feature noun must reach to filter meaningless words in reviews, and select the threshold by grid search. Secondly, considering the co-occurrence rule of different product features in reviews, we propose a novel product similarity calculation, which also performs weighted fusion based on information entropy with a variety of general similarity calculation methods. Finally, compared with traditional methods, the experimental results show that the product feature extraction method proposed in this paper can effectively improve F1 and recall score of product feature extraction.

Keyphrases: LDA, Product Feature Extraction, Shopping Reviews, synonym recognition

@booklet{EasyChair:1307,
  author    = {Jun Feng and Wen Yang and Cheng Gong and Xiaodong Li and Rongrong Bo},
  title     = {Product Feature Extraction via Topic Model and Synonym Recognition Approach},
  howpublished = {EasyChair Preprint 1307},
  year      = {EasyChair, 2019}}