Extracting Feature Words From Customer Reviews

With the further development of the Internet,more and more people buy products and share their reviews on e-commerce sites.The number of reviews online has increased over the years.The useful information hidden in reviews helps potential customers decide whether or not to buy the products.People have become accustomed to browsing the reviews of other customers before buying a product.Manufacturers also want to collect meaningful feedback from customers,so that they can know better about customer needs and preferences.It is very hard for customers and manufacturers to get useful information from a large number of comments quickly.Thus,automatic information extraction in reviews has become a significant problem.This thesis investigates feature word extraction.Feature words are product components or attributes indicating customer interests.In order to provide a systematic study on feature word extraction,this thesis first studies related works and algorithms.After that,a new approach is proposed,the rapid feature word extraction(RFWE)method,to improve the performance.The main work is shown below:1)Implement and partly improve three popular text extraction methods.They are:the frequency-based extraction method,the Web PMI-based extraction method,and the rapid automatic keyword extraction(RAKE)method.Unlike the original algorithms,this thesis chooses a more efficient text analysis tool,Stanford CoreNLP,to parse the reviews.Besides,a local database is designed to store the hit counts from Google search engine.The algorithms can recognize nouns with high frequencies and calculate the relevancy of words in the context.2)To provide an objective evaluation,the performance of each method is validated and compared from the following aspects:precision and recall,time complexity,and robustness.3)According to the extraction methods and their experiment results,RFWE method is proposed combining the techniques used in the popular methods,such as the parsing technique in RAKE method,and the pruning method in the frequency-based method.The new method performs well in precision,recall,and runtime.