Research On Product Recommendation System Based On Review Analysis

With the popularity of the internet,e-commerce get rapid development，variousrecommendation system become an important work.Existing systems make products asthe cener,and recommend more products to customers by their browsing history.In mostcases,it is useful,because it can provide us many useful goods that we didn’t care.Butusually,the cutomer go shopping with a certain purpose.The shopping experience will bereduced if they bought unsatisfactory products.So,in order to make the customers getsatisfactory products,we use a new technique based on reviews analysis andrecommend right goods to the customers by mining useful information in the reviews.The following are the main contributions:①Get explicit comment target.Existing methods construct syntax mode based onnoun and adverb.we add verb form of the syntactic pattern on the basis ofcommon Chinese comments by analyzing lots of product reviews.Thus,we canextract explicit comment target more effectively.②Get implicit comment target.Existing research focus on explicit comment targetand pay little attention to implicit comment target.But in fact,there are manyreviews that don’t any explicit comment target(only sentiment words)..In orderto get implicit comment target,we build a mapping of comment target ontoreviews and can get implicit comment target automatically.③Noumalize the identification process of the product attributes and refine therecognition results.Existing research usually compute the polarity of thecomment target respectively.it will produce too many results and can’t show thecustomers a more clear results.After we compute the polarity, we also build aproduct feature dictionary and corresponding product feature treecollection,then we use a standardized attribute to instead of the non-standardattributes and sub-attributes.At last,we use clustering algorithm to cluster theattribute,thus we can get a more clear results.④Verify the key steps of the review analysis by experiment and prove therationality and effectiveness of the methods that we use.At last we design thefinal recommendation system. The customers can set different priorities of eachproperty according to their level of interest in each property. The system canprovide a final sort result by using the reviews analysis and multiple attribute dicesion method.At the same time, the customer can look over the evaluation ofthe product in the sorting results.Our system is a prototype system and at the present time,it can be used incellphone domain.The customer can get satisfactory products by using our system.So ithas importmant using value.