Research And Implementation Of Recommend System Based On Text Analysis

With the rapid development of Internet Information Technology and the popularity of Network,an increasing number of consumers shop and participation review online.Nevertheless,facing wide variety of similar products on the e-commerce platform,consumers have to spend much time on reading product reviews for making decision.It is difficult for consumers to filter out the products which meet their requirement on the product features in a short time.The result shows that the product features which extract from product reviews veritably reflect the actual object which consumers concern about,through the analysis of product reviews.Therefore,in order to provide consumers with efficient and convenient shopping experience,this thesis designs and realize a product recommend system,which contains extracting product features,ranking the product reviews in order of utility value,sentiment analysis and computing the recommend value,based on the analysis of product reviews.This thesis contribution mainly concentrates on the following parts of research work:1.In the aspect of Extracting Product Features,owing to the candidate product features including of massive similar meaning words which extracted by the sequence of part of Speech(POS)template,a words clustering method which is called Affinity Propagation Clustering based on Cosine Similarity(APC-CS)is proposed.This method can find a word to express different words which have similar meaning.That is to enhance shopping experience.2.In the aspect of Product Reviews' Utility,through the analysis of the factors which influence the utility of product reviews,the original text features are put forward.And combined with the traditional text feature extraction method-Information Gain(IG),a new model for evaluating product reviews' utility based on Combination feature is proposed.3.In the aspect of Sentiment Analysis and Computing Recommend Value,in this thesis,the Long Short-Term Memory Model is used for Sentiment Analysis.And combined with the sentiment value and product reviews' utility value,the model which are used to compute recommend value are constructed.Based on the above method to analysis product reviews,we realize a product recommend system.With this recommend system,consumers can select the product features that they most attention to in the feature bar for similar products.According to the analysis result on product reviews,the product list which ordered by Recommend value in turn will return.This recommend method can save much time for consumers on making purchase decision by comparing product reviews.It also help e-commerce platform enhance the service level,and improve the shopping experience.This system has certain application prospect and commercial value.