Research On Commodity Recommendation Based On Opinion Mining

Nowadays,online shopping has become increasingly popular.When shopping online,users usually see comments to get a real understanding of the commodity.But there are up to ten million comments,so it is difficult for users to view item by item.Also the comments are not all useful.In this way users' desire to buy would reduce.So it is imperative for us to mine online comments and extract useful information to view to the users.To mine useful information from the comments is mainly to extract commodity attributes that comment sentence describes(i.e.evaluation objects),and the modifiers which modify the attributes(i.e.evaluation words).Then for each attribute,classify its whole modifiers to good or poor terms.Make a statistic of the percentage of every category and show these results to users.In this situation users can view intuitively the access information of each commodity attribute,and need not see the comments one by one.If we provide these attributes to users,they can choose the attributes on their preferences,and according to these chosen attributes we can get commodities ranking,then recommend commodities to users.This is the commodity recommendation based on opinion mining.In order to realize this goal,we have done following jobs in this paper:1)First of all,we extract evaluation objects and words based on the method of using noun phrase pattern to match comment text.There are some irrelevant objects to the commodity attribute and some different objects belong to the same attribute,so we need prune and cluster the evaluation objects.Because BIRCH algorithm determines cluster numbers by its own learning,we use it to cluster the objects and then delete the clusters which obtain few objects.In this way we reach the goal of pruning evaluation objects.After pruning process,this paper uses K-Means clustering algorithm to do global cluster for the left clusters and then get final attributes.We adopt the comment information of 284 notebooks in JingDong Mall as test data,and compare our method with two classic methods to verify the effectiveness of our method.2)Secondly we judge the polarity of evaluation word.This paper first judges the polarity of evaluation word based on emotion lexicon.For the words not existing in the lexicon,we compute the similarity of evaluation word and emotional seed words and use the similarity to determine the polarity of the evaluation words.Lastly we use K-Means clustering algorithm to classify the evaluation words and classify them to five categories: very good,good,general,bad and very bad.Also we verify the accuracy of our method and two previous frequently-used methods.3)Finally,we give different weight for attributes according to the order of user has chosen.Then based on multi-attribute decision-making algorithm,we compute the score of each commodity and gain the commodity ranking.This paper designs and realizes a commodity recommendation system based on opinion mining.This system processes and analyses comments and then show the final attributes.After user's choosing preferred attributes,our system views the commodities that meet users' attributes need.What's more,users can see the proportion and special comment sentences of good or poor comments of all the attributes.Also users can choose two commodities to make a comparison,and the system can show the attributes comparison histogram intuitively.