Research On User Opinion Mining In Online Reviews

With the development of the Internet,online e-commerce or online group buying websites has grown exponentially.Online reviews contain important information about products or merchants,have important effect for potential customers' purchase decisions and business management.Therefore,online review mining has a wide range of applica-tion prospects,and its research has important guiding significance for short texts mining.In order to solve the problem of information overload of reviews,we first consider the retrieval of reviews.With user queries,Information Retrieval can help users find the most relevant product reviews.However,the format and expression of comment is free and user's emotional attitude toward product attributes is different.Traditional methods ignores the connection of words in the commentary text and therefore needs to be improved.In addition,to make use of user opinions in reviews,we use the anomaly detection of reviews to obtain explanations for changes of business visits.However,individual user reviews are incomplete and the motives behind the different commentary content may also be related.Existing research lacks a comprehensive analysis of all the comments,so it is necessary to combine the application of targeted methods and to combine other data.This paper proposes an retrieval model that combines query word distances.It em-beds the query word distance into the word frequency estimation framework of the Lan-guage Model.We designed several strategies to use the paired query distance for different conditional probability,and integrate the comments at the commodity level.In addition,with the review under the condition of abnormal traffic conditions,a model is built to obtain the distribution of word features under abnormal conditions.We study the correlation between visits anomalies and utilize the Multi-task Learning method to study the distribution of word features with different access anomalies.We verified our works on public datasets and datasets of a shopping mall respectively.