Research On Application Of Outlier Mining On Online Shopping

With the rapid development of e-commerce industry,people write a large number of comments in online shopping,and these review data which express consumer experience,are complex in structure and changeable in form,making it difficult for readers to find valuable information.However,for consumers,reading a number of meaningful and distinctive reviews can help them make a quick shopping decision before buying goods,which has great significance.For researchers,it has great research value to quickly select a set of commentary subsets from a large number of unstructured comments and push them to the users.The selection of a set of distinctive comment subset can be achieved by outlier mining technology.The word of this thesis mainly includes followings aspects:This thesis focuses on the research of outlier mining algorithm,and explores how to use outlier mining algorithm to make meaningful and characteristic mining for the review text of online shopping goods.Firstly,this thesis analyzes and studies outlier mining algorithms,and proposes an improved hyper sphere-based clustering outlier mining algorithm(IHCOM).Experimental results that this algorithm has lower time complexity and higher outlier mining accuracy.Secondly,on the basis of previous research on the review selection,the methods of feature extraction and the vectorized representation are studied in reviewing text.A method word frequency-based improved in feature selection,making the word frequency threshold value of the first-level feature words more reasonable.And the concept of feature groups is proposed,using the first-level feature words and the second-level feature words together as one feature group,which enhances vector expression ability of text conversion,and reduces text information loss.Finally,an outlier comment mining model that is suitable for online shopping product review mining is designed,and this model is applied to outlier review mining of Amazon Tablet product.Through experimental verification,it shows that the application of the mining model for outlier review selection is feasible,and the outlier review subset selected by applying the mining model is more effective in evaluation indexes such as feature coverage rate,emotional coverage rate,and feature group emotion total value.