Research On F Eature Word Extraction Of APP Based On User's Comments

With the development of network technology and the improvement of public participation awareness,more and more people like to publish their own comments on the Internet,which contain a lot of useful information.At the same time,the explosive growth of APP constantly puts forward higher requirements for people's ability to process information.How to quickly extract valuable information from APP comments is a practical issue.This paper aims to extract the feature words of APP based on users' comments,so as to form a concise summary of APP features,and provide more concise and intuitive decision-making basis for APP developers,mobile application platforms and users.The feature words refer to the users' evaluation of the function or attribute of APP,and its form is "subject words+modifiers".In this paper,Baidu map on 360 mobile assistant is taken as the research object.Firstly,according to the different forms of the feature words contained in the comments,the comments are divided into explicit comments and implicit comments.Secondly,by comparing the differences between explicit comments and implicit comments on feature words extraction,combined with the existing algorithms,a feature word extraction algorithm based on part of speech path rules and TF-IDF weight is proposed,which is suitable for both explicit comments and implicit comments.The algorithm first analyzes the part of speech path of the comments,and counts the high-frequency part of speech combinations of the feature words.Then,TF-IDF algorithm is used to extract the high-frequency words in each part of speech involved,and the high-frequency words are combined based on the part of speech combination template and semantic similarity.At last,the combined candidate feature words are clustered to get the APP feature words.Finally,the extraction effect is measured from the three aspects:accuracy,semantic independence and comprehensiveness.Experimental data show that the accuracy and semantic independence of feature words extracted by this algorithm are very high,which indicates that the algorithm has application value.