Abstract: In order to effectively promote App function iteration,most existing studies generally focus on improving existing functions or adding new functions to promote version update by mining user requirements in user reviews,while neglecting to identify functions that should be eliminated from user reviews.To address the issue,an analysis method about user data-driven App fading functions is proposed.User reviews from App market are collected.Keyword templates are built to filter out reviews that contain fading functions.From these reviews,function phrases are mined by syntax paradigms.A classifier is trained by these phrases to identify fading functions to be studied,so the dataset of fading functions is built.Lifecycle of a fading function is found based on version update log and user reviews backtracking.Then,user reviews for the lifecycle of fading functions are analyzed.A word weight threshold method is proposed to detect and correct false ratings based on text sentiment analysis.BERT algorithm is used to classify the review data.BERTopic-Corex topic model is proposed to generate theme words of reviews.Key user reviews are identified based on the previous analysis results and the word count of reviews.Thus,fading functions can be effectively identified and analyzed from user reviews.Experimental results and examples prove the feasibility and effectiveness of the proposed method.

Key words: Fading function, User review, Review classification, Fake rating, Topic model