Research And Implementation Of Application Store Spam Recognition Based On Deep Learning

With the continuous development of the Internet and communication technology,people have entered the era of mobile Internet from the PC Internet era,and smart phones have occupied an important position in people's daily lives.The mobile application store is an application distribution platform that is used to display and download application software,so that people can choose the appropriate application according to their needs.Software reviews in the application store are the reference standards that people prefer when choosing software.Provides a lot of reference value.However,due to the open nature of the Internet,some users will post some spam comments.These comments not only affect the user experience,but also are not conducive to the integration of information resources.Therefore,it is very important to maintain a good environment for the software review area.In order to solve the above problems,this article comprehensively uses web crawlers,deep learning technologies and experimental comparisons to conduct research.The main contributions of this article are as follows:(1)Aiming at the problem of the lack of Chinese spam comment data sets,this article constructs a Chinese application store comment data set that can be used for the task of identifying spam comments to prepare for the subsequent research on deep learning technology.Aiming at the server's anti-crawler mechanism,this article proposes a Request-based closed-loop crawler architecture,which solves the server's anti-crawler limitation and improves the efficiency of data collection.(2)Aiming at the low efficiency of traditional sensitive word filtering and manual review,this article combines the BERT model and Attention mechanism to build a Bert-AtFnn spam comment recognition model,which not only improves the efficiency of spam comment recognition,but also improves spam comments recognition effect.In view of the fact that the available information of a single pre-training model is relatively single,and the advantages of multiple pre-training models are further integrated,a dual-channel fusion Bert-AtFnn spam comment recognition model is built,which can better extract context information and better realize the recognition of spam comments.(3)In order to test the performance of the Bert-AtFnn spam comment recognition model and the dual-channel fusion Bert-AtFnn spam comment recognition model proposed in this article,model effectiveness experiments and model horizontal comparison experiments were designed to compare with the LSTM model and the BiLSTM-Attention model.This article selects 3 groups of software reviews of different types of applications as a data set for model training.In comparison experiments,the accuracy and F1 value of the Bert-AtFnn model and the dual-channel fusion Bert-AtFnn model in the three sets of data are higher than those of the LSTM and BiLSTM-Attention models,which proves that the spam comment recognition model constructed in this article can better identify spam comments.It provides a certain reference meaning for subsequent research and actual scene applications.