Research On The Application Of Intelligent Deduction In User Portrait

With the tremendous advancement of modern Internet techniques,the Internet has become an indispensable tool for people to obtain various information,such as obtaining information on the Toutiao app and purchasing items on Taobao.However,now entering the era of big data,the information that people face is growing rapidly.It is an imperative task for Internet companies to analyze this large-scale and mostly worthless information,and construct user portraits based on this,to complete accurate services to users.Nowadays,network security issues are frequent and user information is leaked in large quantities,which has caused the increase of Internet users' security awareness.Therefore,Internet users will actively hide certain characteristics of themselves,such as gender and age.Although this will protect users' information,it creates obstacles for Internet companies to provide users with personalized services.Furthermore,due to the user's hiding of their characteristics,the obtained user information only contains part of the user's characteristics,which results in the inability to fully and accurately display the user's characteristics when drawing the user's portrait.Incomplete user portraits will not play a sufficient role in the personalized recommendation and intelligent dialogue systems,resulting in poor practical effects of the entire system.Therefore,drawing a comprehensive user portrait is of great significance.This paper studies the above problems and analyzes the problems of traditional classification prediction methods.This paper proposes a neural network model to complete the prediction.This model combines the text information of Weibo tweets and users' social information and uses an attention mechanism.,And then combine traditional machine learning models such as logistic regression and random forest to improve the prediction effect of the model.On this basis,this paper proposes a generated dialogue model combined with user information.The model generates diverse answers by adding user personal characteristics for training.Here,the coding and decoding of the generative dialogue model adopt the LSTM architecture,and the text generation effect of the model is improved by beam search calculation.In the experimental part,this paper uses the data set crawled on Sina Weibo to train and evaluate the proposed classification prediction model.The experimental results prove that the accuracy of the integrated classification prediction framework proposed in this paper is 3%higher than that of the traditional framework.After that,the classification prediction results and the text of the Weibo data set are applied to the generative dialog frame,and the generated word non-repetition rate is used as an evaluation indicator.The final result shows that the text generated by this model is 5% higher than other models in the non-repetition rate indicator.This fully shows that the diversity of this model is very effective.