Research On Application Of Deep Neural Network In Film Review Big Data

The prediction of user’s basic attributes mainly includes the classification of basic information such as the user’s gender,age,occupation,geographical location,and education level,etc.The research of user’s basic attribute prediction is one of the hot issues in the field of machine learning,and it is also a research hotspot of big data application.The research topic of this thesis is to predict the user’s basic attribute information by analyzing the user’s rating data of movie in the movie review website.Through deep learning theory,matrix factorization techniques and multi-task learning methods,a variety of theoretical methods are used to solve the problem of user attribute classification.The single task learning method and multi-task learning method for user’s attribute prediction based on deep neural network are proposed.The main work is as follows:Based on matrix factorization techniques,this thesis proposes a method of filling sparse user rating data.The user’s rating data for movie is taken as the attribute feature of the user.To overcome the sparseness of the input data,the sparse input data is filled based on an implicit semantic model.This thesis analysis the forms of user’s rating data and proposes a single-task model of deep neural network based on matrix factorization.The user’s filled rating data can be used as the input of the neural network.In combination with deep learning techniques and multi-label learning,we can extract feature of user’s rating data,and learn data’s abstract features.Through the comparative analysis of commonly used classification algorithms,the effectiveness of the algorithm is proved.This thesis proposes a multi-task learning method based on deep neural network.Through the analysis of user’s attributes,the thesis finds the correlation between different attributes of users,and proposes a multi-task model for simultaneously learning and predicting multiple attributes based on the relationship between user’s attributes.Based on the original single-task learning method,a multi-task learning model is proposed by representing the sample tags as structured data.Simultaneous prediction of multiple attributes can be achieved by sharing model parameters between different attribute tasks.The comparative analysis of various methods shows that theproposed method is effective.