Accurate Rating Prediction Based On Deep Learning And Field-aware Factorization Machine

With the development of the Internet age,the rapid increase of information has led to the phenomenon of "information overload".On the Internet,hundreds of millions of users continuously generate a large amount of data every day.How to use these data and extract useful information from it has become the common goal of enterprises and scholars.Under this premise,under this premise,a personalized recommendation system appeared.The personalized recommendation system can analyze the users past behavior characteristics and interest preferences,create an interest model that meets individual needs,and use the model to quickly and accurately find the information that users are interested in.The application range of the personalized recommendation system has been More and more widespread.With the birth of mobile phones,users entertainment behaviors tend to become more fragmented,and new media videos have gradually become users preferred entertainment methods.It is suitable to introduce a recommendation system in the video field.Accurate video recommendation can not only improve the user experience,but also bring traffic to the video website so as to better create greater commercial value.Therefore,research on personalized and personalized video recommendation systems that meet the interests of users has great research value and practical significance.At present,the common classical recommendation algorithms mainly include collaborative filtering based recommendation,content-based recommendation and mixed recommendation.Traditional recommendation algorithms have achieved good results in personalized recommendation,but in different algorithms,sparse data,cold startup and high labor cost will appear.These problems can have a negative impact on the recommended results of the model.Therefore,this paper adopts the deep learning method to extract various heterogeneous data,and proposes a structural model that can be used to overcome the above problems by integrating multiple attention mechanism and domain factorization machine,and applies this model to the field of video recommendation.For video project features and user respectively,using natural language processing the text in the convolution neural network and Word2Vec technology,characteristics of the title of the video,tags and other text processing so as to reduce the cost of artificial to extract the characteristics of the engineering and computational complexity,and is applied to the output fusion long attention mechanism and factorization machine in the field of video score prediction method.This paper mainly does the following work:(1)The method of deep learning can learn the essential characteristics of the data.This article combines personalized recommendation with deep learning,and proposes an accurate scoring prediction method that combines deep learning and domain factorization machine,taking user characteristics and item characteristics as Input data,extract features through deep learning methods,and then use multi-head attention mechanism and domain factorization machine hybrid recommendation method for item recommendation.(2)Apply the model to the field of video recommendation.For text description data such as video titles and tags,first convert them into distributed word vectors through the Word2Vec technology in natural language processing and then input them into the text convolutional neural network.Feature extraction,then combined with other project features and user features to form input data for processing.(3)For the accurate scoring prediction method that integrates deep learning and domain factorization machines,evaluation indicators such as root mean square error,standard average error,and average absolute error are used to test the performance of the method.By using the algorithm of running this article and using the user-based collaborative filtering algorithm for comparison,the data can be evaluated more accurately through the score prediction,and the feasibility and effectiveness of the model are also verified.This is also useful for solving data sparseness play a big role.