Research And Application Of Recommendation Algorithms In Recommendation System

With the rapid development of the Internet,the amount of information has begun to soar rapidly,and users will inevitably browse a lot of useless information.They cannot browse the information they really want in the face of information overload.The recommendation system plays a role in filtering information in solving the problem of information overload,and provides recommendation services to users based on the recommendation algorithm.At present,the recommended algorithm faces problems such as insufficient prediction accuracy and cold start.In this thesis,the research and analysis of the recommended algorithm and the engineering application of the recommendation system are combined to discuss the practical problems,and the following aspects are studied:(1)The recommendation engine streaming system is designed to realize the real-time processing of the comment log.This thesis studies the recommendation algorithm in the recommendation engine that conforms to the business scenario,designs three modules of the recommendation engine:receiving request,processing request and returning result,and realizes the Client connecting the Server to send the request,and the Server returns the result according to the request.(2)The K-Means,MiniBatchKMeans and BIRCH algorithms are used to predict the user's score on the movie,and the prediction accuracy of the three clustering algorithms is verified.For the K-Means algorithm,the direction of algorithm optimization is given.After re-understanding the data,the original discrete data are centralized by using the processing time information,and the better prediction result of K-Means algorithm is obtained.The collaborative filtering algorithm was studied to find the recommendation results related to the user or the item by calculating the cosine similarity between the user or the item.The Apriori and FPGrowth algorithms of association rules are studied.The recommendation system is formed by integrating the algorithm and the recommendation engine,and the list of recommended movies is generated.(3)The TensorFlow fitting prediction model was studied and the shortcomings of the fitted model were analyzed.The mnist data set of deep learning technology training is studied,and it is verified that the number of model iterations can affect the accuracy of the algorithm.The training data of Wide And Deep model is studied,and it is verified that the Wide And Deep model combined with Wide and Deep features can show better training results.The prediction model is built in the music recommendation scene,and the song list recommendation for the user is realized.The errors of different recommendation algorithms such as collaborative filtering and matrix decomposition are evaluated,and some reasons for the errors are given.The Word2Vec algorithm was introduced,and using the method of song sequence modeling to realize the recommendation of similar songs.Finally,the solution to the cold start and user interest prediction problems is given.In this thesis,the research on recommendation system related technology and recommendation algorithm can meet the business requirements in the actual project of recommendation system.Research deep learning is used to recommend the system,and it has certain reference value to solve the recommendation sorting problem.This thesis is of great significance for solving cold start and user interest prediction under the music recommendation scene.