Research And Implementation Of Music Hot Comment Hybrid Recommendation System For Cold Start Users

The recommendation system is an important method to obtain the required information from massive information.recommendation system can connect users with information.On the one hand,recommendation system can help users to quickly find the information they need.On the other hand,recommendation system makes information to be displayed.It can achieve a win-win situation between information consumers and information producers.However,the recommendation basis of most personalized recommendation systems at present is user historical behavior data and user real-time behavior data.Therefore,when the amount of user data in the recommendation system cannot meet the recommendation needs,it will lead to the user's cold start of the recommendation system.Therefore,the recommendation system needs to distinguish and recommend the user's cold start problem to avoid the decrease of the recommendation effect caused by the user's cold start problem and improve the user's satisfaction.The problem of user cold start is also a common problem in the current recommendation field.In order to solve the problem that the user's cold start has no historical behavior or insufficient historical behavior,this paper based on the construction of the music hot comment recommendation platform,researched the user's cold start problem caused by the sparse user data.Use the asynchronous web crawler to obtain the hot music reviews and related data on the current network platform,and use a variety of recommendation algorithms to build a mixed recommendation engine,so that the cold start users of the hot music recommendation platform can also get personalized hot music reviews,satisfy Cold start users' demand for hot reviews of music improves the personalized recommendation effect of the recommendation system for cold start users.First of all,this paper studies the solution mechanism of the user's cold start problem and the user's demand for music reviews.In addition,the paper analyzes the functions of various web crawler frameworks,and constructs an asynchronous crawler system based on the Scrapy framework,combining the MongoDB(distributed file storage database)and SQLite(relational database).This system can acquire all kinds of data needed at a fixed time to meet the data requirements of the platform.Secondly,to solve the problem that the accuracy of the single algorithm is too low for cold-start user recommendation,a hybrid recommendation algorithm based on KNN algorithm,FM algorithm and Bandit algorithm is proposed.The algorithm uses the acquired music hot evaluation data to extract features and build a mixed recommendation model.The hybrid algorithm first uses a recall strategy,using the introduction of time factor popularity list,god comment,similar label based on KNN algorithm,and similar comment text based on the combination of Doc2Vec and KNN algorithm to generate different recall sources for users;then the recalled comments are mixed Perform deduplication and filtering to form a rough list;finally set the "cold start status switch".For users who do not have enough historical behavior,the recommendation platform automatically uses the improved Bandit algorithm to form a recommendation list.For users who have enough historical behavior,recommend the platform The FM algorithm is automatically used to form a recommendation list,and the platform selects TOPK hot comments from the recommendation list to provide to users.The hybrid recommendation algorithm can provide personalized recommendations for cold-start users and achieve good results.Finally,design and implement a hot music recommendation platform.The platform is developed based on the Web framework Bottle and embedded in a hybrid recommendation system for user's cold start.It can provide personalized music hot recommendation service for cold start users.Demonstrate the application of this hybrid recommendation system in the actual environment.