Recommendation Algorithms For User Behavior Data Modeling

With the development of the Internet,the amount of user behavior data collected by online companies is increasing,and as a result,recommendation algorithms for user behavior data modeling has become more and more popular.Modeling user behavior data makes it possible for the businesses to learn more about users' preferences so as to satisfy customers' need and to increase profits.However,in real-world scenarios,user behavior data are diverse,and different algorithms are developed to model different types of user behavior data.As far as we are concerned,most of the classic recommendation algorithms for user behavior data modeling are collaborative filtering recommendation algorithms with matrix factorization on user feedback data,which can be categorized as explicit feedback data modeling and implicit feedback data modeling.As a relatively simple method,matrix factorization provides a powerful tool to model sparse data that only learns the global information about users and items.However,the problem of user behavior data modeling becomes more and more difficult due to the increasing complexity of user behavior data,which are the main challenges when improving the classic recommendation algorithms.In order to address the above problem,in our studies we focus on modeling users' explicit feedback,implicit feedback,and their consumption data.In this thesis,we summarize our contributions as follows.Firstly,we propose a novel context learning algorithm to learn users' preferences.Specifically,we introduce memory-based information of both users and items into three classical recommendation models and improve three new models with better performance.Secondly,we extend the factored item similarity model to obtain better results by incorporating user neighbor and item neighbor information as auxiliary information.Finally,we apply feature engineering techniques on user consumption data from Alibaba Tianchi contest to explore the problem of customer flow prediction on Koubei and then devise algorithms to model consumption data by combining various machine learning techniques.