Research On Transfer Learning-Based Personalized Recommendation Algorithm

The emergence of the Internet and popularization has brought people a lot of information and makenpeoples’ lives convenient and effective.However,too largeamount of information also brings another problem information overload. Personalized recommendation system asan important means to solve the problem of information overload, becomes a research hotspot in recent years. But the current personalized recommendation system also existsmany problems, including data sparseness, cold start problems, extensible problems, etc. Thereby,solving these problemsbecomes a emphasis and difficulty in the research field of personalized recommendation system.Transfer learning,as its name implies, learning the knowledge from the source domain and use the knowledge to solve the problems in the target domain. If we can apply transfer learning in the field of personalized recommendation, The problems of cold start and data sparseness in the field of personalized recommendation system can be solved.This paper studied some typical algorithms in the field of transfer learning andpersonalized recommendation。 Paying more attention on the TradaBoost algorithm, and do the modified on the traditional collaborative filtering algorithm. On this bias, this paper proposes a new collaborative filtering algorithm with transfer learning:TradaBoostCF.The algorithm is used do the personalized recommendation in target domain which the use-item rating data is very sparse. It is through the using of auxiliary data to accomplish classification in the target data set, then using the improved collaborative filtering algorithms to dothe recommendation for the customer, the experiments showed that:TradaBoostCF can achieve a higher recommendations quality and can effectively solve the data sparseness problemby using the auxiliary data set.