Research On The Recommendation Algorithm Based On Knowledge Transfer Inside And Outside The Domain

In today's Internet era,information technology is rapidly iterating.Recommendation system is committed to helping people solve the problem of information overload and improve the efficiency of information processing.However,some established problems still exist in the field of recommendation,such as data sparsity,cold start and so on.The improved model proposed in this paper combines the idea of transfer learning and optimizes the traditional model in two aspects of intra domain recommendation and cross domain recommendation.The first is the improvement of the recommendation system in the domain.The benchmark model is a collaborative filtering recommendation algorithm based on matrix decomposition.According to the quality of user rating data,the whole system is divided into high-quality group and low-quality group,and then the hidden vectors of high-quality group are transferred to low-quality group to improve the overall recommendation effect.The second is the improvement of cross domain recommendation system.According to the direction of transfer learning,it can be divided into content-based transfer and model-based transfer.Content based migration is to use the common tag topics between domains as a bridge to connect two domains to share information.The model-based migration is based on deepfm model,and the parameters of depth model are shared among domains.Finally,the improved recommendation model is combined with intra domain and cross domain.The final recommendation model is Transfer Learning Recommendation consider knowledge inside and outside the domain(TLRec-CKIOD),which is obtained by combining the prediction results of the two parts of the model with linear weighting.In order to verify the effectiveness of the proposed model,three experiments are carried out in the empirical analysis part.The first is the user rating quality experiment in the domain,which proves that the migration between subgroups in the domain is conducive to improving the overall recommendation effect.The second is the model-based migration recommendation experiment,which proves that deep FM model is better than FM model and DNN model,and the recommendation effect after migration is better than that of single domain.Finally,the model effect of knowledge transfer inside and outside the domain is verified.The results show that the recommendation effect of the TLRec-CKIOD model is better than other comparison models.In the last part of the empirical study,TLRec-CKIOD model has a good accuracy and recall rate for the top 10 recommendation list generated by specific users.