Research On Recommendation Algorithm Integrating Knowledge Graph And Deep Learning

The advent of the era of big data has brought a large amount of information to people’s lives.How to select effective information from a large number of information and further recommend it to users has become an urgent problem to be solved,and the recommendation system was born.The major breakthrough in the research of knowledge map and deep learning in various fields makes it introduced into the recommendation system as a key technology to provide more solutions for the application of recommendation algorithm in different scenarios.However,at present,most multi task learning recommendation systems generally have problems such as sparse data,difficult memory ability and generalization ability,and lack of research on user dynamic interest transfer.This study proposes a recommendation algorithm integrating knowledge map and user behavior,which comprehensively considers the static and dynamic characteristics of users.On the one hand,it alleviates the data sparsity of the recommendation system,on the other hand,it provides a way for the recommendation system to dynamically consider the changes of user interest information.Firstly,aiming at the problem of sparse data and lack of user-side entity analysis in the current multi task alternating learning model,a multi task alternating learning recommendation algorithm model based on user-side attribute knowledge map enhancement is proposed.This method introduces the relevant knowledge map information into the multi task recommendation model.From the perspective of users,the knowledge map is used as auxiliary information for cross learning with the original user information,which is used as the initial vector of the model.This method makes full use of the methods of knowledge map feature learning and recommendation system module feature learning,and takes the tasks connected by the knowledge map and recommendation system through a part of the same information as a common task,which effectively alleviates the problem of data sparsity.Finally,experimental demonstration is carried out in CTR prediction scenario and Top-K recommendation scenario,which is improved compared with the benchmark method.Secondly,aiming at the problem that deep learning can not reasonably explain the recommendation results,and the knowledge map multi task recommendation system only considers the user’s static information,a recommendation model integrating user-side knowledge map and user behavior learning is proposed.This method designs a new model framework based on the cyclic neural network variant GRU and attention mechanism in the deep learning model,and tries to model the user behavior sequence.This method can not only alleviate the data sparsity,introduce dynamic interest changes into the knowledge map recommendation system,but also increase the interpretability of the deep learning model.Experiments show that this method improves the recommendation performance.