Research On Personalized Knowledge Recommendation For Software Knowledge Learning Platform

With the innovation and rapid development of Internet technology,Internet information resources have increased dramatically.It is very difficult for users to accurately obtain the information in a large amount of resources.In the meantime,for Internet platforms,it is also very challenging to provide users with the applicable information services and improve the access efficiency.The recommender system has played a crucial role in alleviating information overload and been widely used in many Internet platforms such as e-commerce and online learning websites.With more and more Internet information resources,the application value of recommender system is getting higher.The key of the recommendation system is to model the user's preferences based on their historical access information(such as ratings and clicks),and to predict the user's current needs or interest preferences,thereby helping users to obtain the content they are interested in more quickly from massive data.For the software knowledge learning platform,the platform has a large amount of data and a variety of knowledge categories,such as video tutorials and documentary content.At the same time,the structure of knowledge data is different and the organization forms are different.Faced with the above problems,in order to provide personalized services,improve user experience,and enhance the interactivity between users and knowledge learning platform,we combine the two aspects of the platform side and user side to better make personalized knowledge recommendations.Specifically,on the one hand,we need to consider the organization and processing of heterogeneous data on the platform,such as using knowledge graph to organize heterogeneous knowledge items;on the other hand,we need to consider the differences of knowledge level between different users by modeling the user's current knowledge level information.In the current researches on recommendation,the RNN-based methods ignored the potential relations between interactive items and other items,and it is difficult to capture deeper needs of the users.In this paper,we combine the knowledge recommendation background of the knowledge learning platform scenario and hierarchical modeling for better knowledge recommendation,including short-term session-based recommendation model and recommendation models based on the user's knowledge level.For session-based recommendation,most of the early recommendation methods only considered the relations between the user and the item,which may cause the recommendation to perform poorly.For personalized recommendations based on the user's knowledge level,it is difficult for a recommendation model to capture deep dependencies across categories.Therefore,this paper conducts researches on the above issues deeply:1.A collaborative attention recommendation model based on sequence information and side information is proposed for session-based recommendation.This model combines Session Behavior Learning(SBL)and Side Information Learning(SIL)to better mine the user's potential preference information.After that,a collaborative attention mechanism is introduced into the model,which effectively combines the current sequence information and side information to obtain a better recommendation result.The difference between the session-based recommendation method proposed in this paper and the existing methods is that,first,our model effectively learns the user's direct preference information by embedding different interaction behaviors in the session;secondly,when processing the user's direct interaction items,it considers potential preference information outside the user's session.The potential side information of items interacting within and outside the user's session may have varying degrees of importance to the user's overall preferences.In contrast,the model proposed in this paper can more effectively combine user's intra-session preferences and potential preference propagation to make a more accurate next recommendation within a session.2.A sensitive recommendation model(KLAN)based on user knowledge level is proposed,which takes into account the categories in the knowledge system and the learning process of user discipline knowledge.KLAN mainly includes three main components,namely User Representation Module(URM),Candidate Representation Module(CRM)and Recommendation Module(RM).The user's current knowledge level is expressed to better mine the needs and preferences of the user and recommend more appropriate knowledge for the user.In addition,inspired by the hierarchical attention network in the NLP tasks,we have further improved the KLAN model and proposed a hierarchical attention recommendation model(HKLAN)based on cross-knowledge route sensitivity.The difference from KLAN model is that HKLAN uses a novel hierarchical attention mechanism for encoding and performs preference learning at two levels of knowledge items and categories.As different knowledge items have different contributions to the same knowledge category,different knowledge categories have different contributions to the knowledge level of a given user,so using a network based on a hierarchical attention mechanism to model can learn more deeply about the user's knowledge level representation and current knowledge requirements for more accurate knowledge recommendation.In the end,we select multiple real datasets and perform verification experiments on the proposed model.Compared with the currently popular recommendation algorithms,our proposed models all obtained better experimental results.