Research And Design Of Radio And Television Content Recommendation Algorithm Based On Knowledge Graph

With the service integration development of telecommunication network,radio and television network and computer communication network,there are more and more contents in the application of radio and television system,and the problem of information overload is serious.Providing accurate recommended contents for users is conducive to improving the service quality of radio and television.As a kind of knowledge structure,knowledge graph has been widely used in recommendation system to provide rich auxiliary information for downstream tasks.This dissertation analyzes the characteristics of application scenarios in the field of radio and television,and tries to introduce knowledge graph into the field of radio and television system to solve the problem of content recommendation.This dissertation mainly completes the following four aspects.1.This thesis proposes a radio and television content representation algorithm based on knowledge graph and user interaction features,which combines knowledge graph and user interaction to learn the content representation.This thesis creatively constructs the time session behavior type matrix to calculate the time behavior session similarity,divides the personalized session length for each user according to the similarity,and extracts the user interaction session sequence.After constructing the knowledge graph of radio and television content,the model applies node2 vec algorithm to extract the content sequence from the knowledge graph.The combination of random walk content sequence and user session-based interaction sequence is used as the input learning sequence of item2 vec model.This method is verified by combing the radio and television system data,analyzing the radio and television knowledge graph and content representation.2.In order to solve the problem of content recommendation by modeling the time drift of user preferences,this thesis proposes a content recommendation algorithm for radio and television based on knowledge graph and users' long-term and short-term preferences,namely KSHAN(knowledge enhanced sequential hierarchical attention network)algorithm.In this model,the user behavior sequence is decomposed into content sequence and behavior type sequence,and the knowledge representation of content and behavior type are embedded in the model for splicing and convolution network layer mapping.The user's behavior characteristics are introduced into the two-layer attention network to capture the user's interest changes and model the user's long-term and short-term preferences for recommendation.The effectiveness of this method is verified by the training of application scenario data of radio and television system.3.In order to deeply mine a large amount of implicit feedback behavior information and dynamically capture user preferences in radio and television scene applications,this thesis proposes a radio and television content recommendation algorithm based on knowledge graph and user polymorphic behavior sequence modeling,namely KBT(knowledge enhanced behavior transformer)algorithm.The model further divides the user's implicit feedback behavior into discrete and continuous micro behaviors,and quantifies them from qualitative and quantitative aspects respectively to form a weighted splicing behavior embedding,and then integrates the knowledge representation and time coding of content to map into deep-seated user behavior features.Transformer mechanism is introduced to encode user semantic representation and decode user preferences,and to predict user semantic preferences and target content click through rate.The experimental results show that the proposed method outperforms all baseline models.4.In order to apply the knowledge graph-based radio and television content recommendation algorithm to the radio and television system,this thesis also studies the engineering practice of the proposed algorithm model.By introducing the algorithm model into the actual radio and television recommendation system,the algorithm is run and tested,and the effectiveness of the algorithm application is verified.