Research On Conversation Recommendation Method Based On Neural Network

With the advent of the 5G era,Internet technology has covered all aspects of clothing,food,housing,transportation,etc.,bringing convenience to people,but also making the problem of information overload more serious.As one of the core technologies for overcoming information overload,recommendation system has attracted more and more scientific research scholars’ attention.The user attribute information and user-item interaction matrix data required by the traditional recommendation algorithm are difficult to obtain,while the behavior data left by the user during the conversation can be easily recorded.Therefore,this paper conducts an in-depth study on a conversational recommendation system for modeling conversational behavior data.This dissertation analyzes the current deficiencies in the research of conversational recommendation systems,and proposes two conversational recommendation models,one is the conversational recommendation model SR-IAN based on item characteristics and attention mechanism;the other is based on domain awareness The collaborative session recommendation model CSR-NA.In the SR-IAN model,the GRU network is used to model the current user’s conversational behavior sequence to mine short-term interests;the attention mechanism that combines conversational item characteristics and item location information is used to mine the long-term intentions contained in the conversational behavior sequence;finally,the current is linearly spliced The user’s short-term interests and long-term intentions are characterized by the intent fusion conversation feature,and the matching recommendation list is matched by the feature representation.The CSR-NA model adds inter-session collaboration information on the basis of the SR-IAN model,and uses multi-layer perceptrons to adaptively allocate the relative weights of the intent-fused session feature representation and the inter-session collaboration information to obtain the final representation of the current session;According to the final representation,the recommendation probability score of each candidate item in the candidate set is calculated,and the top K items with the highest recommendation probability score are used as the recommended list of the model.Finally,in order to verify the recommendation effects of the SR-IAN and CSR-NA models,experimental tests were performed on the two data sets Yoochoose and Dignetica to verify the effectiveness of the feature representation of intention fusion and the synergy information between sessions.In addition,SR-IAN and CSR-NA are compared with 9benchmark models.The experimental results show that the model is better than other benchmark models on the two data indicators of Recall and MRR.