Research On Conversational Recommendation Technology Based On Memory Network

With the vigorous development of the Internet and the continuous emergence of new application scenarios,user-generated content has increased dramatically,making it impossible for people to obtain the information they need from massive amounts of data,which has created the problem of "information overload".Recommender system,as a key technology to solve the problem of information overload,can actively recommend items of potential interest to users without a clear purpose.It has been widely used in many fields of the Internet such as e-commerce and social media and has achieved remarkable results.However,traditional recommender systems mainly rely on user historical interaction information to generate recommendations,and lack of interaction with users,which affects the effectiveness of recommendations and user experience.Human-machine dialogue is a fundamental challenge of artificial intelligence.As an emerging research direction in the field of recommendation systems,conversational recommender system refers to the realization of high-quality automatic recommendation through dialogue-based human-computer interaction,which can directly ask users questions during the dialogue and interaction with users,and expect users’ answers obtain the user’s interest explicitly.Therefore,it is possible to flexibly generate recommendations（such as music recommendations generated by Microsoft Xiaobing,etc.）to improve the effectiveness of the recommendation and the user experience.In recent years,with the rise of intelligent voice technology,conversational recommendations have become more and more important,and have received extensive attention from academia and industry.Currently,there are three main challenges in conversational recommendation:（1）How to eliminate redundant information in the conversation;（2）How to distinguish different scenarios in the conversation;（3）How to resolve the entity ambiguity in the conversation.To solve the above problems,this paper studies the conversational recommender system based on the memory network.The main work includes:1.As for the redundant information in the dialogue,this paper proposes an encoder based on the multi-head attention mechanism.By re-encoding the randomly initialized word vector,weight adjustment and linear transformation are performed on each part of the information in the vector,so that the key information in the dialogue can obtain a higher weight.The calculation accuracy of the memory network is improved,and the recommendation effect of the system is improved.2.As for different scenarios in the dialogue,this paper designs a strategy to distinguish between interaction and recommendation scenarios,and proposes a knowledge base-based system response prediction method.Through scene differentiation,candidate reduction,personalized ranking technology,and the use of structured information in the knowledge base to model user preferences,the quality of system responses can be improved.3.To solve the ambiguity of entity recognition in dialogue,this paper proposes a knowledge base-based ambiguity elimination method.By establishing the connection between user preferences and entities,and using the internal connections between entities.It solves the ambiguity problem that a single entity may refer to multiple item attributes,and can improve the retrieval ability of the model.4.Compare the method proposed in this paper with several existing conversational recommendation algorithms.Experimental results show that the proposed method can achieve state-of-the-art performance in a few classical tasks.Besides,this paper analyzes the working principle of the model through ablation experiments and other methods,and further proves the rationality of the method.