Improved conversational recommender system based on dialog context

Abstract Conversational recommender system (CRS) needs to be seamlessly integrated between the two modules of recommendation and dialog, aiming recommend high-quality items users through multiple rounds interactive dialogs. Items can typically refer goods, movies, news, etc. Through this form express their preferences in real time, fully understand user’s thoughts corresponding items. Although mainstream dialog systems have improved performance some extent, there are still key issues, such as insufficient consideration entity’s order different contributions history, low diversity generated responses. To address these shortcomings, we propose an context model based on time-series features. Firstly, augment semantic representation words using external knowledge graphs align space mutual information maximization techniques. Secondly, add a retrieval provide auxiliary for generating replies. We then utilize deep timing network serialize content more accurately learn feature relationship recommendation. In paper, is divided into components, evaluation indicators used evaluate component component. Experimental results widely benchmarks show that proposed method effective.