Research On Aspect Recommendation For Concept And Its Application To Thematic Book Automatic Generation

In the big data era,users are facing the problems of how to organize and find knowledge in such a large-scale content efficiently.It is an efficient way to organize content according to the aspects of concepts.For example,the knowledge about each concept is organized in multiple aspects in Baidu encyclopedia and Wikipedia.However,there are still some weaknesses for the existing encyclopedias,such as incomplete content and insufficient aspects.Thus,it is necessary to recommend aspects for concepts,and to generate a more comprehensive content.The key problem is how to recommend aspects to organize content from digital libraries and the Internet.Therefore,we propose two novel aspect recommendation algorithms for concepts for content organization.The main contributions of this paper is as follows:1.We propose an aspect recommendation based on bipartite product graph.In the algorithm,the recommendation problem is transformed into the prediction problem in bipartite graph.Through the transductive learning on the product graph,the aspects are recommended and ranked.In addition,the computational bottleneck is also optimized in the graph model.2.We propose a neural matrix factorization collaborative filtering recommendation model based on concept tags to improve the performance of aspect recommendation.The model introduces the tags to improve the representation of concepts,and obtains the aspects by fusing generalized matrix factorization and multi-layer perceptron.This algorithm can solve the weakness of the graph model,which is complex and cannot be saved.The tags can improve the semantic representation of concepts and aspects.3.We implement the aspect recommendation algorithms mentioned above,and build the thematic book generation system for CKCEST(China Knowledge Center for Engineering Sciences and Technology).