TAGAN: an academic paper adversarial recommendation algorithm incorporating fine-grained semantic features

Abstract

Academic paper recommendation aims to provide users with personalized paper resources.Collaborative filtering methods face the problems of highly sparse data and lack of negative samples.Considering the above challenges, an academic paper recommendation algorithm TAGAN（title and abstract GAN）which incorporated fine-grained semantic features was presented.Firstly, based on titles and abstracts provide abundant semantic features, convolutional neural networks (CNN) was used to extract the global features of the titles, a two-layer long and short-term memory network (LSTM) was built to model abstract words separately.At the same time, the attention mechanism was proposed to associate the title and the abstract semantically.Then, the semantic features of the paper were integrated into the recommendation framework based on generative adversarial network (GAN).The generative model will fit the user’s interest preferences and can effectively replace the negative sampling process.Finally,through the experimental comparison on the public dataset, TAGAN is better than the baseline models in all indicators, which verifies the effectiveness of TAGAN.

Keywords

• paper recommendation;long and short-term memory network;generative adversarial network;attention mechanism