DIGAT: Modeling News Recommendation with Dual Graph Interaction

Anonymous

Abstract: News recommendation is essential for online news applications. Existing news recommendation approaches typically adopt a two-tower encoder framework, facing two potential limitations. First, in news encoder tower, single candidate news encoding suffers from an insufficient semantic information problem. Second, existing graph learning models for news recommendation are promising but lack effective news-user interaction modeling, which causes the graph modeling suboptimal. To overcome these limitations, we propose dual-interactive graph attention networks (DIGAT) consisting of news- and user-graph channels. In the news-graph channel, we use a semantic-augmented graph to enrich the semantics of the single candidate news by incorporating the semantic information of relevant news. In the user-graph channel, we utilize a news-topic graph to precisely model user interests. Most importantly, we design a dual-graph interaction mechanism to model effective feature interaction between the news and user graphs, which facilitates accurate news-user representation matching. Experiment results on the benchmark dataset MIND show that DIGAT outperforms the existing news recommendation methods. Further ablation studies and analyses validate the effectiveness of semantic-augmented graph encoding and dual-graph interaction.