Cross-loop contrast of heterogeneous graphs for interdisciplinary journal recommendation
Abstract: Interdisciplinary sciences are attracting an increasing number of researchers. Selecting the appropriate journals to disseminate valuable studies is nontrivial for researchers. However, currently, journal recommendation research for interdisciplinary fields still needs to be improved. We propose a graph contrastive learning method to provide rigorous quantitative journal recommendations for interdisciplinary fields. Our method fully explores and exploits text information at various granularities and addresses the excessive intraclass variation in papers published in interdisciplinary journals. We focus on enhancing the semantic features of different granularities by constructing feature-enhancement networks. We also construct multi-granularity heterogeneous graphs and propose cross-loop contrastive learning of global–local graphs to capture comprehensive structural features. Finally, we fuse multi-granularity features via hierarchical adaptive learning into the optimal feature representations for the final journal recommendation. To evaluate our model, we performed benchmark analysis using 28 text classification methods in two typical interdisciplinary fields: lncRNA and miRNA journal recommendations. Comprehensive experiments demonstrate the superior performance of our method, with a 1.03% to 7.94% improvement in accuracy from the Top3 to the Top10. A web server (http://www.csbg-jlu.site/lyc/) was developed to enhance usability and support journal recommendations.
Loading