A Comprehensive Dataset and Benchmark of Textual-Edge Graphs

View a PDF of the paper titled TEG-DB: A Complete Dataset and Benchmark of Textual-Edge Graphs, by Zhuofeng Li and eight different authors

View PDF
HTML (experimental)

Summary:Textual content-Attributed Graphs (TAGs) increase graph constructions with pure language descriptions, facilitating detailed depictions of knowledge and their interconnections throughout numerous real-world settings. Nonetheless, present TAG datasets predominantly function textual data solely on the nodes, with edges usually represented by mere binary or categorical attributes. This lack of wealthy textual edge annotations considerably limits the exploration of contextual relationships between entities, hindering deeper insights into graph-structured information. To deal with this hole, we introduce Textual-Edge Graphs Datasets and Benchmark (TEG-DB), a complete and various assortment of benchmark textual-edge datasets that includes wealthy textual descriptions on nodes and edges. The TEG-DB datasets are large-scale and embody a variety of domains, from quotation networks to social networks. As well as, we conduct intensive benchmark experiments on TEG-DB to evaluate the extent to which present methods, together with pre-trained language fashions, graph neural networks, and their mixtures, can make the most of textual node and edge data. Our aim is to elicit developments in textual-edge graph analysis, particularly in growing methodologies that exploit wealthy textual node and edge descriptions to boost graph evaluation and supply deeper insights into complicated real-world networks. Your complete TEG-DB challenge is publicly accessible as an open-source repository on Github, accessible at this https URL.