Joint Semantic Relation Extraction for Multiple Entity Packets

Yuncheng Shi, Jiahui Wang, Zehao Huang, Shiyao Li, Chengjie Xue, Kun Yue

Abstract: Relation extraction aims to extract and identify relations among entities from unstructured texts. However, existing methods mainly focus on the relations between pairs of single entities, ignoring the joint semantics among multiple ones, leading to insufficient representations of entities and relations in sentences. To address this issue, we propose the joint semantic relation extraction model for extracting multiple entities. Specifically, we first propose the hyperplane cluster based method to find multiple entity packets efficiently. Then, we propose the method to represent multiple entity packets with joint semantic relations including the cooperation and independence features. To promote entity cooperation, we introduce the graph attention network to obtain the potential joint semantics among multiple entities. To promote entity independence, we explore independent semantics by using the fluctuations and regular semantics of entities. Finally, we aggregate the joint willingness among the entities in packets by combining the above two types of features, and thus extract the joint semantic relations effectively. Experimental results on various datasets illustrate that our method outperforms the state-of-the-art competitors by 0.9% to 16.1% and verify the effectiveness of our method.

External IDs:dblp:conf/apweb/ShiWHLXY24