Neural Entity Synonym Set Generation using Association Information and Entity Constraint

Automatically generating entity synonym sets (i.e., sets of terms that represent the same entity) is an important work for many entity-based tasks. Existing studies on entity synonym set generation either use a ranking plus pruning approach or take the problem as a two-phase task (i.e., extracting synonymy pairs, subsequently organizing these pairs into synonym sets). However, these approaches ignore the association semantics of entities and suffer from the error propagation issue. In this paper, we propose a neural-network-based entity synonym set generation approach that exploits association information and entity constraint to generate synonym sets from a given term (i.e., entity) vocabulary. Firstly, to learn whether a new term should be added into the synonym set, an association-aware set-term neural network classifier is proposed. In the classifier, not only the entity representations but also the entity association information is exploited for extracting synonymous features. Secondly, an entity-constraint-based synonym set generation algorithm is employed to apply the trained set-term neural network classifier to generate the entity synonym sets from the term vocabulary. Finally, we conduct the proposed approach on three real-world datasets. The experimental results demonstrate that the entity synonym set generation performance of the proposed approach is better than that of the compared approaches.