Hyponym expansion

To address the issue of topic drift in hyponym expansion tasks, a dataset was constructed by modeling user query interaction information for hyponym expansion tasks. The dataset consists of three main components [20]: the query terms to be expanded, concept chains, and candidate texts.Query terms to be expanded: Ambiguous leaf nodes are selected from the OpenConcepts [24] classification tree. These leaf nodes, specific to certain topics or domains, may trigger ambiguity due to polysemy or unclear hyponymy relationships.Concept chains: Starting from the leaf nodes of the tree, a backward traversal is conducted up to their parent nodes, continuing until the root node is reached, to find the ancestor nodes of the leaf nodes. Leaf nodes and their ancestor nodes are used to construct concept chains, providing contextual information for hyponym expansion, with the leaf nodes serving as the tail nodes of the concept chains.Candidate texts: Sentences containing the tail nodes are selected as candidate texts for the hyponym expansion from a large corpus of text data, including Wikipedia articles, news articles, social media posts, and academic papers.