人工知能学会全国大会論文集
35th (2021)
セッションID: 1N4-IS-1a-02
会議情報

MTabES: Entity Search with Keyword Search, Fuzzy Search, and Entity Popularities
*Phuc NGUYENIkuya YAMADAHideaki TAKEDA
著者情報
キーワード: Knowledge graph, Entity Search, Lookup
会議録・要旨集 フリー

詳細
抄録

Entity search (ES) is a problem finding relevant entities given a query as an entity label. This problem is challenging because many entities could have the same label names, and entities could have many names. Moreover, the query values are noisy, such as abbreviations or misspellings. It is a more challenging problem when the query is expressed in multilingual. (1) Objectives: We introduce an entity search tool called MTabES focused on dealing with noisy queries. In particular, we introduce a reranking function as a weighted fusion of fuzzy search with edit distance, keyword search with BM25 algorithm, and entities' popularities with PageRank scores. MTabES key advantage is the ability to boost the hit rate performance with the fuzzy search. (2) Conclusions: Entity search experimental results on SemTab 2020 and Tough Table datasets show that our toolkit could achieve a higher hit rate than knowledge graphs standard lookups i.g. Wikidata, and Wikipedia. Moreover, MTabES also work efficiently with about five queries/second in MTabES efficiency mode. MTabES toolkit is available at https://github.com/phucty/mtab_tool.

著者関連情報
© 2021 The Japanese Society for Artificial Intelligence
前の記事 次の記事
feedback
Top