ClozeSearch: A Collocation Retrieval Application to Assist in Scientific Writing
Mengru Wang, Omar Alonso · Proceedings of the 31st ACM International Conference on Information & Knowledge Management · 2022
This paper presents a slot-filling retrieval application, ClozeSearch, for searching collocates to assist users in scientific writing. ClozeSearch suggests plausible collocates to fill in user-created slots within the query text. To ease the query formulation, we adapt the autocomplete feature to a slot-filling fashion of querying. Given a query prefix with slots, we select multiple valid terms to replace each slot and then provide complete suggestions based on such hypothetical prefixes. To reduce the search space for sampling the terms, we leveraged histogram pruning. Moreover, we propose two alternatives based on syntactic graph and deep language model for better flexibility in coping with long queries. Experimental results show that our proposed methods outperform the conventional pattern-based matching by a maximum of 0.18 points in F1-score.