BiPaSs: Further Investigation of Fast Pathfinding in Wikidata

Purpose: A previous paper proposed a bidirectional A* search algorithm for quickly finding meaningful paths in Wikidata that leverages semantic distances between entities as part of the heuristics. However, work lacks an optimization algorithm’s hyperparameters and evaluation on large dataset among others. The purpose present is to address these open points. Methodology: Approaches aimed at enhancing accuracy are discussed. Furthermore, different options constructing dual-entity queries pathfinding explored. 20% compiled utilized fine-tune using Simple optimizer. optimized configuration subsequently evaluated against alternative configurations, including baseline, remaining 80% dataset. Findings: additional consideration entity descriptions increases distances. query with 1,196 pairs derived from TREC 2007 Million Query Track yields values 0.699/0.109/0.823 hyperparameters. This achieves higher coverage test set (79.2%) few visits (24.7 average) moderate path lengths (4.4 average). For reproducibility, implementation called BiPaSs, dataset, benchmark results provided. Value: Web engines reliably generate knowledge panels summarizing information only response mentioning single entity. shows unseen feasible. Based paths, can be generated provide explanation mentioned entities’ relationship, potentially satisfying users’ need.