IntentsKB: A Knowledge Base of Entity-Oriented Search Intents

Date: October 25, 2018
Venue: Torino, Italy. The 27th ACM International Conference on Information and Knowledge Management (CIKM '18)
Corresponding article: https://arxiv.org/abs/1809.00345

Please cite the paper, and link to or credit this presentation when using it or part of it in your work.

#InformationRetrieval #IR #EntityOrientedSearch #EntityOrientedSearchIntents #KnowledgeBases

Transcript

1. INTENTSKB: A KNOWLEDGE BASE OF ENTITY-ORIENTED SEARCH INTENTS Darío Garigliotti and Krisztian Balog University of Stavanger Get the paper and the knowledge base: http://bit.ly/cikm2018-intentsKB Refiner Acquisition Refiner Categorization Intent Discovery Knowledge Base Construction [hotel] booking [hotel] deals [hotel] from airport [hotel] address … Website Service Service Property … Hotel_Booking Hotel_Booking Hotel_Booking Hotel_Arriving Hotel_Arriving Hotel_Arriving … ofCategory expressedBy expressedBy ofCategory expressedBy expressedBy … Service "rooms" "reserve" Service "from airport" "taxi" … conf1 conf2 conf3 conf4 conf5 conf6 … from airport taxi … Hotel_Arriving rooms make a reservation … Hotel_Booking [hotel] booking [hotel] deals [hotel] from airport [hotel] address … savoy hotel booking savoy hotel address … savoy hotel sheraton deals sheraton from airport … sheraton APPROACH FOR CONSTRUCTING INTENTSKB SUMMARY AND REPRESENTATION MODEL • Most entity-oriented queries consist of an entity name, complemented with context terms (refiners) to express the underlying intent of the user • • We identify the main search intents for a representative sample of entity types, and design a knowledge model to represent them in a structured fashion • • We propose a pipeline framework and build IntentsKB, a knowledge base comprising over 30k entity-oriented search intents With (S, P, O, confidence) quadruples, we model: • Intents searched for a type of entities turin map, beijing map => [city] map • Categories assigned to refiners messi instagram => Website lebron james net worth => Property congressi lingotto taxi => Service • Multiple refiners expressing an intent "booking", "book", "make a reservation", "rooms" EXPERIMENTAL EVALUATION • How well do the estimated confidence scores correspond to the actual correctness of facts? - An intent profile groups all facts for an intent - We obtain a stratified sample of profiles from 5 equally sized buckets in the confidence range (around 1.29% of the size of IntentsKB) - Experts judge correctness ignoring confidence - We find that the higher the associated confidence score, the more likely it is that the triple is correct [0, 0.87) [0.87, 0.88) [0.88, 0.9) [0.9, 0.93) [0.93, 1] Confidence intervals according to the splitting percentiles 0% 20% 40% 60% 80% 100% Proportion of triples 6,337 6,370 6,335 6,368 6,314 Correct Incorrect, OFCATEGORY Incorrect, EXPRESSEDBY Entity Type Intent ID Intent Category Refiner 1 1 1+ searchedForType ofCategory expressedBy