A Compositional-Distributional Semantic Model for Searching Complex Entity Categories

Proceedings of the Fifth Joint Conference on Lexical and Computational Semantics (*SEM 2016), pages 199–208, Berlin, Germany, August 11-12, 2016. A Compositional-Distributional Semantic Model for Searching Complex Entity Categories Juliano Efson Sales1, Andr´e Freitas1, Brian Davis2, Siegfried Handschuh1 1Department of Computer Science and Mathematics - University of Passau Innstrasse 43, ITZ-110, 94032 Passau, Germany {juliano-sales,andre.freitas,siegfried.handschuh}@uni-passau.de 2Insight Centre for Data Analytics - National University of Ireland Galway IDA Business Park, Lower Dangan, Galway, Ireland brian.davis@insight-centre.org Abstract Users combine attributes and types to describe and classify entities into cate- gories. These categories are fundamen- tal for organising knowledge in a decen- tralised way acting as tags and predi- cates. When searching for entities, ca- tegories frequently describes the search query. Considering that users do not know in which terms the categories are ex- pressed, they might query the same con- cept by a paraphrase. While some ca- tegories are composed of simple expres- sions (e.g. Presidents of Ireland), oth- ers have more complex compositional pat- terns (e.g. French Senators Of The Second Empire). This work proposes a hybrid se- mantic model based on syntactic analysis, distributional semantics and named entity recognition to recognise paraphrases of entity categories. Our results show that the proposed model outperformed the com- parative baseline, in terms of recall and mean reciprocal rank, thus being suitable for addressing the vocabulary gap between user queries and entity categories. 1 Introduction A significant part of search queries on the web tar- get entities (e.g. people, places or events) (Pound et al., 2010). In this context, users frequently use the characteristics of the target entity to describe the search query. For example, to find Barack Obama, it is reasonable that a user types the query Current President of United States. The combination of attributes and types of an entity in a grammatically correct fashion defines an entity category, which groups a set of entities that share common characteristics. Examples of entity categories are French Female Artistic Gym- nasts, Presidents of Ireland and French Senators Of The Second Empire. Considering that users do not know in which terms the categories are ex- pressed, they might query the same concept by a paraphrase, i.e. using synonyms and different syn- tactic structures. The following text excerpt from Wikipedia shows an example where Embraer S.A is defined as Brazilian aerospace conglomerate: “Embraer S.A. is a Brazilian aerospace conglomerate that produces commer- cial, military, executive and agricultural aircraft and provides aeronautical ser- vices. It is headquartered in S˜ao Jos´e dos Campos, S˜ao Paulo State.”1 The flexibility and richness of natural language allow describing Brazilian aerospace conglom- erate both as Brazilian Planemaker2 or as Aircraft manufacturers of Brazil3. In addition to their occurrence in texts, en- tity categories are also available in the form of structured data. The Yago project (Suchanek et al., 2007) shares unary properties associating hun- dreds of thousands of descriptive categories ma- nually created by the Wikipedia community to DBpedia entities (Auer et al., 2007). Thus, a mechanism to recognise paraphrases can make …

Content truncated to 500 words