Brand-Product Relation Extraction Using Heterogeneous Vector Space Representations

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1895–1901 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1895 Brand-Product Relation Extraction Using Heterogeneous Vector Space Representations Arkadiusz Janz♦, Łukasz Kopociński♦, Maciej Piasecki♦, Agnieszka Pluwak♠ ♦ Wrocław University of Science and Technology Wybrzeże Wyspiańskiego 27, 50-370 Wrocław, Poland {arkadiusz.janz|lukasz.kopocinski|maciej.piasecki}@pwr.edu.pl ♠ Institute of Slavic Studies, Polish Academy of Sciences Bartoszewicza 1B/17, 00-337 Warszawa, Poland agnieszka.pluwak@gmail.com Abstract Relation Extraction is a fundamental NLP task. In this paper we investigate the impact of underlying text representation on the performance of neural classification models in the task of Brand-Product relation extraction. We also present the methodology of preparing annotated textual corpora for this task and we provide valuable insight into the properties of Brand-Product relations existing in textual cor- pora. The problem is approached from a practical angle of applications Relation Extraction in facilitating commercial Internet monitoring. Keywords: relation extraction, vector space models, deep learning 1. Introduction Internet monitoring tools, developed and commonly used for business purposes, heavily depend in their internal im- plementation on search engines and keyword or key-phrase based search queries. Commonly such queries are built manually either by the operators of the monitoring sys- tems or in cooperation with the final users. The ability of the monitoring system to generate suggestions in a reac- tive way is a basic requirement of user experience. Such a system should propose query expansion (or more com- plex adaptation) by morpho-syntactically related words (in- flected forms), semantically interrelated words (synonyms, hypernyms, meronyms, but also derivations), etc., and also words connected in a domain specific way (named enti- ties and their attributes or pairs of named-entities related in different ways). In the case of inflection morphological dictionaries can be successfully applied. Semantic rela- tions can be extracted from lexical databases like WordNet1 or acquired by distributional semantics means. However, discovery of relations between named-entities is a complex issue, combining named-entity recognition on one hand and relation detection on the other. In this paper we focus on the problem of discovery and acquisition of named-entity pairs related by the brand (a producer) – product relation as a data source for the query expansion mechanism in an internet monitoring system. Monitoring actions and search queries for them can be related to specific business sectors or product types, e.g. cosmetics, electronic devices, specific brands (Samsung, Nivea, Apple, H&M etc.), or specific products (Lays chips, iPhone, Head&Shoulders shampoo). In all such cases, the extraction of relation between a prod- uct and a brand is crucial from the point of view of the search system or suggestions for the user building the query. In the case of a business sector query, this relation helps finding products related to brands from this sector and so enriching the search results with new data. In the case of a 1 https://wordnet.princeton.edu/ brand query, this relation also helps finding, e.g., opinions not only about the brand, but also its products. In the case of product …

Content truncated to 500 words