Incorporating Trustiness and Collective Synonym/Contrastive Evidence into Taxonomy Construction

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 1013–1022, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Incorporating Trustiness and Collective Synonym/Contrastive Evidence into Taxonomy Construction Luu Anh Tuan #1, Jung-jae Kim #2, Ng See Kiong ∗3 #School of Computer Engineering, Nanyang Technological University, Singapore 1anhtuan001@e.ntu.edu.sg, 2jungjae.kim@ntu.edu.sg ∗Institute for Infocomm Research, Agency for Science, Technology and Research, Singapore 3skng@i2r.a-star.edu.sg Abstract Taxonomy plays an important role in many applications by organizing domain knowl- edge into a hierarchy of is-a relations be- tween terms. Previous works on the taxo- nomic relation identification from text cor- pora lack in two aspects: 1) They do not consider the trustiness of individual source texts, which is important to filter out in- correct relations from unreliable sources. 2) They also do not consider collective evidence from synonyms and contrastive terms, where synonyms may provide ad- ditional supports to taxonomic relations, while contrastive terms may contradict them. In this paper, we present a method of taxonomic relation identification that incorporates the trustiness of source texts measured with such techniques as PageR- ank and knowledge-based trust, and the collective evidence of synonyms and con- trastive terms identified by linguistic pat- tern matching and machine learning. The experimental results show that the pro- posed features can consistently improve performance up to 4%-10% of F-measure. 1 Introduction Taxonomies which serve as backbone of struc- tured knowledge are useful for many applica- tions such as question answering (Harabagiu et al., 2003) and document clustering (Fodeh et al., 2011). Even though there are many hand-crafted, well-structured taxonomies publicly available, in- cluding WordNet (Miller, 1995), OpenCyc (Ma- tuszek et al., 2006), and Freebase (Bollacker et al., 2008), they are incomplete in specific domains, and it is time-consuming to manually extend them or create new ones. There is thus a need for auto- matically extracting taxonomic relations from text corpora to construct/extend taxonomies. Previous works on the task of taxonomy con- struction capture information about potential tax- onomic relations between concepts, rank the can- didate relations based on the captured information, and integrate the highly ranked relations into a tax- onomic structure. They utilize such information as hypernym patterns (e.g. A is a B, A such as B) (Kozareva et al., 2008), syntactic dependency (Drumond and Girardi, 2010), definition sentences (Navigli et al., 2011), co-occurrence (Zhu et al., 2013), syntactic contextual similarity (Tuan et al., 2014), and sibling relations (Bansal et al., 2014). They, however, lack in the three following as- pects: 1) Trustiness: Not all sources are trust- worthy (e.g. gossip, forum posts written by non-experts) (Dong et al., 2015). The trustiness of source texts is important in taxonomic rela- tion identification because evidence from unre- liable sources can be incorrect. For example, the invalid taxonomic relation between “American chameleon” and “chameleon” is mistakenly more popular in the Web than the valid taxonomic rela- tion between “American chameleon” and “lizard”, and statistical methods without considering the trustiness may incorrectly extract the invalid rela- tion instead of …

Content truncated to 500 words