Learning Semantic Hierarchies via Word Embeddings

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 1199–1209, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Learning Semantic Hierarchies via Word Embeddings Ruiji Fu†, Jiang Guo†, Bing Qin†, Wanxiang Che†, Haifeng Wang‡, Ting Liu†∗ †Research Center for Social Computing and Information Retrieval Harbin Institute of Technology, China ‡Baidu Inc., Beijing, China {rjfu, jguo, bqin, car, tliu}@ir.hit.edu.cn wanghaifeng@baidu.com Abstract Semantic hierarchy construction aims to build structures of concepts linked by hypernym–hyponym (“is-a”) relations. A major challenge for this task is the automatic discovery of such relations. This paper proposes a novel and effec- tive method for the construction of se- mantic hierarchies based on word em- beddings, which can be used to mea- sure the semantic relationship between words. We identify whether a candidate word pair has hypernym–hyponym rela- tion by using the word-embedding-based semantic projections between words and their hypernyms. Our result, an F-score of 73.74%, outperforms the state-of-the- art methods on a manually labeled test dataset. Moreover, combining our method with a previous manually-built hierarchy extension method can further improve F- score to 80.29%. 1 Introduction Semantic hierarchies are natural ways to orga- nize knowledge. They are the main components of ontologies or semantic thesauri (Miller, 1995; Suchanek et al., 2008). In the WordNet hierar- chy, senses are organized according to the “is-a” relations. For example, “dog” and “canine” are connected by a directed edge. Here, “canine” is called a hypernym of “dog.” Conversely, “dog” is a hyponym of “canine.” As key sources of knowledge, semantic thesauri and ontologies can support many natural language processing applications. However, these semantic resources are limited in its scope and domain, and their manual construction is knowledge intensive and time consuming. Therefore, many researchers ∗Email correspondence. 生物 organism 植物 plant 毛茛科 Ranunculaceae 乌头属 Aconitum 乌头 aconite 植物药 medicinal plant 药品 medicine 生物 organism 植物 plant 毛茛科 Ranunculaceae 乌头属 Aconitum 乌头 aconite 植物药 medicinal plant 药品 medicine Figure 1: An example of semantic hierarchy con- struction. have attempted to automatically extract semantic relations or to construct taxonomies. A major challenge for this task is the auto- matic discovery of hypernym-hyponym relations. Fu et al. (2013) propose a distant supervision method to extract hypernyms for entities from multiple sources. The output of their model is a list of hypernyms for a given enity (left pan- el, Figure 1). However, there usually also exists hypernym–hyponym relations among these hy- pernyms. For instance, “植 物 (plant)” and “毛茛科 (Ranunculaceae)” are both hyper- nyms of the entity “乌头 (aconit),” and “植 物 (plant)” is also a hypernym of “毛 茛 科 (Ranunculaceae).” Given a list of hypernyms of an entity, our goal in the present work is to construct a semantic hierarchy of these hypernyms (right panel, Figure 1).1 Some previous works extend and refine manually-built semantic hierarchies by using other resources (e.g., Wikipedia) (Suchanek et al., 2008). However, the coverage is limited by the scope of the resources. Several other works relied heavily on lexical patterns, which would …

Content truncated to 500 words