Extracting Discriminative Keyphrases with Learned Semantic Hierarchies

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 932–942, Osaka, Japan, December 11-17 2016. Extracting Discriminative Keyphrases with Learned Semantic Hierarchies Yunli Wang NRC Canada Scientific Data Mining Ottawa ON, Canada Yunli.Wang@nrc.ca Yong Jin U. New Brunswick CS Faculty Fredericton NB, Canada Yong.Jin@unb.ca Xiaodan Zhu NRC Canada Text Analytics Ottawa ON, Canada Xiaodan.Zhu@nrc.ca Cyril Goutte NRC Canada Multilingual Text Processing Ottawa ON, Canada Cyril.Goutte@nrc.ca Abstract The goal of keyphrase extraction is to automatically identify the most salient phrases from doc- uments. The technique has a wide range of applications such as rendering a quick glimpse of a document, or extracting key content for further use. While previous work often assumes keyphrases are a static property of a given documents, in many applications, the appropriate set of keyphrases that should be extracted depends on the set of documents that are being consid- ered together. In particular, good keyphrases should not only accurately describe the content of a document, but also reveal what discriminates it from the other documents. In this paper, we study this problem of extracting discriminative keyphrases. In particularly, we propose to use the hierarchical semantic structure between candidate keyphrases to promote keyphrases that have the right level of specificity to clearly distinguish the target document from others. We show that such knowledge can be used to construct better discriminative keyphrase extraction systems that do not assume a static, fixed set of keyphrases for a document. We show how this helps identify key expertise of authors from their papers, as well as competencies covered by online courses within different domains. 1 Introduction The purpose of keyphrase extraction is to automatically identify the most salient phrases from docu- ments. Keyphrases (of which keywords are a special case) are widely used for providing a quick glimpse of various types of documents, such as news, technical documents, etc. Automatically extracting the rel- evant keyphrases therefore has a wide range of applications and accordingly has attracted much attention from the scientific community. Previous work, however, often assumes that keyphrases are a static property of documents, that is, a given document would always produce a fixed set of keyphrases. Many approaches were developped for that purpose. For example, the Keyphrase Extraction Algorithm, or KEA (Witten et al., 1998), uses a supervised learning method (Na¨ıve Bayes) to predict keyphrases based on their lexical features. Turney (2000) developed a genetic algorithm (GenEx) to extract keyphrases, and showed that this outperformed the well-known C4.5 algorithm. More recent work on supervised keyphrase extraction used, e.g., a combination of lexical and syntactic features (Hulth, 2003) or other statistical classifiers such as support vector machine (SVM) (Zhang et al., 2006) or conditional random fields (CRF) (Zhang et al., 2008). Unsupervised methods were also proposed, based on a graph-based ranking model (Mihalcea and Tarau, 2004), or using co-occurrences (Matsuo and Ishizuka, 2004), enriched with WordNet (Martinez-Romo et al., 2016). Unsupervised keyphrase extraction was also applied to shorter texts from twitter, using multiple random walks to …

Content truncated to 500 words