273. Task 5. Keyphrase Extraction Based on Core Word Identification and Word Expansion

Proceedings of the 5th International Workshop on Semantic Evaluation, ACL 2010, pages 142–145, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics 273. Task 5. Keyphrase Extraction Based on Core Word Identification and Word Expansion You Ouyang Wenjie Li Renxian Zhang The Hong Kong Polytechnic University {csyouyang,cswjli,csrzhang}@comp.polyu.edu.hk Abstract This paper provides a description of the Hong Kong Polytechnic University (PolyU) System that participated in the task #5 of SemEval-2, i.e., the Automatic Keyphrase Extraction from Scientific Articles task. We followed a novel framework to develop our keyphrase extraction system, motivated by differentiating the roles of the words in a keyphrase. We first identified the core words which are defined as the most essential words in the article, and then expanded the identified core words to the target keyphrases by a word expansion approach. 1 Introduction The task #5 in SemEval-2 requires extracting the keyphrases for scientific articles. According to the task definition, keyphrases are the words that capture the main topic of the given document. Currently, keyphrase extraction is usually carried out by a two-stage process, including candidate phrase identification and key phrase selection. The first stage is to identify the candidate phrases that are potential keyphrases. Usually, it is implemented as a process that filters out the obviously unimportant phrases. After the candidate identification stage, the target keyphrases can then be selected from the candidates according to their importance scores, which are usually estimated by some features, such as word frequencies, phrase frequencies, POS-tags, etc.. The features can be combined either by heuristics or by learning models to obtain the final selection strategy. In most existing keyphrase extraction methods, the importance of a phrase is estimated by a composite score of the features. Different features indicate preferences to phrases with specific characteristics. As to the common features, the phrases that consist of important and correlated words are usually preferred. Moreover, it is indeed implied in these features that the words are uniform in the phrase, that is, their degrees of importance are evaluated by the same criteria. However, we think that this may not always be true. For example, in the phrase “video encoding/decoding”, the word “video” appears frequently in the article and thus can be easily identified by simple features, while the word “encoding/decoding” is very rare and thus is very hard to discover. Therefore, a uniform view on the words is not able to discover this kind of keyphrases. On the other hand, we observe that there is usually at least one word in a keyphrase which is very important to the article, such as the word “video” in the above example. In this paper, we call this kind of words core words. For each phrase, there may be one or more core words in it, which serve as the core component of the phrase. Moreover, the phrase may contain some words that support the core words, such as “encoding/decoding” in the above example. These words may be less important to the article, but they are …

Content truncated to 500 words