A Study on Automatically Extracted Keywords in Text Categorization

Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the ACL, pages 537–544, Sydney, July 2006. c⃝2006 Association for Computational Linguistics A Study on Automatically Extracted Keywords in Text Categorization Anette Hulth and Be´ata B. Megyesi Department of Linguistics and Philology Uppsala University, Sweden anette.hulth@gmail.com bea@stp.lingfil.uu.se Abstract This paper presents a study on if and how automatically extracted keywords can be used to improve text categorization. In summary we show that a higher perfor- mance — as measured by micro-averaged F-measure on a standard text categoriza- tion collection — is achieved when the full-text representation is combined with the automatically extracted keywords. The combination is obtained by giving higher weights to words in the full-texts that are also extracted as keywords. We also present results for experiments in which the keywords are the only input to the cat- egorizer, either represented as unigrams or intact. Of these two experiments, the unigrams have the best performance, al- though neither performs as well as head- lines only. 1 Introduction Automatic text categorization is the task of assign- ing any of a set of predefined categories to a doc- ument. The prevailing approach is that of super- vised machine learning, in which an algorithm is trained on documents with known categories. Be- fore any learning can take place, the documents must be represented in a form that is understand- able to the learning algorithm. A trained predic- tion model is subsequently applied to previously unseen documents, to assign the categories. In order to perform a text categorization task, there are two major decisions to make: how to repre- sent the text, and what learning algorithm to use to create the prediction model. The decision about the representation is in turn divided into two sub- questions: what features to select as input and which type of value to assign to these features. In most studies, the best performing representa- tion consists of the full length text, keeping the tokens in the document separate, that is as uni- grams. In recent years, however, a number of ex- periments have been performed in which richer representations have been evaluated. For exam- ple, Caropreso et al. (2001) compare unigrams and bigrams; Moschitti et al. (2004) add com- plex nominals to their bag-of-words representa- tion, while Kotcz et al. (2001), and Mihalcea and Hassan (2005) present experiments where auto- matically extracted sentences constitute the input to the representation. Of these three examples, only the sentence extraction seems to have had any positive impact on the performance of the auto- matic text categorization. In this paper, we present experiments in which keywords, that have been automatically extracted, are used as input to the learning, both on their own and in combination with a full-text representation. That the keywords are extracted means that the se- lected terms are present verbatim in the document. A keyword may consist of one or several tokens. In addition, a keyword may well be a whole ex- pression or phrase, …

Content truncated to 500 words