Learning Timeline Difference for Text Categorization

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 799–804, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Learning Timeline Difference for Text Categorization Fumiyo Fukumoto Graduate Faculty of Interdisciplinary Research Univ. of Yamanashi, Japan fukumoto@yamanashi.ac.jp Yoshimi Suzuki Graduate Faculty of Interdisciplinary Research Univ. of Yamanashi, Japan ysuzuki@yamanashi.ac.jp Abstract This paper addresses text categorization problem that training data may derive from a different time period from the test data. We present a learning framework which extends a boosting technique to learn accurate model for timeline adapta- tion. The results showed that the method was comparable to the current state-of-the- art biased-SVM method, especially the method is effective when the creation time period of the test data differs greatly from the training data. 1 Introduction Text categorization supports and improves several tasks such as creating digital libraries, informa- tion retrieval, and even helping users to interact with search engines (Mourao et al., 2008). growing number of machine learning (ML) tech-A niques have been applied to the text categorization task (Xue et al., 2008; Gopal and Yang, 2010). Each document is represented using a vector of features/terms (Yang and Pedersen, 1997; Hassan et al., 2007). Then, the documents with category label are used to train classifiers. Once category models are trained, each test document is classi- fied by using these models. A basic assumption in the categorization task is that the distributions of terms between training and test documents are identical. When the assumption does not hold, the classification accuracy is worse. However, it is often the case that the term distribution in the training data is different from that of the test data when the training data may drive from a different time period from the test data. Manual annotation of tagged new data is very expensive and time- consuming. The methodology for accurate clas- sification of the new test data by making the max- imum use of tagged old data is needed in learning techniques. In this paper, we present a method for text cat- egorization that minimizes the impact of tempo- ral effects. Our approach extends a boosting tech- nique to learn accurate model for timeline adap- tation. We used two types of labeled training data: One is the same creation time period with the test data. Another is different creation time period with the test data. We call the former same- period training, and the latter diff-period train- ing data. For the same-period training data, the learner shows the same behavior as the boosting. In contrast, for diff-period training data, once they are wrongly predicted by the learned model, these data would be useless to classify test data. We decreased the weights of these data by applying Gaussian function in order to weaken their im- pacts. 2 Related Work The analysis of temporal aspects is a practical problem as well as the process of large-scale heterogeneous data since the World-Wide Web (WWW) is widely used by various sorts of peo- ple. It …

Content truncated to 500 words