Web Mining for Unsupervised Classification

Web Mining for Unsupervised Classification 戴瑋彥 Wei-Yen Day 國立臺灣大學資訊工程學系 Department of Computer Science and Information Engineering National Taiwan University r97067@csie.ntu.edu.tw 紀均易 Chun-Yi Chi 國立臺灣大學資訊工程學系 Department of Computer Science and Information Engineering National Taiwan University ono.ccy@gmail.com 陳瑞呈 Ruey-Cheng Chen 國立臺灣大學資訊工程學系 Department of Computer Science and Information Engineering National Taiwan University cobain@turing.csie.ntu.edu.tw 鄭卜壬 Pu-Jen Cheng 國立臺灣大學資訊工程學系 Department of Computer Science and Information Engineering National Taiwan University pjcheng@csie.ntu.edu.tw 劉培森 Pei-Sen Liu 資訊工業策進會 Institute for Information Industry psliu@iii.org.tw Abstract Data acquisition is a major concern in text classification. The excessive human efforts required by conventional methods to build up quality training collection might not always be available to research workers. In this paper, we look into possibilities to automatically collect training data by sampling the Web with a set of given class names. The basic idea is to populate appropriate keywords and submit them as queries to search engines for acquiring training data. Two methods are presented in this study: One method is based on sampling the common concepts among the classes, and the other based on sampling the discriminative concepts for each class. A series of experiments were carried out independently on two different datasets, and the result shows that the proposed methods significantly improve classifier performance even without using manually labeled training data. Our strategy for 53 retrieving Web samples, we find that, is substantially helpful in conventional document classification in terms of accuracy and efficiency. Keywords: Unsupervised classification, text classification, Web mining 1. Introduction Document classification has been extensively studied in the fields of data mining and machine learning. Conventionally, document classification is a supervised learning task [1, 2] in which adequately labeled documents should be given so that various classification models, i.e., classifiers, can be learned accordingly. However, such requirement for supervised text classification has its limitations in practice. First, the cost to manually label sufficient amount of training documents can be high. Secondly, the quality of labor works is suspicious, especially when one is unfamiliar with the topics of given classes. Thirdly, in certain applications, such as email spam filtering, prototypes for documents considered as spam might change over time, and the need to access the dynamic training corpora specifically-tailored for this kind of application emerges. Automatic methods for data acquisition, therefore, can be very important in real-world classification work and require further exploration. Previous works on automatic acquisition of training sets can be divided in two types. One of which focused on augmenting a small number of labeled training documents with a large pool of unlabeled documents. The key idea from these works is to train an initial classifier to label the unlabeled documents and uses the newly-labeled data to retrain the classifier iteratively. Although classifying unlabeled data is efficient, human effort is still involved in the beginning of the training process. The other type of work focused on collecting training data from the Web. As more data is being put on the Web every day, there is a great potential to exploit the Web and devise algorithms …

Content truncated to 500 words