Suguru Matsuyoshi

Identification of Event and Topic for Multi-document Summarization

This paper focuses on continuous news documents and presents a method for extractive multi-document summarization. Our hypothesis about salient, key sentences in news documents is that they include words related to the target event and topic of a document. Here, an event and a topic are the same as Topic Detection and Tracking (TDT) project: an event is something that occurs at a specific place and time along with all necessary preconditions and unavoidable consequences, and a topic is defined to be “a seminal event or activity along with all directly related events and activities.” The difficulty for finding topics is that they have various word distributions. In addition to the TF-IDF term weighting method to extract event words, we identified topics by using two models, i.e., Moving Average Convergence Divergence (MACD) for words with high frequencies, and Latent Dirichlet Allocation (LDA) for low frequency words. The method was tested on two datasets, NTCIR-3 Japanese news documents and DUC data, and the results showed the effectiveness of the method.

Text classification with relatively small positive documents and unlabeled data

This paper addresses the problem of dealing with a collection of negative training documents which is suitable for relatively small number of positive documents, and presents a method for eliminating the need for manually collecting negative training documents based on supervised machine learning techniques. We applied an error correction technique to the results of negative training data obtained by the Positive Example Based Learning (PEBL). Moreover, we used a boosting technique to learn a set of negative data to train classifiers. The results using Japanese newspaper documents showed that the method contributes for reducing the cost of manual collection of negative training documents.