Search

We propose a method for estimating class membership probabilities of a predicted class, using classification scores not only for the predicted class but also for other classes in a document classification. Class membership probabilities are important in many applications in document classification, in which multiclass classification is often applied. In the proposed method, we first make an accuracy table by counting the number of correctly classified training samples in each range or cell of classification scores. We then apply smoothing methods such as a moving average method with coverage to the accuracy table. In order to determine the class membership probability of an unknown sample, we first calculate the classification scores of the sample, then find the range or cell that corresponds to the scores and output the values associated in the range or cell in the accuracy table. Through experiments on two different datasets with both Support Vector Machines and Naive Bayes classifiers, we empirically show that the use of multiple classification scores is effective in the estimation of class membership probabilities, and that the proposed smoothing methods for the accuracy table work quite well. We also show that the estimated class membership probabilities by the proposed method are useful in the detection of the misclassified samples. [ABSTRACT FROM AUTHOR]

We describe statistical models for detecting causality between two events. Our models are kinds of latent variable models, actually expanded versions of the existing statistical co-occurrence models. The (statistical) dependency information between two events needs to be incorporated into causal models. We handle this information via latent variables in our models. Through experiments, we achieved .678 F-measure value for the evaluation data. [ABSTRACT FROM AUTHOR]

We propose to use semi-supervised learning methods to classify evaluative expressions, that is, tuples of subjects, their attributes, and evaluative words, that indicate either favorable or unfavorable opinions towards a specific subject. Due to its characteristics, the semi-supervised method that we use can classify evaluative expressions in a corpus by their polarities. This can be accomplished starting from a very small set of seed training examples and using contextual information in the sentences to which the expressions belong. Our experimental results with actual Weblog data show that this bootstrapping approach can improve the accuracy of methods for classifying favorable and unfavorable opinions. [ABSTRACT FROM AUTHOR]

Document sentiment classification is a task to classify a document according to the positive or negative polarity of its opinion (favorable or unfavorable). We propose using syntactic relations between words in sentences for document sentiment classification. Specifically, we use text mining techniques to extract frequent word sub-sequences and dependency sub-trees from sentences in a document dataset and use them as features of support vector machines. In experiments on movie review datasets, our classifiers obtained the best results yet published using these data. [ABSTRACT FROM AUTHOR]

We apply a machine learning method to the occupation coding, which is a task to categorize the answers to open-ended questions regarding the respondent's occupation. Specifically, we use Support Vector Machines (SVMs) and their combination with hand-crafted rules. Conducting the occupation coding manually is expensive and sometimes leads to inconsistent coding results when the coders are not experts of the occupation coding. For this reason, a rule-based automatic method has been developed and used. However, its categorization performance is not satisfiable. Therefore, we adopt SVMs, which show high performance in various fields, and compare it with the rule-based method. We also investigate effective combination methods of SVMs and the rule-based method. In our methods, the output of the rule-based method is used as features for SVMs. We empirically show that SVMs outperform the rule-based method in the occupation coding and that the combination of the two methods yields even better accuracy. [ABSTRACT FROM AUTHOR]

Conventional QA systems cannot answer to the questions composed of two or more sentences. Therefore, we aim to construct a QA system that can answer such multiple-sentence questions. As the first stage, we propose a method for classifying multiple-sentence questions into question types. Specifically, we first extract the core sentence from a given question text. We use the core sentence and its question focus in question classification. The result of experiments shows that the proposed method improves F-measure by 8.8% and accuracy by 4.4%. [ABSTRACT FROM AUTHOR]

We propose to use both labeled and unlabeled data with the Expectation-Maximization (EM) algorithm in order to estimate the generative model and use this model to construct a Fisher kernel. The Naive Bayes generative probability is used to model a document. Through the experiments of text categorization, we empirically show that, (a) the Fisher kernel with labeled and unlabeled data outperforms Naive Bayes classifiers with EM and other methods for a sufficient amount of labeled data, (b) the value of additional unlabeled data diminishes when the labeled data size is large enough for estimating a reliable model, (c) the use of categories as latent variables is effective, and (d) larger unlabeled training datasets yield better results. [ABSTRACT FROM AUTHOR]