[in Japanese]

View full abstract

[in Japanese]

View full abstract

This paper proposes a method to detect evaluation points from freely written Web review for overview of Web review structure. On freely written review, what is evaluation point and the relationships between the evaluation points are unclear. The proposed method tries to resolve this problem by overview of Web review structure. The results of the experiments showed high precision for detecting evaluation points to be presented without dependence on kinds of evaluation target. Moreover, the proposed method realized to present relationships among evaluation points to be presented referring syntax features.

View full abstract

Medical information is required for patients to understand the risks and benefits of procedures proposed by a doctor, and to select the treatment plan that they prefer. We aim to create a system that helps patients to collect medical information. The purpose of this paper is to extract descriptions of the effects caused by taking drugs as a triplet of expressions from illness survival blogs' snippets and full blog articles including narrative information. This paper proposes a method to extract such triplets using specific clue words and parsing the results in order to extract from blogs written in natural language. The experiments show that recall was improved by 24.8 points when we combined our proposed method and a baseline system, and precision was improved by 21.1 points when we utilized filters using dictionaries we created from existing medical documents.

View full abstract

This paper studies the issue of comparative analysis of cross-lingual gaps between Japanese and Chinese Web space. Especially, we focus on concerns of those who search for Web pages. We cross-lingually compare Web search information needs that are collected from search engine suggests. We discover certain differences between Japanese and Chinese Web search information needs given a specific search query.

View full abstract

Neuro-fuzzy learning algorithms with Gaussian-type membership functions based on the gradient descent method are well-known methods for generating fuzzy rules. In the conventional method, however, increasing the number of inputs greatly increases the number of parameters. Representation of fuzzy rule tables is thus difficult. We propose a new learning approach, the complex-valued neuro-fuzzy learning algorithm, which extends the conventional method domain to the complex numbers. In this method, inputs, antecedent membership functions, and consequent singletons are complex, and outputs are real. For parameter tuning, we use complex back propagation. The method assigns a two-dimensional real number to the real and imaginary parts of the complex number, which is used as a single complex-valued input. This process greatly reduces the number of tuned parameters, leading to the same or better learning than the conventional method. We compare the proposed and conventional methods using several function identification problems and show that the proposed method outperforms its counterpart, making it a useful tool for learning a fuzzy system model.

View full abstract

Generative Topographic Mapping (GTM) is a nonlinear latent variable model introduced by Bishop et al. as a data visualization technique. In this paper, we propose a supervised GTM model and a semi-supervised GTM model. The conventional supervised GTM uses classi.cation data whose teaching signals have no numerical meaning, and therefore cannot directly handle regression data whose teaching signals have numerical meaning. To overcome the problem, we propose a supervised GTM model which can naturally handle regression data. In order to handle miss-ing labels, we also propose a semi-supervised GTM model that uses both labeled and unlabeled data.

View full abstract