In this paper, we suggest method to categorize item by using item information on the Web. We prepared 4 kinds of item information to categorize item: item names, item names that extracted nouns, item names and item descriptions, and item names and item descriptions that extracted nouns. With these item information, we categorized item by naive Bayes classifier. As a result of categorization, we got classification precision to exceed far a baseline. Classification precision that used only item names was about 15 point higher than used item names and item descriptions. Classification precision that used item information extracted nouns was about 1 point higher than used not extracted nouns.

We invesitgate information diffusion in Blogosphere using URLs described in blog posts. We first divide the set of those URLs into two groups, the group of the URLs that belong to the major media sources and the group of the other URLs. Next, by regarding the blogroll and the comment networks as social networks among bloggers, we extract the word-of-mouth information diffusion patterns in Blogosphere. We compare the word-of-mouth information diffusion patterns through the blogroll and the comment networks with the information diffusion patterns from the major media sources in terms of duration and diffusion speed. Moreover, we analyze the word-of-mouth information diffusion patterns through these social networks in Blogosphere in terms of diffusion probablities and time-delays through links, and influence degrees of nodes.

Information recommendation system attempts to present information that is likely to be useful for the user. Showing recommendation reason is an important role of the system. However, current recommendation system shows very simple or quantitative explanation. In this paper, we aim at showing the recommendation reason that is precise, non-quantitative, and easy to understand. We make use of predicate logic form rules for recommendation and showing reason. In order to perform such recommendation, we use Inductive Logic Programming. We succeed to extract several useful rules from\nBlog.

In recent years, there are increasing attentions to recommendation systems. A recommendation system semi-automatically understands users' tastes and recommends suitable contents for users based on users' profiles. In this paper, we propose a brand new recommendation algorithm based on social bookmarked data. The algorithm is constructed and valuated via tests. The original idea in this paper is to evaluate the contents according their information measure, which is a variant of J-measure. The base set of the system is the set of pairs of keywords, called tags, and users. We attempted to solution of subject of folksonomy studies that polysemy and basic level variation.

We propose a method for extracting bursty latent topics from a document stream that is a time-series data of documents. We utilize Latent Dirichlet Allocation (LDA), which is a probabilistic generative model of documents, for extracting latent topics, and introduce a time-filter for identifying bursty topics. We construct a measure of similarity between two documents with time-stamps on the basis of LDA and the time-filter, and extract bursty latent topics from a document stream by applying a hierarchical agglomerative clustering method. Using real data of document streams, we experimentally demonstrate the effectiveness of the proposed method.

In this paper, we present a system capable of translating natural language questions into SPARQL queries and searching candidate answers from the Linked Open Data (LOD) then returning filtered results to the user. The processes for building queries and searching in the LOD databases are discussed.

In the present paper, an idea deriving user interface for ret rieving question-answer sets from archives of knowledge community sites is proposed. To the user interface, classification implement methods for question type and answer type are also proposed. The methods are evaluated experimentally and provide fairly good precision.