This paper proposes a new method for displaying large-scale tag clouds. We use a topographical image that helps users to grasp the relationship among tags intuitively as a background to the tag clouds. We apply this interface to a blog navigation system and show that the proposed method enables users to find the desired tags easily even if the tag clouds are very large, 5,000 and above tags. Our approach is also effective for understanding the overall structure of a large amount of tagged documents.

There are at least three barriers to utilizing blog tags in classification or navigation: 40% of entries are not (from our observations) tagged, there are many orthographic or synonymous tag variations, and not all tags are informative. We propose a method of multi-autotagging, based on k-NN, which is a case-based classification method. Our method also has the functions of merging tags with the same meaning and identifying informative tags. For realizing these functions, we propose the term weighting method named residual document frequency(RDF); it can score the similarity between tags. Experiments show the effectiveness of our methods. Our autotagging system is generic and can assign tag(s) to any text as well as blog entries although the training data is collected from the blogosophere.

In this article, we propose a navigation system that interests users in contents, based on information on Web site visits and usage. The system guides users to tourist spots by applying a referral system. In cooperation with travel agencies, we verified the system by opening it to Internet access. Moreover, we extend the system to resolve the problems identified in the evaluation results.

In this paper, we describe a method to detect topic words from blog documents. We define "topic words" as words frequently used by people who share the same interests. In this method, each blogger’s interests are extracted from each blog site, and interest similarities between bloggers are calculated. Unusual words that are used by bloggers who have a high level of similarity are then extracted as topic words. We evaluated the precision of this method using blog documents, and the results show that the proposed method is superior (by 4.4 %) to the traditional TF-IDF method in terms of precision.