NTT Research and Development 2008 Review of Activities

Although keyword-based searches are a popular way of finding information from a huge amount of information indexed in a search engine, users sometimes find it hard to specify the appropriate keywords. Tag clouds, in which a set of keywords or categories to be selected are displayed by the system, have become popular since they enable users to input keywords by just clicking on them. It is, however, hard to identify the desired tag if the number of tags is very large, e.g., several hundred, because of the alphabetical-order layout in which relations between tags are not presented.

To resolve this issue, NTT Laboratories developed a new interface for displaying tag clouds. It enables users to grasp relations between tags intuitively by using a topographic image as a background. BLOGRANGER TG, hereafter TG, is a new blog navigation system that provides this interface. TG calculates similarities between tags by analyzing term occurrences in 9 millions tagged blog entries collected from 15 millions entries within the last five weeks. TG generates a topographic image in which similar tags do not overlap with each other. It also enables users to find tags by locating tags with a central concept as the peak of the mountain in the topic area. Moreover, TG automatically assigns tags to untagged blog entries. TG thus allows users to find a wide range of related blog entries just by clicking tags in the topography. "goo labs" is evaluating the feasibility and usability of TG, and its trial will run from December 2007 to November 2008.

We will improve TG by incorporating the results of the trial and commercialize it for portal businesses and content navigation interfaces for enterprise uses.

• BLOGRANGER is a registered trademark of Nippon Telegraph and Telephone Corporation.