Research Area

(1) Data mining from large and complex networks

We study the basic technologies for understanding a huge, irregular and ever-growing complex networks, such as the Web, and then making effective use of it for knowledge navigation. Our approach includes:

• Empirical analysis of the basic properties of networks, such as scale-free properties,
• Application of statistical models and learning techniques to design the generative model of the observed data,
• Application of discrete and numerical optimization techniques that are especially suited to large-scale data.
• More specifically, we study

• Techniques on community extraction, clustering, and visualization for understanding the basic structure of large networks, visualization techniques
• Dynamic modeling techniques to predict how a network changes and grows over time,
• Information propagation modeling and popular topic extraction for understanding how information spreads over network,
• Fast similarity search using network-based indexing.

(2) Understanding the real world situation by sensor networks

We are interested in observing and interpreting the real world by a variety of sensing devices such as acceleration sensor, light sensor, GPS, camera, and microphone. Some research examples are

• Left hand figure： We assume that wrist movement contains vital information for understanding various kinds of human behavior together with the surrounding sound and color. We study techniques for recognizing human behavior by a heterogeneous sensing device attached to a wrist. So far, we succeeded in recognizing 15 different kinds of human behavior.
• Right hand figure： It shows the observed brightness of several points in the room. Data collection by wireless communication network is effective when the situation of a broad space is to be recognized by a large number of scattered sensors. We thus study several techniques for efficient network configuration and efficient data compression.
  

(3) Dialogue system that appeals to user's thinking and emotion

We conduct research on the development of such a dialogue systems that appeals to user's thinking and emotion appropriately. Unlike traditional dialogue system that is focused on the accomplishment of a certain task (such as seat reservation), our intended system aims at increasing the user's satisfaction by joyful and interesting conversations. We study how to estimate the user's thinking and emotional state through a dialogue, and also how to lead the user's state to a better one.
The figure below shows a dialogue system that aims for lively conversations. It leads the user to think by himself/herself through an interactive quiz format. The system generates an appropriate comment that encourages the user's willingness to the conversation, and shows some hints in such a way that the user cannot reach the correct answer very easily.