Contents
English / Japanese
Research Themes
Based on a knowledge engineering approach,
we are building associative knowledge and commonsense knowledge,
and aim to extend them
through natural language processing.
Concept Association
We believe that the ability to associate ideas is the foundation that allows people to communicate smoothly through language. If computers could imitate this human ability, it might become possible to build a new kind of computer that makes more human-like judgments. Our laboratory has long studied the Concept-base as a mechanism for giving computers this associative ability. Today, words are commonly defined by distributed representations such as Word2Vec. The Concept-base instead defines each word by a set of associated words, and uses these sets to give computers the ability to associate.
Emotion Analysis
For computers and people to interact more closely, computers need to understand the emotions that people feel. There are many models of emotional expression, and our current research performs emotion analysis based on Plutchik’s wheel of emotions. The knowledge underlying the analysis is built by hand, and by automatically extending and updating this knowledge base, we achieve emotion analysis that handles a wide variety of expressions.
Kaomoji Analysis
In text-based communication, people often use non-verbal expressions such as kaomoji (Japanese-style emoticons) and emoji. In recent years, stickers on services such as LINE and Facebook have been taking their place, but stickers depend on each service and cannot always be used right away. Our laboratory analyzes in depth the countless kaomoji found on the Web and elsewhere (more than 100,000 kinds have already been identified), and is working to model the process by which new kaomoji are created.
Other Research Themes
- Culprit estimation in mystery novels
Although e-books have become widespread, information such as reviews of novels is still only available if readers write it themselves. This theme focuses on mystery novels. We evaluate the difficulty of a novel by how easy it is to identify the culprit, and aim to build a system that recommends novels to readers.