This page contains general information about my research, previous and present. It also contains description of the systems, tools and methods I developed or took part in developing.

About My Research

I have started my research on expreessions of emotions in 2004...

back to top

ML-Ask: Affect Analysis System

ML-Ask is a system for Affect Analysis of textual input in Japanese. It is based on a linguistic assumption that emotional states of a speaker are conveyed by emotional expressions used in emotive utterances. ML-Ask firstly separates emotive utterances from non-emotive and in the emotive utterances seeks for expressions of specific emotion types. Read more...

back to top

CAO: Emoticon Analysis System

CAO is a system for analysis of emoticons in Japanese online communication. Emoticons are strings of symbols widely used in text-based online communication to convey user emotions. CAO extracts emoticons from input and determines the specific emotion types they express. Firstly, it matches the extracted emoticons to a predetermined raw emoticon database containing over ten thousand emoticon samples extracted from the Web. The emoticons, for which emotion types could not be determined using only this database, are automatically divided into semantic areas representing “mouths” or “eyes”. These areas are automatically annotated according to their co-occurrence in the database. Read more...

back to top

Automatic Evaluation of Conversational Agents

The above two systems are utilized in my further research on methods for enhancing Human-Computer Interaction. The first is a method for automatic evaluation of conversational agents. The affect analysis systems are used to analyze users’ emotional engagement during conversation. This data is reinterpreted to specify general attitudes to the conversational agent and its performance. In the evaluation, the method was used as a background procedure during conversations with two Japanese-speaking conversational agents. The users’ attitudes to the agents were determined automatically during the conversations and compared to the results of a questionnaire taken after the conversations. The results provided by the method revealed similar tendencies to the questionnaire, proving the method as applicable in automatic evaluation of Japanese-speaking conversational agents. Read more...

back to top

Contextual Appropriateness of Emotions

This is a somewhat novel approach to the analysis of emotional states of users. The method not only determines the emotions expressed by speaker, but also verifies whether these emotions are appropriate for the context of the conversation. In this method, affect analysis system estimates the speaker’s affective states and a Web mining technique gathers from the Internet emotive associations consisting of a list of emotions that should be expressed at the moment. The method is still in its developmental stage, but I put a lot of hope in it. I assume implementing this method to a conversational agent could allow it differentiate between sincere/appropriate/natural and insincere/ inappropriate/unnatural expression of emotions (e.g. irony, harrasment, etc.). This in result could help choose appropriate conversational procedures, and therefore significantly enhance human-computer interaction. Read more...

back to top

Automatic Detection of Cyberbullying Activities

Online security has been an important issue for several years. One of the burning online security problems lately in Japan has been online slandering and bullying, which appear on unofficial Web sites. The problem has been becoming especially urgent on unofficial Web sites of Japanese schools. School personnel and members of Parent-Teacher Association (PTA) have started Online Patrol to spot Web sites and blogs containing such inappropriate contents. However, countless number of such data makes the job an uphill task. This paper presents a research aiming to develop a systematic approach to Online Patrol by automatically spotting suspicious entries and reporting them to PTA members and therefore help them do their job. I presented some of the first results of analysis of the inappropriate data collected from unofficial school Web sites. The analysis was performed firstly with an SVM based machine learning method to detect the inappropriate entries. After analysis of the results I performed another analysis of the data, using ML-Ask to find out how the machine learning model could be improved. Read more...

back to top

Sentence Pattern Extraction Architecture

SPEC is a language independent architecture for extracting from sentences more sophisticated patterns than n-grams. In this architecture a "sentence pattern" is considered as n-element ordered combination of sentence elements. Read more...

back to top

Part-of-Speech Tagger for Ainu Language

Short description of POST-AL. Read more...

back to top

Yet Another Corpus of Internet Sentences

Short description of YACIS. Read more...

back to top