Uto Lab

AboutAbout the lab

Removing evaluation bias with mathematics

Building on Bayesian statistics and probability theory, the Uto Lab studies artificial intelligence, machine learning, and natural language processing, from both theoretical and applied perspectives.
The lab studies how to use mathematical models to remove rater-specific bias in situations where multiple people assign scores, such as written tests, interview exams, and reviews on review sites.
Using deep learning and large language models, the lab also studies technology for automatically scoring short essays and for automatically generating reading comprehension questions from long texts.
The lab often works with educational data, so every theme requires foundational knowledge of statistics and probability theory as well as programming skills.

ResearchResearch themes

Building on Bayesian probability theory and test theory, the lab conducts research that treats human evaluations and texts mathematically. The research is presented under three headings: rating techniques that account for rater characteristics, statistical natural language processing, and Bayesian modeling.

*Illustrative image

Techniques for Correcting Rater Bias↗

In written tests and reviews on review sites, if scores are aggregated using only averages or totals, the final score changes depending on who did the rating. The lab therefore studies mathematical models based on item response theory that remove rater bias and estimate the true score of each target.

*Illustrative image

Automated Essay Scoring↗

This is technology for automatically scoring short essays, short-answer written exams, and reports. By combining machine learning methods centered on deep learning with probabilistic and statistical techniques from test theory, psychology, and behaviormetrics, the lab develops new automated scoring technology.

*Illustrative image

Automatic Generation of Reading Comprehension Questions↗

This is technology for automatically generating reading comprehension questions, like long-passage questions in Japanese language exams, from any long text. By combining large generative language models with item response theory, the lab studies technology that generates and presents questions whose difficulty suits the learner's ability.

*Illustrative image

Bayesian Modeling↗

Using Bayesian probability theory, test theory, and language processing technology, the lab develops probabilistic models for a variety of problems. It also develops parameter estimation algorithms based on Markov chain Monte Carlo methods, which are effective for complex probabilistic models.

VoicesStudent voices

Ratings and reviews reflect the experiences of individual current or former students, not the lab's official views. Whether a lab has reviews, or how many, says nothing about the lab itself.

Students who take part in an interview may receive a thank-you gift worth ¥1,000, regardless of what they say.