Generating psychological counseling responses with language models relies heavily on high-quality datasets. Crowdsourced data collection methods require strict worker training, and data from real-world counseling environments may raise privacy and ethical concerns. While recent studies have explored using large language models (LLMs) to augment psychological counseling dialogue datasets, the resulting data often suffers from limited diversity and authenticity. To address these limitations, this study adopts a role-playing approach where trained counselors simulate counselor-client interactions, ensuring high-quality dialogues while mitigating privacy risks. Using this method, we construct KokoroChat, a Japanese psychological counseling dialogue dataset comprising 6,589 long-form dialogues, each accompanied by comprehensive client feedback. Experimental results demonstrate that fine-tuning open-source LLMs with KokoroChat improves both the quality of generated counseling responses and the automatic evaluation of counseling dialogues. The KokoroChat dataset is available at https://github.com/UEC-InabaLab/KokoroChat.

The Werewolf Game is a communication game where players’ reasoning and discussion skills are essential. In this study, we present a Werewolf AI agent developed for the AIWolfDial 2024 shared task, co-hosted with the 17th INLG. In recent years, large language models like ChatGPT have garnered attention for their exceptional response generation and reasoning capabilities. We thus develop the LLM-based agents for the Werewolf Game. This study aims to enhance the consistency of the agent’s utterances by utilizing dialogue summaries generated by LLMs and manually designed personas and utterance examples. By analyzing self-match game logs, we demonstrate that the agent’s utterances are contextually consistent and that the character, including tone, is maintained throughout the game.

My research interests focus on multimodal emotion recognition and personalization in emotion recognition tasks. In multimodal emotion recognition, existing studies demonstrate that integrating various data types like speech, text, and video enhances accuracy. However, real-time constraints and high dataset costs limit their practical application. I propose constructing a multimodal emotion recognition model by combining available unimodal datasets. In terms of personalization, traditional discrete emotion labels often fail to capture the complexity of human emotions. Although recent methods embed speaker characteristics to boost prediction accuracy, they require extensive retraining. I introduce continuous prompt tuning, which updates only the speaker prompts while keeping the speech encoder weights fixed, enabling the addition of new speaker data without additional retraining. This paper discusses these existing research gaps and presents novel approaches to address them, aiming to significantly improve emotion recognition in spoken dialogue systems.