Yoshinobu Kano
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yoshinobu Kano has a total of 92 co-authors. Here are the top co-authors by number of shared publications:
- Ken Satoh
Center for Juris-Informatics, ROIS-DS, Tokyo, Japan
6 shared publications - Masaharu Yoshioka
Hokkaido University, Sapporo, Hokkaido, Japan
6 shared publications - Mi-Young Kim
University of Alberta
6 shared publications - Randy Goebel
University of Alberta
6 shared publications - Juliano Rabelo
University of Alberta
5 shared publications - Takaaki Onaga
Shizuoka University
4 shared publications
Latest Publications
- Open Access
- 1 November 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e98094
Machine Learning Classification of Bipolar Disorder Across Mood States Using Natural Language Processing and Speech Analysis: Observational Study
Background: Bipolar disorder (BP) affects 1% to 3% of adults worldwide, with diagnostic delays averaging 6 years and high initial misdiagnosis rates. The absence of objective diagnostic measures contributes to clinical uncertainty and suboptimal outcomes. Objective: This study aimed to (1) develop a machine learning model for BP diagnosis and mood state classification using linguistic and acoustic features, (2) identify features associated with different mood states by comparing each with healthy controls (HC), and (3) explore whether features distinguish between mood states within BP. Methods: We analyzed data prospectively collected over 7 years from 61 patients with BP and 99 HC across Japan. Speech samples were collected via semistructured interviews over 1 to 5 visits and analyzed cross-sectionally, using aggregated or single-visit observations per participant rather than within-person longitudinal modeling. We extracted 6373 acoustic features (openSMILE 3.0, ComParE 2016) and 211 linguistic features. Least absolute shrinkage and selection operator (LASSO)–regularized logistic regression with nested cross-validation and age- and sex-matched undersampling was used. Mood states were classified as depressive state (DS; Hamilton Depression Rating Scale-17 [HAMD-17] ≥8, Young Mania Rating Scale [YMRS] ≤7), state with manic/mixed symptoms (MS/MX; YMRS ≥8), or euthymic state (ES; HAMD-17 ≤7, YMRS ≤7). Seven classification tasks included BP vs HC, each mood state vs HC, and interstate comparisons. Feature importance was validated via permutation testing ( P <.05). Medication sensitivity analyses were performed. Results: The model achieved high discrimination for BP vs HC (area under the receiver operating characteristic curve [AUC]=0.916, 95% CI 0.854‐0.958; sensitivity 0.933; specificity 0.859) and strong performance for mood states vs HC (AUC=0.805‐0.874). Acoustic features comprised 96% to 98% of the features achieving statistical significance ( P <.05), including mel-frequency cepstral coefficients, spectral characteristics, and voice quality measures, showing strong correlations with diagnostic probabilities (|ρ|=0.35‐0.85) and symptom severity (HAMD-17: ρ=–0.51; YMRS: ρ=–0.37); 2 to 4 linguistic features per comparison also reached significance. With sample independence ensured for interstate comparisons, DS vs ES significantly exceeded chance (AUC=0.760, permutation P =.03), whereas MS/MX vs ES (AUC=0.601, P =.20) and DS vs MS/MX (AUC=0.569, P =.30) did not, suggesting that markers primarily reflect persistent disease characteristics with partial state-associated components in depressive states. Conclusions: The combined acoustic-linguistic model, driven primarily by acoustic features, shows potential as an objective aid for BP diagnosis (AUC>0.9), largely free of medication confounding; some features remained detectable during euthymia, although ES discrimination could not be fully separated from concurrent antipsychotic treatment. Although DS showed partial distinguishability from ES, overall limitations remain for monitoring mood state changes. Findings require validation in diverse populations and against other psychiatric conditions.
- 1 November 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e98094
Open AccessMachine Learning Classification of Bipolar Disorder Across Mood States Using Natural Language Processing and Speech Analysis: Observational Study
Background: Bipolar disorder (BP) affects 1% to 3% of adults worldwide, with diagnostic delays averaging 6 years and high initial misdiagnosis rates. The absence of objective diagnostic measures contributes to clinical uncertainty and suboptimal outcomes. Objective: This study aimed to (1) develop a machine learning model for BP diagnosis and mood state classification using linguistic and acoustic features, (2) identify features associated with different mood states by comparing each with healthy controls (HC), and (3) explore whether features distinguish between mood states within BP. Methods: We analyzed data prospectively collected over 7 years from 61 patients with BP and 99 HC across Japan. Speech samples were collected via semistructured interviews over 1 to 5 visits and analyzed cross-sectionally, using aggregated or single-visit observations per participant rather than within-person longitudinal modeling. We extracted 6373 acoustic features (openSMILE 3.0, ComParE 2016) and 211 linguistic features. Least absolute shrinkage and selection operator (LASSO)–regularized logistic regression with nested cross-validation and age- and sex-matched undersampling was used. Mood states were classified as depressive state (DS; Hamilton Depression Rating Scale-17 [HAMD-17] ≥8, Young Mania Rating Scale [YMRS] ≤7), state with manic/mixed symptoms (MS/MX; YMRS ≥8), or euthymic state (ES; HAMD-17 ≤7, YMRS ≤7). Seven classification tasks included BP vs HC, each mood state vs HC, and interstate comparisons. Feature importance was validated via permutation testing ( P <.05). Medication sensitivity analyses were performed. Results: The model achieved high discrimination for BP vs HC (area under the receiver operating characteristic curve [AUC]=0.916, 95% CI 0.854‐0.958; sensitivity 0.933; specificity 0.859) and strong performance for mood states vs HC (AUC=0.805‐0.874). Acoustic features comprised 96% to 98% of the features achieving statistical significance ( P <.05), including mel-frequency cepstral coefficients, spectral characteristics, and voice quality measures, showing strong correlations with diagnostic probabilities (|ρ|=0.35‐0.85) and symptom severity (HAMD-17: ρ=–0.51; YMRS: ρ=–0.37); 2 to 4 linguistic features per comparison also reached significance. With sample independence ensured for interstate comparisons, DS vs ES significantly exceeded chance (AUC=0.760, permutation P =.03), whereas MS/MX vs ES (AUC=0.601, P =.20) and DS vs MS/MX (AUC=0.569, P =.30) did not, suggesting that markers primarily reflect persistent disease characteristics with partial state-associated components in depressive states. Conclusions: The combined acoustic-linguistic model, driven primarily by acoustic features, shows potential as an objective aid for BP diagnosis (AUC>0.9), largely free of medication confounding; some features remained detectable during euthymia, although ES discrimination could not be fully separated from concurrent antipsychotic treatment. Although DS showed partial distinguishability from ES, overall limitations remain for monitoring mood state changes. Findings require validation in diverse populations and against other psychiatric conditions.
- 28 March 2026
- Published by Springer Nature in The Review of Socionetwork Strategies
- Vol. 20 (1) , 181-182
Preface of Special Issue on 12th Competition on Legal Information of Extraction and Entailment (COLIEE 2025)
- 28 March 2026
- Published by Springer Nature in The Review of Socionetwork Strategies
- Vol. 20 (1) , 181-182
Preface of Special Issue on 12th Competition on Legal Information of Extraction and Entailment (COLIEE 2025)
- Open Access
- 18 March 2026
- Published by Springer Nature in The Review of Socionetwork Strategies
- Vol. 20 (1) , 341-359
KIS: COLIEE 2025 Task 4 Solver Using Japanese LLM
This paper presents a controlled comparison of prompting strategies for the COLIEE Task 4 legal entailment task under a unified experimental setting. Using a Japanese instruction-following LLM, we compare zero-shot (ZS), zero-shot Chain-of-Thought (ZS-CoT), and structured Chain-of-Thought (Structured-CoT). Across seven test years (2018–2024), the overall (sample-weighted) accuracy improves along the progression ZS $$\rightarrow $$ ZS-CoT $$\rightarrow $$ Structured-CoT under the unified setting, suggesting that prompt structure—rather than model changes—drives the gain in this setting. In addition, we report few-shot prompting with similarity-based retrieval and label balancing as a secondary sensitivity analysis over the number of shots. Few-shot effects vary by year and n , and we do not observe a consistent monotonic improvement as n increases.- 18 March 2026
- Published by Springer Nature in The Review of Socionetwork Strategies
- Vol. 20 (1) , 341-359
Open AccessKIS: COLIEE 2025 Task 4 Solver Using Japanese LLM
This paper presents a controlled comparison of prompting strategies for the COLIEE Task 4 legal entailment task under a unified experimental setting. Using a Japanese instruction-following LLM, we compare zero-shot (ZS), zero-shot Chain-of-Thought (ZS-CoT), and structured Chain-of-Thought (Structured-CoT). Across seven test years (2018–2024), the overall (sample-weighted) accuracy improves along the progression ZS $$\rightarrow $$ ZS-CoT $$\rightarrow $$ Structured-CoT under the unified setting, suggesting that prompt structure—rather than model changes—drives the gain in this setting. In addition, we report few-shot prompting with similarity-based retrieval and label balancing as a secondary sensitivity analysis over the number of shots. Few-shot effects vary by year and n , and we do not observe a consistent monotonic improvement as n increases. - Open Access
- 3 March 2026
- Published by Springer Nature in The Review of Socionetwork Strategies
- Vol. 20 (1) , 313-339
Japanese Legal Judgment Prediction Using ModernBERT and Generative AI-based Information Extraction
Legal Judgment Prediction (LJP) has emerged as a promising research topic in the legal domain, aiming to assist decision-making processes by predicting judicial outcomes. In the COLIEE 2025 shared task, a new pilot subtask named LJPJT 2025 was introduced, focusing on Japanese civil tort cases. This subtask consists of two binary classification tasks: Tort Prediction (TP) and Rationale Extraction (RE). In the first part of this paper, we propose a simple fine-tuned ModernBERT-based system for LJPJT 2025, which achieved competitive results, including the highest F1 score among all participants in the RE task and a top-ranked performance in the TP task. In addition to these two classification tasks, we further present an exploratory information extraction task that aims to automatically extract textual segments corresponding to the plaintiff’s or defendant’s claims and the alleged tort instances from raw judgment document. This task is designed to automate the annotation process that has so far relied entirely on manual efforts in previous studies. In contrast to the strong results obtained in the TP and RE tasks, our baseline experiments on this new information extraction task revealed considerably lower performance, highlighting the substantial challenges that remain for achieving end-to-end judgment prediction directly from raw judgment document.
- 3 March 2026
- Published by Springer Nature in The Review of Socionetwork Strategies
- Vol. 20 (1) , 313-339
Open AccessJapanese Legal Judgment Prediction Using ModernBERT and Generative AI-based Information Extraction
Legal Judgment Prediction (LJP) has emerged as a promising research topic in the legal domain, aiming to assist decision-making processes by predicting judicial outcomes. In the COLIEE 2025 shared task, a new pilot subtask named LJPJT 2025 was introduced, focusing on Japanese civil tort cases. This subtask consists of two binary classification tasks: Tort Prediction (TP) and Rationale Extraction (RE). In the first part of this paper, we propose a simple fine-tuned ModernBERT-based system for LJPJT 2025, which achieved competitive results, including the highest F1 score among all participants in the RE task and a top-ranked performance in the TP task. In addition to these two classification tasks, we further present an exploratory information extraction task that aims to automatically extract textual segments corresponding to the plaintiff’s or defendant’s claims and the alleged tort instances from raw judgment document. This task is designed to automate the annotation process that has so far relied entirely on manual efforts in previous studies. In contrast to the strong results obtained in the TP and RE tasks, our baseline experiments on this new information extraction task revealed considerably lower performance, highlighting the substantial challenges that remain for achieving end-to-end judgment prediction directly from raw judgment document.
- Open Access
- 21 August 2025
- Published by Taylor & Francis in The Journal of Legislative Studies
- Vol. 32 (3) , 1-19
Compromise or differentiate? The role of partner popularity in coalition parties’ strategic parliamentary speech
- 21 August 2025
- Published by Taylor & Francis in The Journal of Legislative Studies
- Vol. 32 (3) , 1-19
Open AccessCompromise or differentiate? The role of partner popularity in coalition parties’ strategic parliamentary speech
- Open Access
- 16 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 506-515
An Overview of the COLIEE 2025 Competition: Legal Case Law and Statute Law Information Retrieval and Entailment
- 16 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 506-515
Open AccessAn Overview of the COLIEE 2025 Competition: Legal Case Law and Statute Law Information Retrieval and Entailment
- Open Access
Few Dimensions are Enough: Fine-tuning BERT with Selected Dimensions Revealed Its Redundant Nature
When fine-tuning BERT models for specific tasks, it is common to select part of the final layer's output and input it into a newly created fully connected layer. However, it remains unclear which part of the final layer should be selected and what information each dimension of the layers holds. In this study, we comprehensively investigated the effectiveness and redundancy of token vectors, layers, and dimensions through BERT fine-tuning on GLUE tasks. The results showed that outputs other than the CLS vector in the final layer contain equivalent information, most tasks require only 2-3 dimensions, and while the contribution of lower layers decreases, there is little difference among higher layers. We also evaluated the impact of freezing pre-trained layers and conducted cross-fine-tuning, where fine-tuning is applied sequentially to different tasks. The findings suggest that hidden layers may change significantly during fine-tuning, BERT has considerable redundancy, enabling it to handle multiple tasks simultaneously, and its number of dimensions may be excessive.
Open AccessFew Dimensions are Enough: Fine-tuning BERT with Selected Dimensions Revealed Its Redundant Nature
When fine-tuning BERT models for specific tasks, it is common to select part of the final layer's output and input it into a newly created fully connected layer. However, it remains unclear which part of the final layer should be selected and what information each dimension of the layers holds. In this study, we comprehensively investigated the effectiveness and redundancy of token vectors, layers, and dimensions through BERT fine-tuning on GLUE tasks. The results showed that outputs other than the CLS vector in the final layer contain equivalent information, most tasks require only 2-3 dimensions, and while the contribution of lower layers decreases, there is little difference among higher layers. We also evaluated the impact of freezing pre-trained layers and conducted cross-fine-tuning, where fine-tuning is applied sequentially to different tasks. The findings suggest that hidden layers may change significantly during fine-tuning, BERT has considerable redundancy, enabling it to handle multiple tasks simultaneously, and its number of dimensions may be excessive.
- Open Access
- 1 November 2024
- Published by Japanese Society for Artificial Intelligence in Transactions of the Japanese Society for Artificial Intelligence
- Vol. 39 (6) , D-O33_1-O33_1
Liar Detector Trained by Japanese Language Expressions and Validation of Domain Independence
The paper discusses the challenges of detecting deception in text communication, where non-verbal cues like tone of voice and facial expressions are absent, making it more difficult than in-person communication. It addresses the issue of scarcity in labeled corpora for lie detection research. The study uses data from a Werewolf BBS (bulletin board system), which, while extensive and useful, might be too specific to the Werewolf game context in terms of vocabulary and expressions. To address this, the research filters data from a general domain, X’s posting data, using only common vocabulary to enhance generalizability. Additionally, a new dataset for a murder mystery game was created to further validate the model trained on Werewolf BBS logs for cross-domain applicability. Experiments in lie detection were conducted using several different learning methods, with the BERT+GRU hierarchical model achieving an F1 score of 0.70 before filtering and 0.68 after, demonstrating the feasibility of lie detection in general domains.
- 1 November 2024
- Published by Japanese Society for Artificial Intelligence in Transactions of the Japanese Society for Artificial Intelligence
- Vol. 39 (6) , D-O33_1-O33_1
Open AccessLiar Detector Trained by Japanese Language Expressions and Validation of Domain Independence
The paper discusses the challenges of detecting deception in text communication, where non-verbal cues like tone of voice and facial expressions are absent, making it more difficult than in-person communication. It addresses the issue of scarcity in labeled corpora for lie detection research. The study uses data from a Werewolf BBS (bulletin board system), which, while extensive and useful, might be too specific to the Werewolf game context in terms of vocabulary and expressions. To address this, the research filters data from a general domain, X’s posting data, using only common vocabulary to enhance generalizability. Additionally, a new dataset for a murder mystery game was created to further validate the model trained on Werewolf BBS logs for cross-domain applicability. Experiments in lie detection were conducted using several different learning methods, with the BERT+GRU hierarchical model achieving an F1 score of 0.70 before filtering and 0.68 after, demonstrating the feasibility of lie detection in general domains.
- 28 May 2024
- Published by Springer Nature
- p. 140-155
LLM Tuning and Interpretable CoT: KIS Team in COLIEE 2024
- 28 May 2024
- Published by Springer Nature
- p. 140-155
LLM Tuning and Interpretable CoT: KIS Team in COLIEE 2024
- 28 May 2024
- Published by Springer Nature
- p. 109-124
Overview of Benchmark Datasets and Methods for the Legal Information Extraction/Entailment Competition (COLIEE) 2024
- 28 May 2024
- Published by Springer Nature
- p. 109-124
Overview of Benchmark Datasets and Methods for the Legal Information Extraction/Entailment Competition (COLIEE) 2024