原田 慧
| 情報学専攻 | 教授 |
| Ⅰ類（情報系） | 教授 |
| 産学官連携センター | 教授 |
研究者情報
研究活動情報
受賞
- 受賞日 2025年11月
The 26th International Symposium on Advanced Intelligent Systems (ISIS 2025)
Best Presentation Award, Madoka Hagiri;Kazushi Okamoto;Koki Karube;Kei Harada;Atsushi Shibata - 受賞日 2025年09月
Recommender Systems Challenge 2025 (RecSys Challenge 2025)
2nd Position Award (Academic Track), Kaito Terasaki;Taketo Yoneda;Kiyotakashi Takagawa;Hayato Maruyama;Yongzhi Jin;Hibiki Ayabe;Kei Harada;Kazushi Okamoto - 受賞日 2025年06月
Comparison of Motion Study and Data Science Methods for Proficiency in Disassembly Task Movement via Motion Capture
Sullivan Best Paper Award - 2nd Position, Taku Hayashi;Kei Harada;Koki Karube;Munenori Kakehi;Masao Sugi;Tetsuo Yamada
論文
- Similarity-Based Supervised User Session Segmentation Method for Behavior Logs
Yongzhi Jin; Kazushi Okamoto; Kei Harada; Atsushi Shibata; Koki Karube
Japan Society for Fuzzy Theory and Intelligent Informatics, 出版日 2026年05月, 査読付, In information recommendation, a session refers to a sequence of user actions
within a specific time frame. Session-based recommender systems aim to capture
short-term preferences and generate relevant recommendations. However, user
interests may shift even within a session, making appropriate segmentation
essential for modeling dynamic behaviors. In this study, we propose a
supervised session segmentation method based on similarity features derived
from action embeddings and attributes. We compute the similarity scores between
items within a fixed-size window around each candidate segmentation point,
using four types of features: item co-occurrence embeddings, text embeddings of
titles and brands, and price. These features are used to train supervised
classifiers (LightGBM, XGBoost, CatBoost, support vector machine, and logistic
regression) to predict the session boundaries. We construct a manually
annotated dataset from real user browsing histories and evaluate the
segmentation performance using F1-score, area under the precision-recall curve
(PR-AUC), and area under the receiver operating characteristic curve. The
LightGBM model achieves the best performance, with an F1-score of 0.806 and a
PR-AUC of 0.831. These results demonstrate the effectiveness of the proposed
method for session segmentation and its potential to capture dynamic user
behaviors.
英語 - Correction: Comparison of motion study and data science methods for proficiency in disassembly task movement via motion capture
Taku Hayashi; Kei Harada; Koki Karube; Munenori Kakehi; Masao Sugi; Tetsuo Yamada
責任著者, The International Journal of Advanced Manufacturing Technology, 出版日 2026年04月22日, 査読付
研究論文（学術雑誌）, 英語 - BookAsSumQA: An Evaluation Framework for Aspect-Based Book Summarization via Question Answering
Ryuhei Miyazato; Ting-Ruen Wei; Xuyang Wu; Hsin-Tai Wu; Kei Harada
責任著者, Proceedings of the IJCNLP-AACL2025 SRW, 出版日 2025年12月, 査読付, Aspect-based summarization aims to generate summaries that highlight specific aspects of a text, enabling more personalized and targeted summaries. However, its application to books remains unexplored due to the difficulty of constructing reference summaries for long text. To address this challenge, we propose BookAsSumQA, a QA-based evaluation framework for aspect-based book summarization. BookAsSumQA automatically generates aspect-specific QA pairs from a narrative knowledge graph to evaluate summary quality based on its question-answering performance. Our experiments using BookAsSumQA revealed that while LLM-based approaches showed higher accuracy on shorter texts, RAG-based methods become more effective as document length increases, making them more efficient and practical for aspect-based book summarization.
研究論文（国際会議プロシーディングス）, 英語 - Estimation of Fireproof Structure Class and Construction Year for Disaster Risk Assessment
Hibiki Ayabe; Kazushi Okamoto; Koki Karube; Atsushi Shibata; Kei Harada
Proceedings of the 7th ACM International Conference on Multimedia in Asia, Workshop Visual and Signal Communication Technologies in Design of Housing, Urban Spaces, Local Communities, and Human Behavior, 掲載ページ 1-7, 出版日 2025年12月, 査読付, Structural fireproof classification is vital for disaster risk assessment and insurance pricing in Japan. However, key building metadata such as construction year and structure type are often missing or outdated, particularly in the second-hand housing market. This study proposes a multi-task learning model that predicts these attributes from facade images. The model jointly estimates the construction year, building structure, and property type, from which the structural fireproof class - defined as H (non-fireproof), T (semi-fireproof), or M (fireproof) - is derived via a rule-based mapping based on official insurance criteria. We trained and evaluated the model using a large-scale dataset of Japanese residential images, applying rigorous filtering and deduplication. The model achieved high accuracy in construction-year regression and robust classification across imbalanced categories. Qualitative analyses show that it captures visual cues related to building age and materials. Our approach demonstrates the feasibility of scalable, interpretable, image-based risk-profiling systems, offering potential applications in insurance, urban planning, and disaster preparedness.
研究論文（国際会議プロシーディングス）, 英語 - Unveiling Key Movements in Disassembly Task through Motion Capture Analysis and Machine Learning
Taku Hayashi; Kei Harada; Koki Karube; Tetsuo Yamada; Masao Sugi
責任著者, Asian Journal of Management Science and Applications, 出版日 2025年12月, 査読付
研究論文（学術雑誌）, 英語 - Generation and annotation of item usage scenarios in e-commerce using large language models
Madoka Hagiri; Kazushi Okamoto; Koki Karube; Kei Harada; Atsushi Shibata
Proceedings of the 26th International Symposium on Advanced Intelligent Systems, 出版日 2025年11月, 査読付, Complementary recommendations suggest combinations of useful items that play
important roles in e-commerce. However, complementary relationships are often
subjective and vary among individuals, making them difficult to infer from
historical data. Unlike conventional history-based methods that rely on
statistical co-occurrence, we focus on the underlying usage context that
motivates item combinations. We hypothesized that people select complementary
items by imagining specific usage scenarios and identifying the needs in such
situations. Based on this idea, we explored the use of large language models
(LLMs) to generate item usage scenarios as a starting point for constructing
complementary recommendation systems. First, we evaluated the plausibility of
LLM-generated scenarios through manual annotation. The results demonstrated
that approximately 85% of the generated scenarios were determined to be
plausible, suggesting that LLMs can effectively generate realistic item usage
scenarios.
研究論文（国際会議プロシーディングス）, 英語 - A Universal Framework for Offline Serendipity Evaluation in Recommender Systems via Large Language Models
Yu Tokutake; Kazushi Okamoto; Kei Harada; Atsushi Shibata; Koki Karube
Proceedings of the 34th ACM International Conference on Information and Knowledge Management, 出版日 2025年11月, 査読付, Serendipity in recommender systems (RSs) has attracted increasing attention
as a concept that enhances user satisfaction by presenting unexpected and
useful items. However, evaluating serendipitous performance remains challenging
because its ground truth is generally unobservable. The existing offline
metrics often depend on ambiguous definitions or are tailored to specific
datasets and RSs, thereby limiting their generalizability. To address this
issue, we propose a universally applicable evaluation framework that leverages
large language models (LLMs) known for their extensive knowledge and reasoning
capabilities, as evaluators. First, to improve the evaluation performance of
the proposed framework, we assessed the serendipity prediction accuracy of LLMs
using four different prompt strategies on a dataset containing user-annotated
serendipitous ground truth and found that the chain-of-thought prompt achieved
the highest accuracy. Next, we re-evaluated the serendipitous performance of
both serendipity-oriented and general RSs using the proposed framework on three
commonly used real-world datasets, without the ground truth. The results
indicated that there was no serendipity-oriented RS that consistently
outperformed across all datasets, and even a general RS sometimes achieved
higher performance than the serendipity-oriented RS.
研究論文（国際会議プロシーディングス）, 英語 - 大学図書館利用者の学習状況を考慮した授業関連の図書推薦手法の提案
宮里,龍平; 西野,哲朗; 原田,慧
責任著者, 情報処理学会論文誌数理モデル化と応用（TOM）, 情報処理学会, 18巻, 1号, 掲載ページ 1-9, 出版日 2025年02月25日, 査読付, 推薦システムは，利用者にとって価値があるアイテムを特定することで，利用者のアイテム探しを補助するシステムである．大学図書館においても，貸出履歴や図書の内容情報を使って利用者に図書を推薦する手法が提案されている．しかし，異なる学習状況や目的を持って図書を探しに訪れる利用者を考慮した図書推薦手法は実現されていない．本研究では，授業と関連した図書に，授業科目や用途，難易度，その他図書の特徴を表すラベルを手作業で付加した蔵書DBを構築し，利用者の希望を選択式で受付けることで，利用者の目的と状況に合った図書を推薦することを可能にした．この手法を実装し，既存システムOPACを比較対象とした被験者実験を行った．その結果，「借りてみたい図書を見つけることができる」，「要望が推薦結果によく反映されている」といった観点において，提案手法がOPACを有意に上回ることが示された．
Recommendation system is designed to assist users in finding their desired items by identifying users' preference. In university libraries, recommendation methods have been proposed that use borrowing histories and book content information to recommend books to users. However, existing approaches do not adequately consider users' diverse and dynamic learning situations and objectives when recommending books. The purpose of this study is to propose a recommendation method for recommending class-related books in university libraries. In this study, we created a library database that each book is manually labeled with information on relevant course subjects, intended usage, difficulty level, and other characteristics. Allowing users to select their preferences through a simple choice-based interface, we elicit users' preference and recommend books with considering their purpose and current learning needs. We implemented this method and conducted a comparative experiment with OPAC. The results showed that our proposed system significantly outperformed OPAC in terms of the following aspects: whether you could find desired books and you felt that the recommendations reflected your preferences.
研究論文（学術雑誌）, 日本語 - Generators of Brownian motions on abstract Wiener spaces
Kei Harada
筆頭著者, Noncommutative Harmonic Analysis with Applications to Probability II, Institute of Mathematics Polish Academy of Sciences, 掲載ページ 135-142, 出版日 2010年, 査読付
研究論文（国際会議プロシーディングス）, 英語
MISC
- EnsemHalDet: Robust VLM Hallucination Detection via Ensemble of Internal State Detectors
Ryuhei Miyazato; Shunsuke Kitada; Kei Harada
Vision-Language Models (VLMs) excel at multimodal tasks, but they remain vulnerable to hallucinations that are factually incorrect or ungrounded in the input image. Recent work suggests that hallucination detection using internal representations is more efficient and accurate than approaches that rely solely on model outputs. However, existing internal-representation-based methods typically rely on a single representation or detector, limiting their ability to capture diverse hallucination signals. In this paper, we propose EnsemHalDet, an ensemble-based hallucination detection framework that leverages multiple internal representations of VLMs, including attention outputs and hidden states. EnsemHalDet trains independent detectors for each representation and combines them through ensemble learning. Experimental results across multiple VQA datasets and VLMs show that EnsemHalDet consistently outperforms prior methods and single-detector models in terms of AUC. These results demonstrate that ensembling diverse internal signals significantly improves robustness in multimodal hallucination detection., 出版日 2026年04月03日, arXiv:2604.02784 - NLP2025 Theme Session “AIWolf: Conversation Game of Liar Detection and Persuation with LLM”
Yoshinobu Kano; Fujio Toriumi; Michimasa Inaba; Hirotaka Osawa; Daisuke Katagami; Takashi Otsuki; Claus Aranha; Kei Harada; Takeshi Ito
Association for Natural Language Processing, 出版日 2025年, Journal of Natural Language Processing, 32巻, 2号, 掲載ページ 720-726, 1340-7619, 2185-8314 - A Completely Locale-independent Session-based Recommender System by Leveraging Trained Model
Yu Tokutake; Chihiro Yamasaki; Yongzhi Jin; Ayuka Inoue; Kei Harada
In this paper, we propose a solution that won the 10th prize in the KDD Cup
2023 Challenge Task 2 (Next Product Recommendation for Underrepresented
Languages/Locales). Our approach involves two steps: (i) Identify candidate
item sets based on co-visitation, and (ii) Re-ranking the items using LightGBM
with locale-independent features, including session-based features and product
similarity. The experiment demonstrated that the locale-independent model
performed consistently well across different test locales, and performed even
better when incorporating data from other locales into the training., 出版日 2023年10月11日, CoRR, abs/2310.07281巻, journals/corr/abs-2310-07281, arXiv:2310.07281 - AIWolfDial 2023: Summary of Natural Language Division of 5th International AIWolf Contest.
Yoshinobu Kano; Neo Watanabe; Kaito Kagaminuma; Claus Aranha; Jaewon Lee; Benedek Hauer; Hisaichi Shibata; Soichiro Miki; Yuta Nakamura; Takuya Okubo; Soga Shigemura; Rei Ito; Kazuki Takashima; Tomoki Fukuda; Masahiro Wakutani; Tomoya Hatanaka; Mami Uchida; Mikio Abe; Akihiro Mikami; Takashi Otsuki; Zhiyang Qi; Kei Harada; Michimasa Inaba; Daisuke Katagami; Hirotaka Osawa; Fujio Toriumi
出版日 2023年, INLG (Generation Challenges), 掲載ページ 84-100, 英語, conf/inlg/KanoWKALHSMNOSITFWHUAMOQHIKOT23 - The space of tempered distributions as a k-space
Kei Harada; Hayato Saigo
In this paper, we investigate the roles of compact sets in the space of
tempered distributions $\mathscr{S}^{\prime}$. The key notion is "k-spaces",
which constitute a fairly general class of topological spaces. In a k-space,
the system of compact sets controls continuous functions and Borel measures.
Focusing on the k-space structure of $\mathscr{S}^{\prime}$, we prove some
theorems which seem to be fundamental for infinite dimensional harmonic
analysis from a new and unified view point. For example, the invariance
principle of Donsker for the white noise measure is shown in terms of infinite
dimansional characteristic functions., 出版日 2010年09月08日, arXiv:1009.1429
書籍等出版物
講演・口頭発表等
- TabSynDiT：高カーディナリティな質的変数を含む混合型表形式データの合成データ生成
米田 岳斗; 原田 慧; 岡本 一志; 柴田 淳司; 軽部 幸起
第18回データ工学と情報マネジメントに関するフォーラム
発表日 2026年03月 - 画像情報と画像表現がVLM エージェントのユーザー行動模倣に及ぼす影響の分析
畑中 希栞; 宮里 龍平; 岡本 一志; 軽部 幸起; 柴田 淳司; 原田 慧
言語処理学会第32回年次大会
発表日 2026年03月 - プロービングとアンサンブル学習による大規模視覚言語モデルのハルシネーション検出
宮里 龍平; 岡本 一志; 軽部 幸起; 柴田 淳司; 原田 慧
言語処理学会第32回年次大会
発表日 2026年03月 - GNNを用いた群集心理が影響するバブル発生の分析
宮本 春那; 岡本 一志; 軽部 幸起; 柴田 淳司; 原田 慧
第219回知能システム研究発表会
発表日 2026年02月 - 料理レシピにおける可読性指標の検討
上田 茜; 岡本 一志; 軽部 幸起; 原田 慧; 柴田淳司
IDRユーザフォーラム2025
発表日 2025年11月 - Heterogeneous Feature Integration for Behavioral Profiles
Kaito Terasaki; Taketo Yoneda; Kiyotakashi Takagawa; Hayato Maruyama; Yongzhi Jin; Hibiki Ayabe; Kei Harada; Kazushi Okamoto
口頭発表（一般）, 英語, RecSys Challenge 2025 Workshop, 査読付
発表日 2025年09月22日 - Visualizing and Improving Assembly Task Using Motion Capture Analysis
Koki Karube; Masao Sugi; Kei Harada; Tetsuo Yamada
口頭発表（一般）, 英語, IISE Annual Conference and Expo 2025, 査読付
発表日 2025年06月 - Comparison of Motion Study and Data Science Methods for Proficiency in Disassembly Task Movement via Motion Capture
Taku Hayashi; Kei Harada; Koki Karube; Munenori Kakehi; Masao Sugi; Tetsuo Yamada
口頭発表（一般）, 英語, The 34rd International Conference on Flexible Automation and Intelligent Manufacturing (FAIM 2025), 査読付
発表日 2025年06月 - 深層学習を用いた物件外観画像による築年代推定法の検討
綾部 響己; 岡本 一志; 柴田 淳司; 原田 慧; 軽部 幸起
口頭発表（一般）, 日本語, 第39回人工知能学会全国大会
発表日 2025年05月 - 大規模言語モデルによる商品利用シナリオの生成と評価
羽切 まどか; 岡本 一志; 原田 慧; 柴田 淳司; 軽部 幸起
口頭発表（一般）, 日本語, 第39回人工知能学会全国大会
発表日 2025年05月 - 大規模言語モデルを用いた料理レシピの曖昧表現補完
上田 茜; 岡本 一志; 原田 慧; 柴田 淳司; 軽部 幸起
口頭発表（一般）, 日本語, 第39回人工知能学会全国大会
発表日 2025年05月 - 柑橘類を対象とした推薦システムの実現に向けた基礎的調査
細尾 佳意; 岡本 一志; 原田 慧; 柴田 淳司; 軽部 幸起
口頭発表（一般）, 日本語, 第39回人工知能学会全国大会
発表日 2025年05月 - 大規模言語モデルを用いた関係性抽出に基づく小説のアスペクトベース要約手法の提案
宮里 龍平; Hsin-Tai Wu; 原田 慧; 岡本 一志; 柴田 淳司; 軽部 幸起
口頭発表（一般）, 日本語, 第39回人工知能学会全国大会
発表日 2025年05月 - アスペクトに着目した読者に影響を与える映画レビューの分析
井上 歩香; 岡本 一志; 柴田 淳司; 原田 慧; 軽部 幸起
口頭発表（一般）, 日本語, 第17回データ工学と情報マネジメントに関するフォーラム
発表日 2025年03月 - 色彩調和に基づくファッションコーディネートの分析
齋藤 香莉菜; 岡本 一志; 柴田 淳司; 原田 慧; 軽部 幸起
口頭発表（一般）, 日本語, 第17回データ工学と情報マネジメントに関するフォーラム
発表日 2025年02月 - 文字起こしデータを用いたラジオ番組の基礎的調査
西岡 興平; 岡本 一志; 柴田 淳司; 原田 慧; 軽部 幸起
口頭発表（一般）, 日本語, 第17回データ工学と情報マネジメントに関するフォーラム
発表日 2025年02月 - 築年代推定におけるViTとCNNモデルの比較
綾部 響己; 岡本 一志; 柴田 淳司; 原田 慧; 軽部 幸起
ポスター発表, 日本語, IDRユーザフォーラム2024
発表日 2024年12月 - 料理レシピの曖昧表現に関する調査
上田 茜; 岡本 一志; 原田 慧; 柴田 淳司; 軽部 幸起
ポスター発表, 日本語, IDRユーザフォーラム2024
発表日 2024年12月
