| アイテムタイプ |
学術雑誌論文 / Journal Article(1) |
| 公開日 |
2026-03-02 |
| タイトル |
|
|
タイトル |
Toward Diversified Graph Recommendation via Semantic and Topology Augmentation With LLMs |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Accuracy |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Semantics |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Recommender systems |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Visualization |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Large language models |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Graph neural networks |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Bars |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Topology |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Sensitivity |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Reviews |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
Recommender systems |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
diversity |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
graph neural networks |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
large language models |
| 資源タイプ |
|
|
資源タイプ |
journal article |
| アクセス権 |
|
|
アクセス権 |
open access |
| 著者 |
Pangan, Zachary S.
Peng, Shaowen
若宮, 翔子
荒牧, 英治
|
| 抄録 |
|
|
内容記述タイプ |
Abstract |
|
内容記述 |
Graph-based recommender systems are frequently affected by popularity bias, limiting user exposure to a variety of content through precise but repetitive recommendations. Despite the structural strengths of the model, neighborhood aggregation exacerbates this problem by reinforcing dominant preferences. In this work, we propose a model-agnostic augmentation framework that relies on LLM-generated prompts to enhance both accuracy and diversity in graph recommendation. Our approach introduces a dual augmentation strategy guided by constrained prompt generation, where large language models produce semantic embeddings and topological links from structured user and item prompts. Semantic embeddings and topological links are generated from user and item prompts, while proximal categories are inferred to constrain augmentation within the scope of user preferences. This mechanism prevents over-diversification and ensures relevance, effectively balancing exploration and personalization which is an issue that is underexplored in existing LLM-based recommendation approaches. Experiments on MovieLens-1M and Steam show that our method consistently improves both accuracy and diversity. Specifically, we achieve a 28.95% increase in Recall@20, an 8.00% improvement in Category Coverage, and a 2.80% gain in the Frequency-Aware Discounted Category Coverage (FADCC) metric on MovieLens-1M; and a 50.06%, 1.59%, and 32.04% improvement, respectively, on Steam. These results demonstrate the potential of controlled LLM-guided augmentation to mitigate popularity bias while maintaining semantic fidelity and user relevance. |
| 書誌情報 |
en : IEEE Access
巻 13,
p. 205957-205977,
ページ数 21,
発行日 2025-11-25
|
| 出版者 |
|
|
出版者 |
IEEE |
| ISSN |
|
|
収録物識別子タイプ |
EISSN |
|
収録物識別子 |
2169-3536 |
| 出版者版DOI |
|
|
関連タイプ |
isReplacedBy |
|
|
識別子タイプ |
DOI |
|
|
関連識別子 |
https://doi.org/10.1109/ACCESS.2025.3637140 |
| 出版者版URI |
|
|
関連タイプ |
isReplacedBy |
|
|
識別子タイプ |
URI |
|
|
関連識別子 |
https://ieeexplore.ieee.org/document/11267390 |
| 権利 |
|
|
権利情報Resource |
https://creativecommons.org/licenses/by/4.0/ |
|
権利情報 |
© 2025 The Authors. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ . |
| 著者版フラグ |
|
|
出版タイプ |
NA |
| 助成情報 |
|
|
|
助成機関名 |
Japan Society for the Promotion of Science (JSPS) |
|
|
研究課題番号 |
25K21353 |
|
|
研究課題番号URI |
https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-25K21353/ |
|
|
研究課題名 |
温故知新：古典的推薦アルゴリズムと最先端LLMの融合 |
| 助成情報 |
|
|
|
助成機関名 |
Cabinet Office |
|
|
研究課題番号 |
JPJ012425 |
|
|
研究課題名 |
Cross-Ministerial Strategic Innovation Promotion Program (SIP) on “Integrated Health Care System” |
