Selected work

Representative Papers

BookAsSumQA: An Evaluation Framework for Aspect-Based Book Summarization via Question Answering

Nov 09, 2025

Aspect-based summarization of long texts—such as books—suffers from a scarcity of high-quality reference summaries, prohibitively high costs of human evaluation, and poor scalability. Method: This paper proposes BookAsSumQA, the first automated evaluation framework for aspect-based summarization of long literary texts. It leverages narrative knowledge graphs to automatically generate aspect-specific question-answer (QA) pairs, eliminating the need for manually annotated reference summaries. By integrating large language models (LLMs) with retrieval-augmented generation (RAG), it uses QA accuracy as a proxy metric for summary quality. Contribution/Results: Experiments demonstrate that BookAsSumQA effectively discriminates among diverse summarization methods. Notably, it is the first to empirically reveal RAG’s significant superiority over LLM-only approaches in long-document aspect-based summarization. The framework is both scalable and practically applicable, enabling efficient, reference-free evaluation.

0 citationsRead paper

Enhancing Japanese Large Language Models with Reasoning Vectors

Aug 04, 2025

To address the limited reasoning capabilities of Japanese large language models (LLMs) stemming from insufficient training resources, this paper proposes a training-free reasoning capability transfer method. The core idea is to extract “reasoning vectors”—i.e., weight-difference representations—derived from high-performing multilingual or English reasoning models under the task vector paradigm, and then inject them directionally into Japanese LLMs. This approach is augmented with a lightweight post-training strategy to enable cross-lingual, low-overhead reasoning enhancement. Experiments demonstrate significant performance gains across multiple Japanese reasoning benchmarks, without requiring additional labeled data or full-parameter fine-tuning. Importantly, the method preserves the original model architecture and parameters while achieving robust generalization. This work establishes a scalable, highly generalizable paradigm for enhancing reasoning capabilities in low-resource language LLMs.

0 citationsRead paper

Recent publications

Latest Papers

BookAsSumQA: An Evaluation Framework for Aspect-Based Book Summarization via Question Answering

Nov 09, 2025

Aspect-based summarization of long texts—such as books—suffers from a scarcity of high-quality reference summaries, prohibitively high costs of human evaluation, and poor scalability. Method: This paper proposes BookAsSumQA, the first automated evaluation framework for aspect-based summarization of long literary texts. It leverages narrative knowledge graphs to automatically generate aspect-specific question-answer (QA) pairs, eliminating the need for manually annotated reference summaries. By integrating large language models (LLMs) with retrieval-augmented generation (RAG), it uses QA accuracy as a proxy metric for summary quality. Contribution/Results: Experiments demonstrate that BookAsSumQA effectively discriminates among diverse summarization methods. Notably, it is the first to empirically reveal RAG’s significant superiority over LLM-only approaches in long-document aspect-based summarization. The framework is both scalable and practically applicable, enabling efficient, reference-free evaluation.

0 citationsRead paper

Enhancing Japanese Large Language Models with Reasoning Vectors

Aug 04, 2025

To address the limited reasoning capabilities of Japanese large language models (LLMs) stemming from insufficient training resources, this paper proposes a training-free reasoning capability transfer method. The core idea is to extract “reasoning vectors”—i.e., weight-difference representations—derived from high-performing multilingual or English reasoning models under the task vector paradigm, and then inject them directionally into Japanese LLMs. This approach is augmented with a lightweight post-training strategy to enable cross-lingual, low-overhead reasoning enhancement. Experiments demonstrate significant performance gains across multiple Japanese reasoning benchmarks, without requiring additional labeled data or full-parameter fine-tuning. Importantly, the method preserves the original model architecture and parameters while achieving robust generalization. This work establishes a scalable, highly generalizable paradigm for enhancing reasoning capabilities in low-resource language LLMs.

0 citationsRead paper