Selected work

Representative Papers

Continuous Improvement and Parallel Autonomous Exploration: An LLM-Agent Framework for Searching Large Solution Spaces

Aug 04, 2026

This work addresses the challenge of efficiently searching for high-quality, diverse solutions in large-scale structured solution spaces by proposing a novel framework based on large language model (LLM) agents. The framework employs a leaderboard-driven retention set to guide continuous improvement of a single agent, while enabling multiple agents to operate in parallel—autonomously performing analysis, implementation, self-evaluation, and iteration—with coordination handled solely by a dedicated orchestrator agent. By innovatively integrating a continuous-improvement reward loop with fully autonomous parallel exploration, the approach transcends the limitations of conventional single-trajectory optimization paradigms. Evaluated on a product-to-catalog matching task, the single-agent configuration achieves best-in-class coverage rates of 47.8–57.4%, which further improves to 62.8–69.4% with five parallel agents, substantially outperforming the 33.3% baseline.

0 citationsRead paper

Towards Better Search with Domain-Aware Text Embeddings for C2C Marketplaces

Dec 24, 2025

Search on Mercari—the largest Japanese C2C marketplace—faces challenges including short, ambiguous user queries, noisy item titles, and stringent online latency constraints. Method: We propose a domain-aware text embedding method tailored for Japanese C2C search. It introduces role-specific prefixes to model semantic asymmetry between queries and titles, integrated with Matryoshka representation learning to yield compact, truncation-robust embeddings. The model is fine-tuned on query-title pairs derived from real purchase behavior, augmented by prompt engineering and log-driven evaluation. Contribution/Results: Offline experiments demonstrate substantial gains over general-purpose encoders. Human evaluation confirms improvements in proper noun recognition, platform-specific semantic understanding, and term importance modeling. Online A/B testing shows statistically significant increases in revenue per user and search efficiency (p < 0.01).

0 citationsRead paper

Improving Visual Recommendation on E-commerce Platforms Using Vision-Language Models

Oct 15, 2025

To address insufficient visual similarity modeling in e-commerce product recommendation, this work pioneers the integration of the SigLIP vision-language model into recommender systems, proposing a cross-modal semantic alignment framework grounded in image–text contrastive learning. We fine-tune SigLIP using a sigmoid-based contrastive loss and design a lightweight image encoder to generate highly discriminative item embeddings, thereby enhancing visual representation capability. Offline evaluation demonstrates a 9.1% improvement in nDCG@5; online A/B testing shows a 50% increase in click-through rate and a 14% uplift in conversion rate. This study not only validates SigLIP’s effectiveness for e-commerce recommendation but also establishes a practical, business-oriented vision–language joint modeling paradigm. It provides a reusable technical pathway for multimodal recommendation, bridging the gap between foundational vision-language models and real-world industrial applications.

0 citationsRead paper

Using item recommendations and LLMs in marketing email titles

Aug 27, 2025

To address low user engagement and poor interaction rates caused by template-based email subject lines in e-commerce marketing, this paper proposes a large language model (LLM)-driven method for generating personalized email subject lines, tightly integrating user profiling with item recommendation content to enable thematic and user-specific headline generation. This work represents the first systematic deployment of LLMs for email subject line generation at scale—across a user base exceeding one million—in a real-world e-commerce setting. We establish a comprehensive evaluation framework encompassing offline simulation, rigorous A/B testing, and online experimentation. Empirical results demonstrate statistically significant improvements in both email open rates and click-through rates. Crucially, these gains are achieved under strict safety and controllability constraints, ensuring operational stability and business impact—thereby overcoming key limitations of conventional template-driven approaches.

0 citationsRead paper

Recent publications

Latest Papers

Continuous Improvement and Parallel Autonomous Exploration: An LLM-Agent Framework for Searching Large Solution Spaces

Aug 04, 2026

This work addresses the challenge of efficiently searching for high-quality, diverse solutions in large-scale structured solution spaces by proposing a novel framework based on large language model (LLM) agents. The framework employs a leaderboard-driven retention set to guide continuous improvement of a single agent, while enabling multiple agents to operate in parallel—autonomously performing analysis, implementation, self-evaluation, and iteration—with coordination handled solely by a dedicated orchestrator agent. By innovatively integrating a continuous-improvement reward loop with fully autonomous parallel exploration, the approach transcends the limitations of conventional single-trajectory optimization paradigms. Evaluated on a product-to-catalog matching task, the single-agent configuration achieves best-in-class coverage rates of 47.8–57.4%, which further improves to 62.8–69.4% with five parallel agents, substantially outperforming the 33.3% baseline.

0 citationsRead paper

Towards Better Search with Domain-Aware Text Embeddings for C2C Marketplaces

Dec 24, 2025

Search on Mercari—the largest Japanese C2C marketplace—faces challenges including short, ambiguous user queries, noisy item titles, and stringent online latency constraints. Method: We propose a domain-aware text embedding method tailored for Japanese C2C search. It introduces role-specific prefixes to model semantic asymmetry between queries and titles, integrated with Matryoshka representation learning to yield compact, truncation-robust embeddings. The model is fine-tuned on query-title pairs derived from real purchase behavior, augmented by prompt engineering and log-driven evaluation. Contribution/Results: Offline experiments demonstrate substantial gains over general-purpose encoders. Human evaluation confirms improvements in proper noun recognition, platform-specific semantic understanding, and term importance modeling. Online A/B testing shows statistically significant increases in revenue per user and search efficiency (p < 0.01).

0 citationsRead paper

Improving Visual Recommendation on E-commerce Platforms Using Vision-Language Models

Oct 15, 2025

To address insufficient visual similarity modeling in e-commerce product recommendation, this work pioneers the integration of the SigLIP vision-language model into recommender systems, proposing a cross-modal semantic alignment framework grounded in image–text contrastive learning. We fine-tune SigLIP using a sigmoid-based contrastive loss and design a lightweight image encoder to generate highly discriminative item embeddings, thereby enhancing visual representation capability. Offline evaluation demonstrates a 9.1% improvement in nDCG@5; online A/B testing shows a 50% increase in click-through rate and a 14% uplift in conversion rate. This study not only validates SigLIP’s effectiveness for e-commerce recommendation but also establishes a practical, business-oriented vision–language joint modeling paradigm. It provides a reusable technical pathway for multimodal recommendation, bridging the gap between foundational vision-language models and real-world industrial applications.

0 citationsRead paper

Using item recommendations and LLMs in marketing email titles

Aug 27, 2025

To address low user engagement and poor interaction rates caused by template-based email subject lines in e-commerce marketing, this paper proposes a large language model (LLM)-driven method for generating personalized email subject lines, tightly integrating user profiling with item recommendation content to enable thematic and user-specific headline generation. This work represents the first systematic deployment of LLMs for email subject line generation at scale—across a user base exceeding one million—in a real-world e-commerce setting. We establish a comprehensive evaluation framework encompassing offline simulation, rigorous A/B testing, and online experimentation. Empirical results demonstrate statistically significant improvements in both email open rates and click-through rates. Crucially, these gains are achieved under strict safety and controllability constraints, ensuring operational stability and business impact—thereby overcoming key limitations of conventional template-driven approaches.

0 citationsRead paper