Diji Yang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Diji Yang has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Jinmeng Rao
Mineral.ai, Mountain View, CA, USA
2 shared publications - Yi Zhang
University of California Santa Cruz, Santa Cruz, CA, USA
2 shared publications - Abhishek Kumar
Meta, Menlo Park, CA, USA
1 shared publication - Benyu Zhang
Meta Platforms, Menlo Park, CA, USA
1 shared publication - Chunxiao Wang
School of Science, Shandong Jianzhu University, Jinan, China
1 shared publication - Hanchao Yu
Meta Platforms, Menlo Park, CA, USA
1 shared publication
Latest Publications
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 5741-5752
Guiding Generative Recommender Systems with Structured Human Priors via Multi-head Decoding
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 5741-5752
Open AccessGuiding Generative Recommender Systems with Structured Human Priors via Multi-head Decoding
- Top Cited PapersOpen Access
- 13 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1305-1315
Knowing You Don't Know: Learning When to Continue Search in Multi-round RAG through Self-Practicing
- 13 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1305-1315
Top Cited PapersOpen AccessKnowing You Don't Know: Learning When to Continue Search in Multi-round RAG through Self-Practicing
- Open Access
- 10 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 730-740
IM-RAG: Multi-Round Retrieval-Augmented Generation Through Learning Inner Monologues
Although the Retrieval-Augmented Generation (RAG) paradigms can use external knowledge to enhance and ground the outputs of Large Language Models (LLMs) to mitigate generative hallucinations and static knowledge base problems, they still suffer from limited flexibility in adopting Information Retrieval (IR) systems with varying capabilities, constrained interpretability during the multi-round retrieval process, and a lack of end-to-end optimization. To address these challenges, we propose a novel LLM-centric approach, IM-RAG, that integrates IR systems with LLMs to support multi-round RAG through learning Inner Monologues (IM, i.e., the human inner voice that narrates one's thoughts). During the IM process, the LLM serves as the core reasoning model (i.e., Reasoner ) to either propose queries to collect more information via the Retriever or to provide a final answer based on the conversational context. We also introduce a Refiner that improves the outputs from the Retriever, effectively bridging the gap between the Reasoner and IR modules with varying capabilities and fostering multi-round communications. The entire IM process is optimized via Reinforcement Learning (RL) where a Progress Tracker is incorporated to provide mid-step rewards, and the answer prediction is further separately optimized via Supervised Fine-Tuning (SFT). We conduct extensive experiments with the HotPotQA dataset, a popular benchmark for retrieval-based, multi-step question-answering. The results show that our approach achieves state-of-the-art (SOTA) performance while providing high flexibility in integrating IR modules as well as strong interpretability exhibited in the learned inner monologue.
- 10 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 730-740
Open AccessIM-RAG: Multi-Round Retrieval-Augmented Generation Through Learning Inner Monologues
Although the Retrieval-Augmented Generation (RAG) paradigms can use external knowledge to enhance and ground the outputs of Large Language Models (LLMs) to mitigate generative hallucinations and static knowledge base problems, they still suffer from limited flexibility in adopting Information Retrieval (IR) systems with varying capabilities, constrained interpretability during the multi-round retrieval process, and a lack of end-to-end optimization. To address these challenges, we propose a novel LLM-centric approach, IM-RAG, that integrates IR systems with LLMs to support multi-round RAG through learning Inner Monologues (IM, i.e., the human inner voice that narrates one's thoughts). During the IM process, the LLM serves as the core reasoning model (i.e., Reasoner ) to either propose queries to collect more information via the Retriever or to provide a final answer based on the conversational context. We also introduce a Refiner that improves the outputs from the Retriever, effectively bridging the gap between the Reasoner and IR modules with varying capabilities and fostering multi-round communications. The entire IM process is optimized via Reinforcement Learning (RL) where a Progress Tracker is incorporated to provide mid-step rewards, and the answer prediction is further separately optimized via Supervised Fine-Tuning (SFT). We conduct extensive experiments with the HotPotQA dataset, a popular benchmark for retrieval-based, multi-step question-answering. The results show that our approach achieves state-of-the-art (SOTA) performance while providing high flexibility in integrating IR modules as well as strong interpretability exhibited in the learned inner monologue.
- 1 April 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 205-209
Multiobjective Particle Swarm Optimization with Improved Selection Strategy for Route Optimization
The Multiobjective Route Optimization (MORO) problem is an extension of the traditional single-object route optimization problem, which aims to find an effective path with several conflicting objectives. Many research studies used multiobjective evolutionary algorithm (MOEA) to solve MORO problem, however, they cannot achieve satisfactory results in both quality and computational speed. In this paper, an improved selection strategy based multiobjective particle swarm optimization (ISSMOPSO) is proposed for MORO. The improved selection strategy tactfully combines the advantages of vector evaluated genetic algorithm (VEGA) and Pareto dominating and dominated relationship based fitness function (PDDR-FF). The selection strategy based on VEGA has a preference for the edge region of the Pareto front, the PDDR-FF-based selection strategy has the tendency converging toward the center area of the Pareto front, which preserve both the convergence rate and the distribution performance. The experimental results show that the convergence of ISSMOPSO is better than comparison algorithms, and the diversity is close to NSGA-II and SPEA2 but better than traditional MOPSO.
- 1 April 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 205-209
Multiobjective Particle Swarm Optimization with Improved Selection Strategy for Route Optimization
The Multiobjective Route Optimization (MORO) problem is an extension of the traditional single-object route optimization problem, which aims to find an effective path with several conflicting objectives. Many research studies used multiobjective evolutionary algorithm (MOEA) to solve MORO problem, however, they cannot achieve satisfactory results in both quality and computational speed. In this paper, an improved selection strategy based multiobjective particle swarm optimization (ISSMOPSO) is proposed for MORO. The improved selection strategy tactfully combines the advantages of vector evaluated genetic algorithm (VEGA) and Pareto dominating and dominated relationship based fitness function (PDDR-FF). The selection strategy based on VEGA has a preference for the edge region of the Pareto front, the PDDR-FF-based selection strategy has the tendency converging toward the center area of the Pareto front, which preserve both the convergence rate and the distribution performance. The experimental results show that the convergence of ISSMOPSO is better than comparison algorithms, and the diversity is close to NSGA-II and SPEA2 but better than traditional MOPSO.