Skip to content

Research

II-Research develops methods that help people find, understand, and trust information. Our work spans information retrieval, generative AI, multimodal learning, knowledge representation, and human-centered computing.

Trustworthy Information Retrieval

Information retrieval systems influence how people acquire knowledge and make decisions. We study retrieval technologies that are effective, transparent, privacy-aware, and robust across domains such as digital libraries, news, healthcare, patents, and e-commerce.

We explore search systems that use large language models to help users formulate needs, retrieve evidence, synthesize information, and complete complex information-seeking tasks.

Privacy-preserving Information Retrieval

Interaction logs are valuable for improving search, but they can contain sensitive personal information. We investigate scalable learning and retrieval methods that protect user privacy while preserving system utility.

Multimodal Information Retrieval

We align text, images, audio, video, and other modalities to support richer and more context-aware search experiences.

Conversational Information Retrieval

We study context-aware retrieval that learns from previous queries and interactions. Topics include user modeling, intent understanding, behavior analysis, and adaptive ranking.

Retrieval-Augmented Generation (RAG)

Large language models can produce inaccurate or unsupported content. We develop retrieval and evidence-grounding methods that improve factuality, attribution, controllability, and evaluation in generative systems.

Evaluation Methods

As search evolves from ranked links to generated answers, evaluation must evolve as well. We design metrics and experimental methods for relevance, diversity, grounding, reliability, and human utility.

Commonsense Reasoning

Commonsense knowledge is natural for people but difficult for machines to acquire and apply. We investigate generative and multimodal reasoning methods that produce outputs consistent with everyday knowledge and real-world constraints.

User Understanding

Search interactions reveal how users express information needs and make choices. We study user modeling, intent analysis, and behavior understanding for information retrieval and recommender systems, with careful attention to privacy and responsible use.

Knowledge Graphs

Knowledge graphs represent entities, concepts, and their relationships in a structured form. Our interests include persona knowledge graphs, affordance knowledge graphs, graph representation learning, and knowledge-enhanced retrieval.

Computational Human-value Understanding

Advanced AI systems can affect individuals and societies in ways that are not fully anticipated. We study datasets, models, and evaluation methods that help AI systems understand human values and align their behavior with human needs.

Current Topics

  • Dataset construction for human-value understanding
  • Human-value-aware evaluation
  • Alignment of large language models
  • Misinformation mitigation