Research

Knowledge Structuring, Agentic Search & Retrieval

  • • Knowledge structuring and indexing: Automatically construct and refine knowledge structures, ontologies, and search indexes from heterogeneous documents.
  • • Agent-oriented retrieval interfaces: Design how agents query knowledge bases and receive evidence based on their current task state.
  • • Context and search state management: Organize retrieved evidence, search history, and unresolved questions to support continued exploration and reasoning.
  • • Web search and search-augmented LLMs: Develop methods for finding and integrating information from the web and other knowledge sources to support grounded reasoning and generation.
  • • Retrieval learning and evaluation: Generate queries and develop benchmarks to evaluate retrieval quality and its contribution to downstream agent performance.

Data Analysis, Prediction & Decision-Making

  • • Data understanding and insight discovery: Reason over tables, time series, and heterogeneous datasets to identify patterns and generate actionable insights.
  • • Reasoning- and experience-based prediction: Combine data, external knowledge, and past experience for classification, forecasting, and anomaly detection.
  • • Explainable modeling and decision support: Connect predictions and recommendations to relevant evidence, conditions, and relationships in the data.
  • • Feedback-driven learning: Improve analytical strategies and decision policies through interaction with tools, environments, and observed outcomes.

Personalization & User-Centric AI

  • • Personalization benchmarks and evaluation: Assess personalized search, recommendation, and agent behavior across users, tasks, and long-term interactions.
  • • User profiling and memory: Infer preferences and goals from interaction histories, and build memories that can be retrieved and updated over time.
  • • Personalized reward modeling: Learn user-specific criteria for evaluating and optimizing generated content and agent behavior.
  • • User simulation: Model users and their interactions to support the training and evaluation of personalized systems.
  • • Personalized policy learning: Learn what to recommend and how to act using user histories, memories, and feedback.