IDEA Lab - Research

Research Areas

Our current research interests include, but are not limited to, the following areas:

Retrieval-Augmented Generation (RAG): Designing retrieval pipelines and knowledge grounding strategies to enhance LLM generation, including ontology construction, document indexing, query formulation, and evidence selection.
Recommendation & Personalization: Modeling user preferences, contexts, and long-term interests for personalized ranking, retrieval, and recommendation, including user modeling and user simulation for evaluation, data generation, and training.
Search in Specialized Domains: Retrieval for domain-specific corpora (e.g., scientific literature, e-commerce, enterprise data), with an emphasis on knowledge structuring and concept-aware retrieval.
Multi-modal and Multi-domain Generalization: Integrating text, image, graph, and other modalities, as well as multiple domains, for retrieval and recommendation.
Continual Retrieval and Recommendation: Methods for handling evolving data distributions, user interest drift, and non-stationary environments in retrieval and recommendation systems.
Efficient and Scalable Systems: Lightweight models, indexing strategies, and efficiency–effectiveness trade-offs for large-scale and resource-constrained settings.

Recommender System

Recommender systems are pivotal in modern digital ecosystems, driving user engagement and satisfaction across various domains such as e-commerce, media streaming, and online education. By filtering vast amounts of information and presenting users with personalized suggestions, these systems enhance decision-making, improve user experiences, and significantly impact business revenue.

We focus on addressing key challenges in real-world applications, such as learning user preferences from limited feedback, optimizing accuracy-efficiency trade-offs, and personalizing recommendations across multi-domain and multi-modal contexts.

Below are examples of recent research projects:

Information Retrieval

Information retrieval lies at the heart of how humans interact with vast amounts of digital information. From web search engines to domain-specific retrieval systems, IR technologies enable users to efficiently locate relevant information in response to their needs. This capability is essential in today’s data-driven world, powering applications such as e-commerce, healthcare, legal research, and scientific discovery.

We focus on tackling practical challenges in various domains, where traditional methods often fall short, including reducing reliance on human annotations, leveraging structured knowledge to enhance relevance prediction, building interactive and conversational systems for improved user engagement.

Below are examples of recent research projects:

Data and Web Mining

Data and web mining play a critical role in extracting actionable insights from the vast and ever-growing volumes of unstructured information on the web. These fields enable us to analyze, understand, and utilize patterns in data to power a wide range of applications, from search engines and recommender systems to fraud detection and social network analysis. By transforming raw data into meaningful knowledge, data and web mining contribute significantly to both user-centric and business-driven decision-making.

The value of data and web mining lies in their ability to uncover hidden structures and relationships within data, enabling systems to make accurate predictions, recommendations, and optimizations. In the context of search and recommendation systems, these techniques facilitate the understanding of user behavior, preferences, and intent, allowing for the highly personalized and context-aware experiences.

Below are examples of recent research projects:

Collaborators

We have closely collaborated with various institutions and companies, including (this list might not be exhaustive):