Speaker: Eduard Dragut, Temple University - Speaker will be in person
Talk Title: Neural Approaches for Resource Selection in Distributed Search
Date: Friday, March 13, 2026 - 1:30 - 2:30 PM EST (North American Daylight Saving Time)
Abstract: The rapid growth and increasing heterogeneity of online information sources have renewed interest in Distributed Information Retrieval (DIR) as a scalable alternative to centralized web search. In DIR systems, a query is routed to the most relevant resources before documents are retrieved and ranked. While this problem has been studied extensively, many classical approaches rely on term-based statistics and do not fully exploit modern representation learning methods. This talk revisits DIR using contemporary neural techniques for resource representation and selection. First, we present a graph neural network (GNN)–based approach that models structural relationships between queries and resources. The method uses a pre-trained language model to obtain semantic representations and constructs a heterogeneous graph capturing both query–resource and resource–resource interactions. A GNN then learns representations that integrate semantic and structural signals, leading to improved resource ranking. Second, we introduce Resource2Box, a representation learning method that models resources as box embeddings in latent space. Unlike traditional vector representations, box embeddings capture the semantic diversity within a resource by representing it as a region rather than a point. The model aggregates document information through attentive pooling and learns query–resource relationships using a box–vector distance metric. Together, these approaches demonstrate how modern neural models can significantly improve resource selection in distributed search, revisiting a classical IR problem through the lens of contemporary representation learning.
Bio: Eduard Dragut is a Professor in the Department of Computer and Information Sciences at Temple University. He received his Ph.D. in Computer Science from the University of Illinois at Chicago. His research focuses on data management, information retrieval, and applied artificial intelligence, with an emphasis on building scalable systems for extracting and integrating knowledge from large and heterogeneous data sources. He also pursues interdisciplinary AI projects for social good, including work on assistive technologies such as augmentative and alternative communication (AAC) and AI-driven tools for knowledge discovery. He has published widely in leading venues in databases, natural language processing, and data mining.
Zoom Access: Zoom Link and reach out to Hamed Zamani or Dan Parker for the passcode.
