Team 3DS2A at LongEval: Performance Evaluation over Time of IR Systems with Proximity Search and Reranking Components

This report gives an overview of the system developed by Team 3DS2A for Task 1 of the LongEval Lab at CLEF 2025. The team members are students enrolled in the Computer Engineering master’s program at the University of Padua. The team developed an information retrieval system tailored to run on a French-language document corpus. The system was evaluated over a nine-month timespan to assess its robustness in terms of precision and recall over time. Throughout development, multiple techniques were explored, including alternative text analyzers, proximity search, chunk-based indexing, and semantic reranking using sentence embeddings. This report presents the system architecture, the experimental strategies adopted, and the performance achieved during the training phase.