LongEval: Longitudinal Evaluation of Model Performance at CLEF 2024
摘要
This paper introduces the planned second LongEval Lab, part of the CLEF 2024 conference. The aim of the lab’s two tasks is to give researchers test data for addressing temporal effectiveness persistence challenges in both information retrieval and text classification, motivated by the fact that model performance degrades as the test data becomes temporally distant from the training data. LongEval distinguishes itself from traditional IR and classification tasks by emphasizing the evaluation of models designed to mitigate performance drop over time using evolving data. The second LongEval edition will further engage the IR community and NLP researchers in addressing the crucial challenge of temporal persistence in models, exploring the factors that enable or hinder it, and identifying potential solutions along with their limitations.