Approximate Retrieval over Learned Sparse Representations: An Investigation of the Scalability on Massive Datasets

Approximate sparse retrieval over learned sparse representations has emerged as a promising approach for efficient document retrieval. However, scaling these techniques to massive datasets introduces significant challenges related to indexing efficiency, query latency, and memory consumption. This thesis investigates the scalability of state-of-the-art approximate sparse retrieval methods, with a particular focus on Seismic, a recently proposed retrieval algorithm.