Rational Retrieval Acts: Leveraging Pragmatic Reasoning to Improve Sparse Retrieval
This addresses the problem of improving retrieval accuracy for IR practitioners by introducing a novel adaptation of a linguistics framework, though it is incremental as it builds on existing sparse retrieval methods.
The paper tackles the problem that sparse neural IR methods and traditional models like BM25 do not account for document collection context and term weight interplay, by adapting the Rational Speech Acts (RSA) framework from linguistics to dynamically modulate token-document interactions. The result is consistent improvement in multiple sparse retrieval models, achieving state-of-the-art performance on out-of-domain datasets from the BEIR benchmark.
Current sparse neural information retrieval (IR) methods, and to a lesser extent more traditional models such as BM25, do not take into account the document collection and the complex interplay between different term weights when representing a single document. In this paper, we show how the Rational Speech Acts (RSA), a linguistics framework used to minimize the number of features to be communicated when identifying an object in a set, can be adapted to the IR case -- and in particular to the high number of potential features (here, tokens). RSA dynamically modulates token-document interactions by considering the influence of other documents in the dataset, better contrasting document representations. Experiments show that incorporating RSA consistently improves multiple sparse retrieval models and achieves state-of-the-art performance on out-of-domain datasets from the BEIR benchmark. https://github.com/arthur-75/Rational-Retrieval-Acts