Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

RECOMP: Improving Retrieval-Augmented LMs with Compression and Selective Augmentation

About

Retrieving documents and prepending them in-context at inference time improves performance of language model (LMs) on a wide range of tasks. However, these documents, often spanning hundreds of words, make inference substantially more expensive. We propose compressing the retrieved documents into textual summaries prior to in-context integration. This not only reduces the computational costs but also relieves the burden of LMs to identify relevant information in long retrieved documents. We present two compressors -- an extractive compressor which selects useful sentences from retrieved documents and an abstractive compressor which generates summaries by synthesizing information from multiple documents. Both compressors are trained to improve LMs' performance on end tasks when the generated summaries are prepended to the LMs' input, while keeping the summary concise.If the retrieved documents are irrelevant to the input or offer no additional information to LM, our compressor can return an empty string, implementing selective augmentation.We evaluate our approach on language modeling task and open domain question answering task. We achieve a compression rate of as low as 6% with minimal loss in performance for both tasks, significantly outperforming the off-the-shelf summarization models. We show that our compressors trained for one LM can transfer to other LMs on the language modeling task and provide summaries largely faithful to the retrieved documents.

Fangyuan Xu, Weijia Shi, Eunsol Choi• 2023

Related benchmarks

TaskDatasetResultRank
Multi-hop Question Answering2WikiMultihopQA
EM40.5
559
Multi-hop Question AnsweringHotpotQA (test)
F158.02
311
Multi-hop Question AnsweringHotpotQA
F1 Score40.1
294
Question Answering2Wiki
EM35.8
241
Question AnsweringBamboogle
EM21.7
227
Multi-hop Question Answering2Wiki
Exact Match29.5
215
Question AnsweringHotpotQA
EM29.9
173
Multi-hop Question Answering2WikiMQA
F1 Score34.3
161
Question AnsweringNQ (test)
EM Accuracy32.85
133
Question AnsweringHotpotQA
F143.2
132
Showing 10 of 100 rows
...

Other info

Follow for update