Enhancing Factual Consistency of Abstractive Summarization
This addresses a critical issue for users relying on accurate summaries, though it is incremental as it builds on existing summarization systems.
The paper tackles the problem of factual inconsistencies in automatic abstractive summarization by proposing a fact-aware model (FASum) and a factual corrector (FC), resulting in summaries with higher factual consistency and corrections via minimal keyword modifications.
Automatic abstractive summaries are found to often distort or fabricate facts in the article. This inconsistency between summary and original text has seriously impacted its applicability. We propose a fact-aware summarization model FASum to extract and integrate factual relations into the summary generation process via graph attention. We then design a factual corrector model FC to automatically correct factual errors from summaries generated by existing systems. Empirical results show that the fact-aware summarization can produce abstractive summaries with higher factual consistency compared with existing systems, and the correction model improves the factual consistency of given summaries via modifying only a few keywords.