Inverse Reinforcement Learning for Text Summarization
This work addresses the challenge of generating high-quality summaries for text summarization tasks, offering a novel approach that improves performance over existing methods, though it is incremental as it builds on IRL and summarization techniques.
The paper tackled the problem of training abstractive summarization models by introducing inverse reinforcement learning (IRL) to imitate human summarization behaviors, resulting in summaries that outperformed MLE and RL baselines on metrics like ROUGE, coverage, novelty, compression ratio, factuality, and human evaluations across datasets such as CNN/DailyMail and WikiHow.
We introduce inverse reinforcement learning (IRL) as an effective paradigm for training abstractive summarization models, imitating human summarization behaviors. Our IRL model estimates the reward function using a suite of important sub-rewards for summarization and concurrently optimizes the policy network. Experimental results across datasets in different domains (CNN/DailyMail and WikiHow) and various model sizes (BART-base and BART-large) demonstrate the superiority of our proposed IRL model for summarization over MLE and RL baselines. The resulting summaries exhibit greater similarity to human-crafted gold references, outperforming MLE and RL baselines on metrics such as ROUGE, coverage, novelty, compression ratio, factuality, and human evaluations.