Neural Headline Generation with Minimum Risk Training

By Ayana, Shiqi Shen, Zhiyuan Liu, Maosong Sun. 2016

55 citations recorded by the source corpus.

Automatic headline generation is an important research area within text summarization and sentence compression. Recently, neural headline generation models have been proposed to take advantage of well-trained neural networks in learning sentence representations and mapping sequence to sequence. Nevertheless, traditional neural network encoder utilizes maximum likelihood estimation for parameter optimization, which essentially constraints the expected training objective within word level instead of sentence level. Moreover, the performance of model prediction significantly relies on training data distribution. To overcome these drawbacks, we employ minimum risk training strategy in this paper, which directly optimizes model parameters with respect to evaluation metrics and statistically leads to significant improvements for headline generation. Experiment results show that our approach outperforms state-of-the-art systems on both English and Chinese headline generation tasks.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Furu Wei
2. Tie-Yan Liu
3. Tao Qin
4. Xu Tan
5. Mohit Bansal
6. Jianfeng Gao
7. M. Zhou
8. Zhe Gan

In Computer Vision and Pattern Recognition

1. Tao Qin
2. Tie-Yan Liu
3. Xu Tan
4. Furu Wei
5. M. Zhou
6. Di He
7. Jianfeng Gao
8. Xueqi Cheng

In Natural Language Processing

1. Furu Wei
2. Tie-Yan Liu
3. Tao Qin
4. M. Zhou
5. Xu Tan
6. Mohit Bansal
7. Graham Neubig
8. Jun Suzuki

Citations in the corpus

• MASS: Masked Sequence to Sequence Pre-training for Language Generation (2019, background): “…some inputs, including tasks like neural machine translation (NMT) (Cho et al., 2014; Bahdanau et al., 2015a; Vaswani et al., 2017), text summarization (Ayana et al., 2016; Suzuki & Nagata, 2017; Gehring et al., 2017) and conversational response generation (Shang et al., 2015; Vinyals & Le, 2015).”
• Learning to Extract Coherent Summary via Deep Reinforcement Learning (2018, background): “(Ayana et al. 2016; Ranzato et al. 2015) show that directly optimizing the evaluation metrics via reinforcement learning is more effective than optimizing likelihood for the sequence generation problems.”
• Controlling Output Length in Neural Encoder-Decoders (2016, methodology): “There are already many studies that address this task (Nallapati et al., 2016; Ayana et al., 2016; Ranzato et al., 2015; Lopyrev, 2015; Gulcehre et al., 2016; Gu et al., 2016; Chopra et al., 2016).”
• Cutting-off Redundant Repeating Generations for Neural Abstractive Summarization (2016, methodology): “We expect to further improve its performance by applying MRT for its training since recent progress of NMT has suggested leveraging a sequence-wise optimization technique for improving performance Wise-man and Rush (2016); Shen et al. (2016).”
• Retrieve, Rerank and Rewrite: Soft Template Based Neural Summarization (2018, methodology): “For instance, Ayana et al. (2016) applied the Minimum Risk Training strategy to maximize the ROUGE scores of generated sum- maries.”
• Faithful to the Original: Fact Aware Neural Abstractive Summarization (2017, methodology): “For instance, (Ayana, Liu, and Sun 2016) applied Minimum Risk Training strategy to maximize the ROUGE scores of generated summaries.”
• Towards Explainable NLP: A Generative Explanation Framework for Text Classification (2018, background): “Minimum risk training (MRT) aims to minimize the expected loss, i.e., risk over the training data (Ayana et al., 2016).”
• On the Weaknesses of Reinforcement Learning for Neural Machine Translation (2019, methodology): “CMRT was proposed by Och (2003), adapted to NMT by Shen et al. (2016), and often used since (Ayana et al., 2016; Neubig, 2016; Shen et al., 2017; Edunov et al., 2018; Makarov and Clematide, 2018; Neubig et al., 2018).”
• Recent advances in document summarization (2017, methodology): “Meanwhile, directly optimizing ROUGE via reinforcement learning has been shown to be more effective than optimizing likelihood for the decoder generation [3, 159].”
• Selective Encoding for Abstractive Sentence Summarization (2017, methodology): “Ayana et al. (2016) apply MRT on abstractive sentence summarization task and the results show that optimizing for ROUGE improves the test performance.”
• Double Path Networks for Sequence to Sequence Learning (2018, methodology): “4https://github.com/moses-smt/mosesdecoder/blob/master/scripts/generic/multi-bleu. perl We compare our model with some previous work including RNN MRT (Ayana et al., 2016) which adopts RNNSearch on this task, WFE (Suzuki and Nagata, 2017) which addresses the problem of repeated sequence generation, ”
• PROP: Pre-training with Representative Words Prediction for Ad-hoc Retrieval (2020, background): “Recent advances have shown that pre-trained language representation models, such as OpenAI GPT [31], BERT [12] and XLNET [41], can capture rich language information from text, and achieve stateof-the-art accuracy in many downstream natural language processing (NLP) tasks including summarization [3],”