Text Summarization

Given single-documents or multi-documents, summarizing the opinions expressed of the input is a vital task in NLP. I divide the current researches into two categories: keyphrase generation and opinion summarization.

Keyphrase generation: A lot of research has been conducted on generating keyphrases tosummarize various types of text. Early approaches to keyphrase generation extract important phrasesfrom the document as the results. Sequence tagging models havebeen applied to identify keyphrases. Retrieval-based approaches utilize a two-step pipeline to extract and rank candidate keyphrases. Sun et al.[1] adopt an extractive graph-based approach, which applies a point network to generate a set of diverse keyphrases. More recently, abstractive approaches havebeen explored. Chan et al.[2] propose a reinforcement learning approach for neural keyphrase generation that encourages a model to generate both sufficient and accurate keyphrases. Wang et al.[3] propose a topic-aware neural keyphrase generation method toidentify topic words. The methods listed above only consider keyphrase generation given a single document. Our work Abstractive Opinion Tagging considers opinion tagging from multiple documents, that is, from all of the reviewsfor a given item.

Opinion summarization: Opinion summarization has become an emerging research topicin recent years. Early studies on opinion summarization focus onextracting salient sentences from text: Hu and Liu[4] identify item features mentioned in the reviews and thenextract opinion sentences for the identified features. Unsupervised learning methods are utilized to extract a review summary by exploiting review helpfulness ratings[5]. To reduce redundancy, a greedy algorithm is also applied to form the final summaries[6]. Reflecting the most representative opinions from reviewers, manyrecent studies have shown that abstractive approaches are more appropriate for summarizing review text: Geraniet al.[7] utilize a template filling strategy to generate a review summary; Wang and Ling[8] apply an encoder-decoder attentionmodel to generate an abstractive summary for opinionated documents. The objective of the above summarization approaches is togenerate coherent sentences to summarize opinions.

In contrast, we propose the abstractive opinion tagging task so as to generate opinion tags from a large number of user-generated reviews. In our scenario, opinion tags are more concise but without loss of essential information; they should help users comprehend reviews quickly and conveniently (this work has been accepted by WSDM 2021).


[1] Zhiqing Sun, Jian Tang, Pan Du, Zhi-Hong Deng, and Jian-Yun Nie. 2019. Div-GraphPointer: A Graph Pointer Network for Extracting Diverse Keyphrases. In SIGIR.

[2] Hou Pong Chan, Wang Chen, Lu Wang, and Irwin King. 2019. Neural KeyphraseGeneration via Reinforcement Learning with Adaptive Rewards. In ACL.

[3] Yue Wang, Jing Li, Hou Pong Chan, Irwin King, Michael R. Lyu, and Shuming Shi. 2019. Topic-Aware Neural Keyphrase Generation for Social Media Language. In ACL.

[4] Minqing Hu and Bing Liu. 2004. Mining and Summarizing Customer Reviews. In SIGKDD.

[5] Wenting Xiong and Diane J. Litman. 2014. Empirical Analysis of Exploiting Review Helpfulness for Extractive Summarization of Online Reviews. In COLING.

[6] Stefanos Angelidis and Mirella Lapata. 2018. Summarizing Opinions: Aspect Extraction Meets Sentiment Prediction and They Are Both Weakly Supervised. In EMNLP.

[7] Shima Gerani, Yashar Mehdad, Giuseppe Carenini, Raymond T. Ng, and BitaNejat. 2014. Abstractive Summarization of Product Reviews Using Discourse Structure. In EMNLP.

[8] Lu Wang and Wang Ling. 2016. Neural Network-Based Abstract Generation for Opinions and Arguments. In NAACL.

Qintong Li
Qintong Li

My research interests include machine learning, natural language processing, and knowledge reasoning.