Comparative Opinion Summarization via Collaborative Decoding

Iso, Hayate; Wang, Xiaolan; Angelidis, Stefanos; Suhara, Yoshihiko

doi:10.18653/v1/2022.findings-acl.261

Cited by 8 publications

(8 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Early research primarily focused on mining explicit comparative information from massive corpora, such as identifying comparative sentences (Jindal and Liu, 2006), extracting comparable entities (Li et al, 2011), and classifying components of comparison (Beloucif et al, 2022). Recent work focused more on text generation tasks such as generating arguments to answer comparative questions (Chekalina et al, 2021), generating comparable questions from news (Beloucif et al, 2022), and summarizing comparative opinions (Lerman and McDonald, 2009;Iso et al, 2022). The existing techniques were designed for specific tasks and could not generalize across all types of comparative reasoning tasks.…”

Section: Comparative Reasoningmentioning

confidence: 99%

“…Recent research has developed models for a few NLP tasks related to comparing texts, including identifying comparative sentences (Jindal and Liu, 2006), mining comparable entities (Li et al, 2011), identifying comparative aspects from a set of questions (Bondarenko et al, 2022;Beloucif et al, 2022), extracting comparative summaries (Bista et al, 2019), and summarizing different opinions (Iso et al, 2022). Yet, the data collection for these tasks relies on expensive and time-consuming manual annotation.…”

Section: Introductionmentioning

confidence: 99%

“…Yet, the data collection for these tasks relies on expensive and time-consuming manual annotation. As a result, low-resource scenarios are common when it comes to new comparative tasks (Iso et al, 2022). Moreover, the task-specific design of such models limits their general comparative reasoning abilities.…”

Section: Introductionmentioning

confidence: 99%

“…To comprehensively evaluate the comparative reasoning abilities of LMs, we introduce a new benchmark with a suite of comparative reasoning tasks. It contains: (1) comparative question answering (QA), sourced from subsets of HotpotQA and 2WikiQA datasets (Yang et al, 2018;Ho et al, 2020); (2) comparative question generation (QG), including HotpotQG and 2WikiQG which are converted from the QA datasets; (3) comparative summarization, including the Diffen dataset that we crawled and the existing CocoTrip dataset (Iso et al, 2022).…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Pre-training Language Models for Comparative Reasoning

Yu,

Zhang,

et al. 2023

Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing

View full text Add to dashboard Cite

Comparative reasoning is a process of comparing objects, concepts, or entities to draw conclusions, which constitutes a fundamental cognitive ability. In this paper, we propose a novel framework to pre-train language models for enhancing their abilities of comparative reasoning over texts. While there have been approaches for NLP tasks that require comparative reasoning, they suffer from costly manual data labeling and limited generalizability to different tasks. Our approach introduces a novel method of collecting scalable data for text-based entity comparison, which leverages both structured and unstructured data. Moreover, we present a framework of pre-training language models via three novel objectives on comparative reasoning. Evaluation on downstream tasks including comparative question answering, question generation, and summarization shows that our pre-training framework significantly improves the comparative reasoning abilities of language models, especially under low-resource conditions. This work also releases the first integrated benchmark for comparative reasoning.

show abstract

Section: Comparative Reasoningmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Pre-training Language Models for Comparative Reasoning

Yu,

Zhang,

et al. 2023

Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing

View full text Add to dashboard Cite

show abstract

“…Abstractive Opinion Summarization Due to the difficulty of collecting the gold training dataset at scale, many multi-document abstractive opinion summarization studies work on an unsupervised solution to build opinion summarization systems (Chu and Liu, 2019;Bražinskas et al, 2020b;Amplayo and Lapata, 2020;Amplayo et al, 2021;Suhara et al, 2020;Elsahar et al, 2021;Im et al, 2021;Iso et al, 2021Iso et al, , 2022Isonuma et al, 2021;Wang and Wan, 2021).…”

Section: Related Workmentioning

confidence: 99%

Noisy Pairing and Partial Supervision for Opinion Summarization

Iso¹,

Wang²,

Suhara³

2022

Preprint

View full text Add to dashboard Cite

Current opinion summarization systems simply generate summaries reflecting important opinions from customer reviews, but the generated summaries may not attract the reader's attention. Although it is helpful to generate professional reviewer-like summaries from customer reviews automatically, collecting many training pairs of customer and professional reviews is generally tricky. We propose a weakly supervised opinion summarization framework, Noisy Pairing and Partial Supervision (NAPA ) that can build the stylized opinion summarization system with no customerprofessional review pairs. Experimental results show consistent improvements in automatic evaluation metrics, and qualitative analysis shows that our weakly supervised opinion summarization system can generate summaries that look more like those written by professional reviewers.

show abstract

Contrastive text summarization: a survey

Ströhle

Campos

Jatowt

2023

Int J Data Sci Anal

View full text Add to dashboard Cite

In our data-flooded age, an enormous amount of redundant, but also disparate textual data is collected on a daily basis on a wide variety of topics. Much of this information refers to documents related to the same theme, that is, different versions of the same document, or different documents discussing the same topic. Being aware of such differences turns out to be an important aspect for those who want to perform a comparative task. However, as documents increase in size and volume, keeping up-to-date, detecting, and summarizing relevant changes between different documents or versions of it becomes unfeasible. This motivates the rise of the contrastive or comparative summarization task, which attempts to summarize the text of different documents related to the same topic in a way that highlights the relevant differences between them. Our research aims to provide a systematic literature review on contrastive or comparative summarization, highlighting the different methods, data sets, metrics, and applications. Overall, we found that contrastive summarization is most commonly used in controversial news articles, controversial opinions or sentiments on a topic, and reviews of a product. Despite the great interest in the topic, we note that standard data sets, as well as a competitive task dedicated to this topic, are yet to come to be proposed, eventually impeding the emergence of new methods. Moreover, the great breakthrough of using deep learning-based language models for abstract summaries in contrastive summarization is still missing.

show abstract

Comparative Opinion Summarization via Collaborative Decoding

Cited by 8 publications

References 41 publications

Pre-training Language Models for Comparative Reasoning

Pre-training Language Models for Comparative Reasoning

Noisy Pairing and Partial Supervision for Opinion Summarization

Contrastive text summarization: a survey

Contact Info

Product

Resources

About