Abstract:There are large and growing textual corpora in which people express contrastive opinions about the same topic. This has led to an increasing number of studies about contrastive opinion mining. However, there are several notable issues with the existing studies. They mostly focus on mining contrastive opinions from multiple data collections, which need to be separated into their respective collections beforehand. In addition, existing models are opaque in terms of the relationship between topics that are extrac… Show more
“…Other studies (Kim and Zhai, 2009;Huang et al, 2011;Sipos and Joachims, 2013;Ren et al, 2017) tackled similar tasks by developing extracting sentences/phrases from given sets of documents for comparative document analysis. Topic models have also been used to capture comparative topics for better understanding text corpora, but they do not generate textual summaries (Ren and de Rijke, 2015;He et al, 2016;Ibeke et al, 2017).…”
Section: Target Entity Id: 614392 Vs Counterpart Entity Id: 256595mentioning
Opinion summarization focuses on generating summaries that reflect popular subjective information expressed in multiple online reviews. While generated summaries offer general and concise information about a particular hotel or product, the information may be insufficient to help the user compare multiple different choices. Thus, the user may still struggle with the question "Which one should I pick?" In this paper, we propose the comparative opinion summarization task, which aims at generating two contrastive summaries and one common summary from two different candidate sets of reviews. We develop a comparative summarization framework COCOSUM, which consists of two base summarization models that jointly generate contrastive and common summaries. Experimental results on a newly created benchmark COCOTRIP show that COCOSUM can produce higher-quality contrastive and common summaries than stateof-the-art opinion summarization models. The dataset and code are available at https:// github.com/megagonlabs/cocosum.
“…Other studies (Kim and Zhai, 2009;Huang et al, 2011;Sipos and Joachims, 2013;Ren et al, 2017) tackled similar tasks by developing extracting sentences/phrases from given sets of documents for comparative document analysis. Topic models have also been used to capture comparative topics for better understanding text corpora, but they do not generate textual summaries (Ren and de Rijke, 2015;He et al, 2016;Ibeke et al, 2017).…”
Section: Target Entity Id: 614392 Vs Counterpart Entity Id: 256595mentioning
Opinion summarization focuses on generating summaries that reflect popular subjective information expressed in multiple online reviews. While generated summaries offer general and concise information about a particular hotel or product, the information may be insufficient to help the user compare multiple different choices. Thus, the user may still struggle with the question "Which one should I pick?" In this paper, we propose the comparative opinion summarization task, which aims at generating two contrastive summaries and one common summary from two different candidate sets of reviews. We develop a comparative summarization framework COCOSUM, which consists of two base summarization models that jointly generate contrastive and common summaries. Experimental results on a newly created benchmark COCOTRIP show that COCOSUM can produce higher-quality contrastive and common summaries than stateof-the-art opinion summarization models. The dataset and code are available at https:// github.com/megagonlabs/cocosum.
“…Based on the sentiment dictionary, Gupta and Yang 40 developed a system to understand and predict the emotion intensity of tweets. Ibeke et al 41 proposed a novel unified latent variable model (contraLDA) to extract sentences expressing comparative opinions and compare opinions’ intensity.…”
“Social sensors” refer to those who provide opinions through electronic communication channels such as social networks. There are two major issues in current models of sentiment analysis in social sensor networks. First, most existing models only analyzed the sentiment within the text but did not analyze the users, which led to the experimental results difficult to explain. Second, few studies extract the specific opinions of users. Only analyzing the emotional tendencies or aspect-level emotions of social users brings difficulties to the analysis of the opinion evolution in public emergencies. To resolve these issues, we propose an explainable sentiment prediction model based on the portraits of users sharing representative opinions in social sensors. Our model extracts the specific opinions of the user groups on the topics and fully considers the impacts of their diverse features on sentiment analysis. We conduct experiments on 51,853 tweets about the “COVID-19” collected from 1 May 2020 to 9 July 2020. We build users’ portraits from three aspects: attribute features, interest features, and emotional features. Six machine learning algorithms are used to predict emotional tendency based on users’ portraits. We analyze the influence of users’ features on the sentiment. The prediction accuracy of our model is 64.88%.
“…The pandemic disrupted activities all over the world, as schools and education systems were halted and eventually proceeded to online studies [5,6]. Another worrying development of the pandemic was information overload, which included misinformation and confusing messages (e.g., 'fake news') [7], leading to contrasting opinions among the general public [8,9].…”
This study aimed to analyse public sentiments of UK-originated tweets related to COVID-19 vaccines, and it applied six chronological time periods, between January and December 2021. The dates were related to six BBC news reports about the most significant developments in the three main vaccines that were being administered in the UK at the time: Pfizer-BioNTech, Moderna, and Oxford-AstraZeneca. Each time period spanned seven days, starting from the day of the news report. The study employed the bidirectional encoder representations from transformers (BERT) model to analyse the sentiments in 4172 extracted tweets. The BERT model adopts the transformer architecture and uses masked language and next sentence prediction models. The results showed that the overall sentiments for all three vaccines were negative across all six periods, with Moderna having the least negative tweets and the highest percentage of positive tweets overall while AstraZeneca attracted the most negative tweets. However, for all the considered time periods, Period 3 (23–29 May 2021) received the least negative and the most positive tweets, following the related BBC report—’COVID: Pfizer and AstraZeneca jabs work against Indian variant’—despite reports of blood clots associated with AstraZeneca during the same time period. Time periods 5 and 6 had no breaking news related to COVID vaccines, and they reflected no significant changes. We, therefore, concluded that the BBC news reports on COVID vaccines significantly impacted public sentiments regarding the COVID-19 vaccines.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.