Uncovering Flaming Events on News Media in Social Media Praboda Rajapaksha, Reza Farahbakhsh, Noel Crespi, Bruno Defude
Total Page:16
File Type:pdf, Size:1020Kb
Uncovering flaming events on news media in social media Praboda Rajapaksha, Reza Farahbakhsh, Noel Crespi, Bruno Defude To cite this version: Praboda Rajapaksha, Reza Farahbakhsh, Noel Crespi, Bruno Defude. Uncovering flam- ing events on news media in social media. IPCCC 2019: 38th International Performance Computing and Communications Conference, Oct 2019, Londres, United Kingdom. pp.1-9, 10.1109/IPCCC47392.2019.8958759. hal-02363443 HAL Id: hal-02363443 https://hal.archives-ouvertes.fr/hal-02363443 Submitted on 14 Nov 2019 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. Uncovering Flaming Events on News Media in Social Media Praboda Rajapaksha, Reza Farahbakhsh, Noel Crespi, Bruno Defude To cite this version: Praboda Rajapaksha, Reza Farahbakhsh, Noel Crespi, Bruno Defude. Uncovering Flaming Events on News Media in Social Media. International Performance Computing and Communications Conference, Oct 2019, London, United Kingdom. hal-02363443 HAL Id: hal-02363443 https://hal.archives-ouvertes.fr/hal-02363443 Submitted on 14 Nov 2019 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. Uncovering Flaming Events on News Media in Social Media Praboda Rajapaksha∗y, Reza Farahbakhsh∗, Noel¨ Crespi∗, Bruno Defude∗ ∗ Institut Polytechnique de Paris, Telecom SudParis,. CNRS Lab UMR5157, Evry, France. yUva Wellassa University, 90000, Badulla, Sri Lanka. fpraboda.rajapaksha, reza.farahbakhsh, noel.crespi, [email protected] Abstract—Social networking sites (SNSs) facilitate the sharing These comments might be posted by genuine users and fraud of ideas and information through different types of feedback or spam generated content [1]. Two examples of flaming including publishing posts, leaving comments and other type of include 1) ‘Delete your account’: a Clinton-Trump Twitter reactions. However, some comments or feedback on SNSs are 1 inconsiderate and offensive, and sometimes this type of feedback flame war and 2) flame war between Donald Trump and has a very negative effect on a target user. The phenomenon Pope Francis on the popes calling Trump ‘disgraceful’ for known as flaming goes hand-in-hand with this type of posting that his immigration recommendations2. As a result of this high- can trigger almost instantly on SNSs. Most popular users such level visibility, flaming has became an interesting topic among as celebrities, politicians and news media are the major victims social researchers as they seek to understand the phenomena of the flaming behaviors and so detecting these types of events will be useful and appreciated. Flaming event can be monitored and explore the impact of these types of activities on targets. and identified by analyzing negative comments received on a One use-case we target in this study is news media in post. Thus, our main objective of this study is to identify a way Facebook as many American adults consume news on social to detect flaming events in SNS using a sentiment prediction media and majority of them are commonly use Facebook [2]. method. We use a deep Neural Network (NN) model that can In addition we can observe that the number of fans in news me- identity sentiments of variable length sentences and classifies 3 the sentiment of SNSs content (both comments and posts) to dia Facebook pages are considerably higher similar to other discover flaming events. Our deep NN model uses Word2Vec categories. News media tend to publish news items of interest and FastText word embedding methods as its training to explore to more diversified and varied readers, and therefore many which method is the most appropriate. The labeled dataset for news readers interact with news items daily via commenting, training the deep NN is generated using an enhanced lexicon sharing and reacting. A set of news media is relying on the based approach. Our deep NN model classifies the sentiment of a sentence into five classes: Very Positive, Positive, Neutral, content of most popular news media as content consumers Negative and Very Negative. To detect flaming incidents, we [3], [4] and so, it is important to understand these flaming focus only on the comments classified into the Negative and types of events in the news media domain. Thus, our main Very Negative classes. As a use-case, we try to explore the objective is to explore news items’ flaming events in terms of flaming phenomena in the news media domain and therefore negative feedback with insults and other offensive words. The we focused on news items posted by three popular news media on Facebook (BBCNews, CNN and FoxNews) to train and test existence of flamings on news items may reduce number of the model. The experimental results show that flaming events can followers, or sometimes these types of posts can go viral and be detected with our proposed approach, and we explored main increase the number of followers. Therefore detecting flamings characteristics that trigger a flaming event and topics discussed and identifying the topics that the community most strongly in the flaming posts. disagree with is a useful conception for news media. Index Terms—Flaming detection, Sentiment analysis, deep neural networks, social media, Facebook, News media, FastText, Sentiment polarity prediction is one of the main ways of Word2Vec. detecting flaming events in SNS [5], [6]. Many advancements arXiv:1909.07181v1 [cs.SI] 16 Sep 2019 have been made to the sentiment classification methods to I. INTRODUCTION date. However, these methods are domain-specific and there- In modern Internet parlance, cyberbullying has become fore their results are strongly biased on the words used in increasingly common, especially in the Internet communities that domain. As a result, we build a word embedding-based and SNSs such as Facebook, Twitter, Instagram and YouTube. multiclass (5-classes) sentiment classification deep NN-based In this era, several new phenomena have appeared, leading approach by focusing on Facebook news items and variable to important debates and discussions and among them one length user feedbacks that can be modified and applied in any hot topic is ‘Flamings’. The Flaming can be considered as other social media category other than news. In addition, we a serious issue on SNSs where many users express disagree- use an improved lexicon-based sentiment classification method ments, insults or offensive words in the form of comments to generate a true labeling list with which to train the deep NN on a forum, blog or chat room intended to inflame emotions 1https://www.sbs.com.au/news/delete-your-account-clinton-trump-in- and sensibilities of others. These comments do not contribute twitter-flame-war any useful content to the discussion groups and instead at- 2https://www.lifewire.com/what-is-flaming-2483253 tempt to wound another person socially and psychologically. 3https://fanpagelist.com/category/news/ model. The flaming effect analyses will be done based on the TABLE I comments classified as Negative and Very Negative. A flaming STATISTICS OF PUBLIC POSTS OBTAINED FROM BBC, CNN, AND FOXNEWS FACEBOOK PAGES DURING FEBRUARY 2018. (PP REPRESENTS event takes place when many users give negative feedback, and THE ABBREVIATION FOR PRE-PROCESSING) so a post with a large number of negative comments received News Number of comments of within a short time will possibly a flaming event. We will also #Fans media 300 posts 300 posts 100 PPposts 200 PPposts explore what types of topics were mainly affected by flamings Before PP After PP Training Experiment that are published by BBCNews, CNN and FoxNews. BBCNews 46.2M 398,453 312,881 107,874 205,007 CNN 29.9M 595,268 312,881 217,386 288,606 This paper offers the following contributions: FoxNews 16.3M 1,162,734 312,881 280,342 773,439 1) a word embedding-based sentiment prediction deep NN model focused on Facebook comments. 2) an exploration of which word embedding method dataset. Also, the model will consider multi-class labels to (Word2Vec or FastText) works better on Facebook comments. train and experiment with a real datasets. 3) identification of flaming posts on BBCNews, CNN and FoxNews Facebook pages published in February 2018. III. EXPERIMENTS 4) identification of flaming posts’ associated topics. News media flaming events can be detected by identifying negative comments received on shared news items in SNSs. II. LITERATURE SURVEY Therefore, sentiment prediction models can be adapted to Researchers have investigated flamings on YouTube [7], cluster senses of the user feedbacks in order to explore the email threads [8] and comments received on Twitter and existence of flaming events in news media in SNS using a Facebook [5], [6], news sites and news channels [9]. These rule based technique. Many previous sentiment analysis works works show that flaming events can be appeared in any online have used Twitter as the SNS to analyze and build their platforms especially in the SNSs as any user can comment on sentiment prediction models. As Facebook employ different a public content. Sentiment polarity prediction is one of the content properties such as permitting to share variable length main ways of detecting flaming events in SNSs [5], [6].