AI helps you reading Science

AI generates interpretation videos

AI extracts and analyses the key points of the paper to generate videos automatically


pub
Go Generating

AI Traceability

AI parses the academic lineage of this thesis


Master Reading Tree
Generate MRT

AI Insight

AI extracts a summary of this paper


Weibo:
On NewsFN dataset the pre-trained Bidirectional Encoder Representations from Transformers model is able to classify news articles with accuracy of 97.021% which is a significant improvement over other traditional approach

Classification of Fake News by Fine-tuning Deep Bidirectional Transformers based Language Model

EAI Endorsed Trans. Scalable Inf. Syst., no. 27 (2020): 163973

Cited by: 0|Views12
EI

Abstract

With the ever-increasing rate of information dissemination and absorption, “Fake News” has become a real menace. People these days often fall prey to fake news that is in line with their perception. Checking the authenticity of news articles manually is a time-consuming and laborious task, thus, giving rise to the requirement for automate...More

Code:

Data:

0
Introduction
  • The veracity and trustworthiness of news is a crucial issue of the modern world.
  • Fake news can be termed as “news articles that are intentionally and verifiably false”.
  • A situation where people don’t completely trust their news sources has been created due to promulgation of fake news, as per the Gallup polls [2], only thirty-two percent of Americans trust their news sources to be fully accurate and fair.
Highlights
  • The veracity and trustworthiness of news is a crucial issue of the modern world
  • A powerful and time efficient approach is proposed for accurately classifying news articles into two classes: Fake and Real
  • In this research study, we utilized the robustness of pre-trained Bidirectional Encoder Representations from Transformers (BERT) language model and applied a highly successful approach of transfer learning for converting the model into a classification model
  • We built two more models for comparison and computed values for various evaluation metrics to support the performance of the proposed method of classifying fake news
  • On NewsFN dataset the pre-trained BERT model is able to classify news articles with accuracy of 97.021% which is a significant improvement over other traditional approach
  • The overall performance of this approach can be improved by fine-tuning larger BERT models, provided the available dataset is large enough and there are enough computational resources to handle the increased computational complexity
Methods
  • Fake news keeps evolving every day, which brings the need of creating an end-to-end classification model which is robust and requires minimal computation and preprocessing.
  • Previous work in pre-training representations like in OpenAI GPT and ELMo are unidirectional and shallow bidirectional respectively, as opposed to BERT which is deeply bidirectional.
  • BERT removes the constraint provided by the unidirectional approach by using Masked LM (Masked Language Model) as a pre-training objective.
  • BERT crossed the threshold of eleven state-of-art NLP tasks.
  • BERT provided them with an approach that can yield state-of-art results without using heavily engineered and task-specific architectures.
Results
  • For performance evaluation of the model, accuracy is chosen as the primary metric for evaluation since the training set, as well as the test set, are completely balanced.
  • Each of the words in the texts needs to be represented in numerical form before passing data through LSTM layers, an embedding layer of size 400 is added whose parameters are trained along with LSTM’s parameters.
  • Values for other evaluation metrics are compared in Table 5
Conclusion
  • Detection of Fake News is gaining a lot of traction among researchers because of its complexity, and the requirement for an algorithm that can filter thousands of news articles and judge their authenticity in a matter of minutes.
  • In this paper a framework based on natural language processing (NLP) is proposed to address this task of classifying news articles as either fake or real using Fine Tuned BERT model.
  • On NewsFN dataset the pre-trained BERT model is able to classify news articles with accuracy of 97.021% which is a significant improvement over other traditional approach.
  • One of the most concerning factors for this research task is to get a properly labeled dataset, as currently, there is no particular dataset that is diverse enough to build a state-of-the-art mechanism for fake news detection
Tables
  • Table1: Statistical information about dataset before and after removing outliers
  • Table2: Demographics of Training and Testing sets
  • Table3: Variation of BERT models open-sourced by Google Research
  • Table4: Accuracy comparison for three models
  • Table5: Comparison of Precision; Recall; F1Score and ROC AUC Score for predictions on
Download tables as Excel
Related work
  • In this section, we start by discussing how definition of “Fake news” has evolved over time. Then, we discuss existing works and methods that are applicable for the task of fake news classification. Different researchers have applied significantly different approaches for tackling fake news and to achieve decent progress on combating this challenge, so we study these approaches based on their type i.e. whether the method applied is content-based, feedbackbased or is based on the social media engagement of users. After this, we give an overview of available datasets that have been used in the past for this task. Disinformation and misinformation which is colloquially known as “Fake News”, isn’t a new phenomenon. It has recently garnered much attention due to 2016 US presidential elections, as can be observed by looking at the term on Google Trends [22]. Misinformation was present before 2016 election as well, as is evident by studies conducted on misinformation before 2016 which shows that misinformation has wide ranging effects that range from financial loss, to politics. One such instance is a 2008 false bankruptcy story about UAL parent company which led to 76% drop in stock price [23]. In content-based based methods, the fundamental basis is that the textual and linguistic features of a real news will differ from that of a fake news. There are hand-engineered ways of extracting these cues as well as more recent DL methods. One of the earliest hand-engineered featuresbased methods, Scientific Content Analysis (SCAN) proposed in 1987 was primarily developed for polygraph examinations and consisted of cues such as grammatical errors, continuity in written paragraphs and provided information [24]. While the method did seem promising in its early days but was later proved ineffective [25, 26]. SCAN also required experts to rigorously analyse the content. As, there always has been efforts to decrease human labour for these tasks, another linguistic-based method was developed by Fuller et al [27]. Authors created a comprehensive set of 31 linguistic cues which were further refined using 3 classifiers to have only 8 cues for deception detection. These cues were based on the previously proposed different cue sets in the linguistic field [28 – 30]. One main limitation of this work was that the cues were largely dependent upon topic or domain of the text and the model was not able to generalize well when tested on contents from different domains [31]. Relatively recent feature-based methods include analysis-based methods such as punctuation marks [32], regular expressions [33], platform (Twitter, Facebook, Wikipedia, etc.) specific cues such as like counts, hashtags [34, 35]. While the method of hand-designing the features and cues is much interpretable but also has disadvantages such as need of re-drawing based on domain, platform or situation of the content, human involvements and lack of generalization. To improve generalization ability of detection models, researchers have also utilized more effective ways of extracting features such as N-gram [36, 37]. Term frequently vectors are created using N-grams and then these are sent to different classifiers like Support Vector Machine (SVM). While using N-grams did improve the performance but being a simple approach could not capture all the features in the different writing styles. Some researchers also devised the classification models that instead of being word-based like N-grams, are based on the syntactical part of writing that exploits Part of speech (POS) tags or are derived from Probabilistic Context Free Grammars (PCFG) [38,39,40]. These approaches lacked ability to capture clues across long news articles and were even weaker as compared to word-based approaches. Process of feature extraction is now automated with the advent of DL. Deep neural networks are able to extract simple as well as complex features that are not intuitive. Wang et al used two popular forms of neural networks, one is Convolutional Neural Network (CNN) and the other is bidirectional LSTMs for embedding the statement text and speaker metadata information into lower dimensions and then fed to classifier for classifying fake news based on the content [41]. They also made use of word embedding known as word2vec for capturing useful contextual properties [42]. Quian et al proposed a Two-level CNN which first generates sentence embedding using words and then utilize the sentence embedding to create article embedding [43]. Their proposed variant of CNN was able to outperform the generic CNN. Variants of CNNs and RNNs have occasionally been used over the past decade for the same task [44 - 46]. While most of the work done on detection of Fake News has been on building supervised models, there are unsupervised techniques that have been employed to detect the credibility of a post. Yang et al uses opinions of users on social media towards authenticity of a story and uses Bayesian networks to build a probabilistic graphical model that treats truth of news and user credibility as latent random variables [47]. The underlying basis of feed-back based methods are the secondary information such as user’s comments, news’ propagation graph in the social media and other userrelated information. Researchers have tried developing hand-engineered features for these methods such as number of followers, content of tweets, depth of retweets, geographical location etc. [48 - 50]. The route of retweets or shares of a news articles and how it propagates through the social media web has been extensively analysed by researchers. Ma et al utilized Jaccard similarity to compute similarity scores of propagation trees of users [51, 52]. Texts of user’s comment along with article’s text also give rise to an informative model for fake news classification as it is highly likely that fake news articles will have fewer positive comments as compared to real news articles [53]. Shu et al proposes a novel method of detection of fake news that uses TriFN which is a tri relationship embedding framework between the users, publishers and news pieces; this auxiliary information improves significantly improves upon the baseline models [54]. Propagation patterns of articles can also be useful features in detection of fake news, as was demonstrated by Monti et al where geometric deep learning was used for creating a model to detect fake news. Heterogenous data like user profile and activity, content, news spreading patterns and structure of social network is fused together underlying by using algorithms that are a generalization of classical convolutional neural networks to graphs [55]. Methodologies discussed so far have been applied on variety of datasets in the past. There are several novel datasets that have been made available solely for the task of fake news detection. These datasets do vary largely with each other as some may solely comprise of articles related to politics while some may be related to any other particular domain. Additionally, datasets also vary on the kind of data present in them as some may contain very short statements while other can have large articles. In the following paragraph, we summarize some of the popular datasets. Dataset that we use is discussed in detail in the later section. LIAR dataset available for detection of fake news has 12.8k labelled short political statements collected over a period of 9 years (from 2007 to 2016) from POLITIFACT.COM. Precisely, labels in this dataset are: true, mostly-true, halftrue, barely-true, false and pants-fire. Number of claims per class are roughly equal in size. Another dataset, FEVER, short for Fact Extraction and Verification has 185,445 claims. These claims were created by extracting data from Wikipedia and then the claims were verified without prior knowledge of the sentences of origin. These claims have been classified into three classes: supported, refuted or notenoughinfo and have also been verified by skilled annotators [56]. As present form of fake news is mostly present on social media, datasets such as BUZZFEEDNEWS contain 2282 samples published using Facebook by 9 news agencies one week before the 2016 US elections. Every post or link is checked and verified by 5 BuzzFeed journalists. Labels in this dataset are: mostly true, mixture, mostly false and no factual content [57]. A similar dataset, Some-like-it-hoax dataset consists of 15,500 Facebook posts and 909,236 users that are classified as either hoax or not hoax [58]. PHEME dataset is a collection of 6425 tweets that are rumours and nonrumours and were posted during the time of some breaking news. 60% of samples are non-rumours, 16% are true rumours, 10% are false and rest are unverified. Most of the contents in the dataset have been verified by journalists and via crowd-sourcing. The CREDBANK dataset is a set of tweets that were traced over a period of around 4 months during 2014-2015. Along with the tweet’s content, it consists of topics classified as events or non events that are annotated with ratings stating their credibility [59]. FAKENEWSNET [60] is yet another popular database of News Content and gives a better understanding of how fake news is present on the social media.
Funding
  • Fine-tuned BERT model could achieve an accuracy of 97.021% on NewsFN data and is able to outperform the other two models by approximately eight percent
  • Misinformation was present before 2016 election as well, as is evident by studies conducted on misinformation before 2016 which shows that misinformation has wide ranging effects that range from financial loss, to politics. One such instance is a 2008 false bankruptcy story about UAL parent company which led to 76% drop in stock price [23]
  • Accuracy on the test set is 97.021% which itself is promising and amongst state-ofthe-art models for classification of fake news
  • The fine-tuned BERT system can achieve an accuracy of 97.021 per cent on NewsFN data and is capable of surpassing the other two models by approximately eight per cent
  • On NewsFN dataset the pre-trained BERT model is able to classify news articles with accuracy of 97.021% which is a significant improvement over other traditional approach
Study subjects and analysis
samples: 2282
These claims have been classified into three classes: supported, refuted or notenoughinfo and have also been verified by skilled annotators [56]. As present form of fake news is mostly present on social media, datasets such as BUZZFEEDNEWS contain 2282 samples published using Facebook by 9 news agencies one week before the 2016 US elections. Every post or link is checked and verified by 5 BuzzFeed journalists

BuzzFeed journalists: 5
As present form of fake news is mostly present on social media, datasets such as BUZZFEEDNEWS contain 2282 samples published using Facebook by 9 news agencies one week before the 2016 US elections. Every post or link is checked and verified by 5 BuzzFeed journalists. Labels in this dataset are: mostly true, mixture, mostly false and no factual content [57]

Facebook posts: 15500
Labels in this dataset are: mostly true, mixture, mostly false and no factual content [57]. A similar dataset, Some-like-it-hoax dataset consists of 15,500 Facebook posts and 909,236 users that are classified as either hoax or not hoax [58]. PHEME dataset is a collection of 6425 tweets that are rumours and nonrumours and were posted during the time of some breaking news. 60% of samples are non-rumours, 16% are true rumours, 10% are false and rest are unverified

tweets: 6425
A similar dataset, Some-like-it-hoax dataset consists of 15,500 Facebook posts and 909,236 users that are classified as either hoax or not hoax [58]. PHEME dataset is a collection of 6425 tweets that are rumours and nonrumours and were posted during the time of some breaking news. 60% of samples are non-rumours, 16% are true rumours, 10% are false and rest are unverified. Most of the contents in the dataset have been verified by journalists and via crowd-sourcing

articles: 3164
This dataset has 6335 items, consisting of the headline and text of the news articles on politics from a wide range of news sources that are classified as either “Fake” or “Real”. Precisely, 3164 articles are labeled as “Fake” and 3171 as “Real”. The ratio of the number of “Fake” articles to that of “Real” articles is roughly 1:1 hence dataset is well balanced with respect to the two classes, and there is no need for oversampling or under sampling

Reference
  • Allcott, H., & Gentzkow, M. (2017). Social media and fake news in the 2016 election. Journal of economic perspectives, 31(2), 211-36.
    Google ScholarLocate open access versionFindings
  • Swift, A. (2016). Americans’ trust in mass media sinks to new low. Gallup News, 14(6). Retrieved from https://news.gallup.com/poll/195542/americans-trust-massmedia-sinks-new-low.aspx.
    Locate open access versionFindings
  • Reilly, R., & Nye, R. (2012). Power, principles and the press. Retrieved from http://www. The open road.com/wpcontent/uploads/2012/09/Powerprinciples-and-thepress-OpenRoad-and-Populus1.pdf.
    Locate open access versionFindings
  • Oshikawa, R., Qian, J., & Wang, W. Y. (2018). A survey on natural language processing for fake news detection. arXiv preprint arXiv:1811.00770.
    Findings
  • Fact-checking U.S. politics. PolitiFact. Retrieved from https://www.politifact.com/.
    Findings
  • Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems (pp. 1097-1105).
    Google ScholarLocate open access versionFindings
  • Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
    Google ScholarLocate open access versionFindings
  • Khan, M. A., Khan, M. A., Ahmed, F., Mittal, M., Goyal, L. M., Hemanth, D. J., & Satapathy, S. C. (2020). Gastrointestinal diseases segmentation and classification based on duo-deep architectures. Pattern Recognition Letters, 131, 193-204.
    Google ScholarLocate open access versionFindings
  • Gautam, R., & Sharma, M. (2020). Prevalence and Diagnosis of Neurological Disorders Using Different Deep Learning Techniques: A Meta-Analysis. Journal of Medical Systems, 44(2), 49.
    Google ScholarLocate open access versionFindings
  • Mittal, M., Goyal, L. M., Kaur, S., Kaur, I., Verma, A., & Hemanth, D. J. (2019). Deep learning based enhanced tumor segmentation approach for MR brain images. Applied Soft Computing, 78, 346-354.
    Google ScholarLocate open access versionFindings
  • Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S.,& Berg, A. C. (2015). Imagenet large scale visual recognition challenge. International journal of computer vision, 115(3), 211-252.
    Google ScholarLocate open access versionFindings
  • Yosinski, J., Clune, J., Bengio, Y., & Lipson, H. (2014). How transferable are features in deep neural networks?. In Advances in neural information processing systems (pp. 3320-3328).
    Google ScholarLocate open access versionFindings
  • Bawa, V. S., & Kumar, V. (2019). Emotional sentiment analysis for a group of people based on transfer learning with a multi-modal system. Neural Computing and Applications, 31(12), 9061-9072.
    Google ScholarLocate open access versionFindings
  • Tan, C., Sun, F., Kong, T., Zhang, W., Yang, C., & Liu, C. (2018). A survey on deep transfer learning. In International conference on artificial neural networks (pp. 270-279). Springer, Cham.
    Google ScholarLocate open access versionFindings
  • Ramírez, I., Cuesta-Infante, A., Pantrigo, J. J., Montemayor, A. S., Moreno, J. L., Alonso, V., & Palombarani, L. (2018). Convolutional neural networks for computer vision-based detection and recognition of dumpsters. Neural Computing and Applications, 1-9.
    Google ScholarLocate open access versionFindings
  • Peters, M. E., Neumann, M., Iyyer, M., Gardner, M., Clark, C., Lee, K., & Zettlemoyer, L. (2018). Deep contextualized word representations. arXiv preprint arXiv:1802.05365.
    Findings
  • Howard, J., & Ruder, S. (2018). Universal language model fine-tuning for text classification. arXiv preprint arXiv:1801.06146.
    Findings
  • Radford, A., Narasimhan, K., Salimans, T., & Sutskever, I. (2018). Improving language understanding by generative pretraining. URL https://s3-us-west-2.amazonaws.com/openaiassets/researchcovers/languageunsupervised/language understanding paper.pdf.
    Findings
  • Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2018). Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805.
    Findings
  • Han, W., Peng, M., Xie, Q., Hu, G., Gao, W., Wang, H.,... & Liu, Z. (2019). DTC: Transfer Learning for Common sense Machine Comprehension. Neurocomputing.
    Google ScholarFindings
  • Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N.,& Polosukhin, I. (2017). Attention is all you need. In Advances in neural information processing systems (pp. 5998-6008).
    Google ScholarLocate open access versionFindings
  • Fake news - Explore - Google Trends. Retrieved from https://trends.google.com/trends/explore?date=2013-12-06 2018-01-06 &geo=US&q=fake news.
    Findings
  • Castillo, C., Mendoza, M., & Poblete, B. (2011). Information credibility on twitter. In Proceedings of the 20th international conference on World wide web (pp. 675-684).
    Google ScholarLocate open access versionFindings
  • Sapir, A. (1987). The LSI course on scientific content analysis (SCAN). Phoenix, ZA: Laboratory for Scientific Interrogation.
    Google ScholarFindings
  • Bogaard, G., Meijer, E. H., Vrij, A., & Merckelbach, H. (2016). Scientific content analysis (SCAN) cannot distinguish between truthful and fabricated accounts of a negative event. Frontiers in psychology, 7, 243.
    Google ScholarLocate open access versionFindings
  • Nahari, G., Vrij, A., & Fisher, R. P. (2012). Does the truth come out in the writing? Scan as a lie detection tool. Law and Human Behavior, 36(1), 68.
    Google ScholarLocate open access versionFindings
  • Fuller, C. M., Biros, D. P., & Wilson, R. L. (2009). Decision support for determining veracity via linguistic-based cues. Decision Support Systems, 46(3), 695-703.
    Google ScholarLocate open access versionFindings
  • Zhou, L., Burgoon, J. K., Nunamaker, J. F., & Twitchell, D. (2004). Automating linguistics-based cues for detecting deception in text-based asynchronous computer-mediated communications. Group decision and negotiation, 13(1), 81106.
    Google ScholarLocate open access versionFindings
  • Buller, D. B., & Burgoon, J. K. (1996). Interpersonal deception theory. Communication theory, 6(3), 203-242.
    Google ScholarLocate open access versionFindings
  • Pennebaker, J. W., Francis, M. E., & Booth, R. J. (2001). Linguistic inquiry and word count: LIWC 2001. Mahway: Lawrence Erlbaum Associates, 71(2001), 2001.
    Google ScholarFindings
  • Ali, M., & Levine, T. (2008). The language of truthful and deceptive denials and confessions. Communication Reports, 21(2), 82-91.
    Google ScholarLocate open access versionFindings
  • Rubin, V. L., Conroy, N., Chen, Y., & Cornwell, S. (2016). Fake news or truth? using satirical cues to detect potentially misleading news. In Proceedings of the second workshop on computational approaches to deception detection (pp. 7-17).
    Google ScholarLocate open access versionFindings
  • Zhao, Z., Resnick, P., & Mei, Q. (2015). Enquiring minds: Early detection of rumours in social media from enquiry posts. In Proceedings of the 24th international conference on world wide web (pp. 1395-1405).
    Google ScholarLocate open access versionFindings
  • Morris, M. R., Counts, S., Roseway, A., Hoff, A., & Schwarz, J. (2012). Tweeting is believing? Understanding microblog credibility perceptions. In Proceedings of the ACM 2012 conference on computer supported cooperative work (pp. 441-450).
    Google ScholarLocate open access versionFindings
  • Chen, Y., Conroy, N. J., & Rubin, V. L. (2015). Misleading online content: recognizing clickbait as" false news". In Proceedings of the 2015 ACM on workshop on multimodal deception detection (pp. 15-19).
    Google ScholarLocate open access versionFindings
  • Mihalcea, R., & Strapparava, C. (2009). The lie detector: Explorations in the automatic recognition of deceptive language. In Proceedings of the ACL-IJCNLP 2009 Conference Short Papers (pp. 309-312). Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Ott, M., Choi, Y., Cardie, C., & Hancock, J. T. (2011). Finding deceptive opinion spam by any stretch of the imagination. In Proceedings of the 49th annual meeting of the association for computational linguistics: Human language technologies-volume 1 (pp. 309-319). Association for Computational Linguistics.
    Google ScholarLocate open access versionFindings
  • Biber, D., Johansson, S., Leech, G., Conrad, S., Finegan, E., & Quirk, R. (1999). Longman grammar of spoken and written English (Vol. 2). London: longman.
    Google ScholarLocate open access versionFindings
  • Rayson, P., Wilson, A., & Leech, G. (2002). Grammatical word class variation within the British National Corpus sampler. In New Frontiers of Corpus Research (pp. 295-306). Brill Rodopi.
    Google ScholarLocate open access versionFindings
  • Johnson, M. (1998). PCFG models of linguistic tree representations. Computational Linguistics, 24(4), 613-632.
    Google ScholarLocate open access versionFindings
  • Wang, W. Y. (2017). " liar, liar pants on fire": A new benchmark dataset for fake news detection. arXiv preprint arXiv:1705.00648.
    Findings
  • Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781.
    Findings
  • Qian, F., Gong, C., Sharma, K., & Liu, Y. (2018). Neural User Response Generator: Fake News Detection with Collective User Intelligence. In IJCAI (pp. 3834-3840).
    Google ScholarLocate open access versionFindings
  • Rashkin, H., Choi, E., Jang, J. Y., Volkova, S., & Choi, Y. (2017). Truth of varying shades: Analyzing language in fake news and political fact-checking. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing (pp. 2931-2937).
    Google ScholarLocate open access versionFindings
  • Volkova, S., Shaffer, K., Jang, J. Y., & Hodas, N. (2017). Separating facts from fiction: Linguistic models to classify suspicious and trusted news posts on twitter. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) (pp. 647-653).
    Google ScholarLocate open access versionFindings
  • Chopra, S., Jain, S., & Sholar, J. M. (2017). Towards automatic identification of fake news: Headline-article stance detection with LSTM attention models. In Stanford CS224d Deep Learning for NLP final project.
    Google ScholarFindings
  • Yang, S., Shu, K., Wang, S., Gu, R., Wu, F., & Liu, H. (2019). Unsupervised fake news detection on social media: A generative approach. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 33, pp. 5644-5651).
    Google ScholarLocate open access versionFindings
  • Castillo, C., Mendoza, M., & Poblete, B. (2011). Information credibility on twitter. In Proceedings of the 20th international conference on World wide web (pp. 675-684).
    Google ScholarLocate open access versionFindings
  • Kwon, S., Cha, M., & Jung, K. (2017). Rumor detection over varying time windows. PloS one, 12(1).
    Google ScholarLocate open access versionFindings
  • Yang, F., Liu, Y., Yu, X., & Yang, M. (2012). Automatic detection of rumour on Sina Weibo. In Proceedings of the ACM SIGKDD Workshop on Mining Data Semantics (pp. 1-7).
    Google ScholarLocate open access versionFindings
  • Ma, J., Gao, W., & Wong, K. F. (2017). Detect rumors in microblog posts using propagation structure via kernel learning. Association for Computational Linguistics.
    Google ScholarFindings
  • Ma, J., Gao, W., & Wong, K. F. (2018). Rumour detection on twitter with tree-structured recursive neural networks. Association for Computational Linguistics.
    Google ScholarFindings
  • Chen, T., Li, X., Yin, H., & Zhang, J. (2018). Call attention to rumours: Deep attention based recurrent neural networks for early rumour detection. In Pacific-Asia Conference on Knowledge Discovery and Data Mining (pp. 40-52). Springer, Cham.
    Google ScholarLocate open access versionFindings
  • Shu, K., Wang, S., & Liu, H. (2019). Beyond news contents: The role of social context for fake news detection. In Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining (pp. 312-320).
    Google ScholarLocate open access versionFindings
  • Monti, F., Frasca, F., Eynard, D., Mannion, D., & Bronstein, M. M. (2019). Fake news detection on social media using geometric deep learning. arXiv preprint arXiv:1902.06673.
    Findings
  • Thorne, J., Vlachos, A., Christodoulopoulos, C., & Mittal, A. (2018). FEVER: a large-scale dataset for fact extraction and verification. arXiv preprint arXiv:1803.05355.
    Findings
  • Silverman, C. (2016). Viral fake election news outperformed real news on Facebook in final months of the us election. Retrieved from https://www.buzzfeednews.com/article/craigsilverman/viralfake-election-news-outperforme-real-news-onFacebook.
    Findings
  • Tacchini, E., Ballarin, G., Della Vedova, M. L., Moret, S., & de Alfaro, L. (2017). Some like it hoax: Automated fake news detection in social networks. arXiv preprint arXiv:1704.07506.
    Findings
  • Mitra, T., & Gilbert, E. (2015). Credbank: A large-scale social media corpus with associated credibility annotations. In Ninth International AAAI Conference on Web and Social Media.
    Google ScholarLocate open access versionFindings
  • Shu, K., Mahudeswaran, D., Wang, S., Lee, D., & Liu, H. (2018). Fakenewsnet: A data repository with news content, social context and dynamic information for studying fake news on social media. arXiv preprint arXiv:1809.01286.
    Findings
  • Zhu, Y., Kiros, R., Zemel, R., Salakhutdinov, R., Urtasun, R., Torralba, A., & Fidler, S. (2015). Aligning books and movies: Towards story-like visual explanations by watching movies and reading books. In Proceedings of the IEEE international conference on computer vision (pp. 19-27).
    Google ScholarLocate open access versionFindings
  • fake_real_news_dataset · GitHub. Retrieved from https://github.com/GeorgeMcIntire/fake_real_news_dataset.
    Findings
  • Wu, Y., Schuster, M., Chen, Z., Le, Q. V., Norouzi, M., Macherey, W.,& Klingner, J. (2016). Google's neural machine translation system: Bridging the gap between human and machine translation. arXiv preprint arXiv:1609.08144.
    Findings
  • Kingma, D. P., & Ba, J. (2014). Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980.
    Findings
  • Hinton, G., Srivastava, N., & Swersky, K. (2012). Neural networks for machine learning lecture 6a overview of minibatch gradient descent. Cited on, 14(8).
    Google ScholarLocate open access versionFindings
  • Robbins, H., & Monro, S. (1951). A stochastic approximation method. The annals of mathematical statistics, 400-407.
    Google ScholarFindings
  • Chen, T., & Guestrin, C. (2016). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining (pp. 785-794).
    Google ScholarLocate open access versionFindings
Author
Akshay Aggarwal
Akshay Aggarwal
Aniruddha Chauhan
Aniruddha Chauhan
Deepika Kumar
Deepika Kumar
Mamta Mittal
Mamta Mittal
Sharad Verma
Sharad Verma
Your rating :
0

 

Tags
Comments
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn
小科