KR102340542B1 - Device and method for automatic fake news detection - Google Patents

Abstract

A fake news classification device is disclosed. The fake news classification apparatus includes a pre-training unit for pre-training a BERT (Bidirectional Encoder Representations from Transformers model) model using news composed of at least a headline and a body text, and predetermined data. Determining whether or not news to be classified is fake news using a fine-tuning unit that fine-tunes the BERT model using a dataset, and a detection model generated by the pre-learning unit and the fine-tuning unit It includes a classification section that

Description

Fake news detection device and method based on automation

The present invention relates to a method for automatic detection of fake news.

Fake information is expressed in various forms such as videos, audio, images, and text. In addition, fake information in text form may be classified into news, social network services, speeches, documents, and the like. The present invention proposes a model for a method of detecting fake news by focusing on text-based fake news. Fraudulent or false information spreads rapidly and becomes a problem if readers fail to detect that the information is fake news.

In 2015, the International Fact-Checking Network (IFCN) was established. IFCN observes fact check trends and provides training programs for fake checkers. In addition, various efforts are being made to prevent the spread of fake news. Politifact (https://www.politifact.com) and snopes (https://www.snopes.com) have developed fake news detection tools. However, these tools have a problem in that time and money are consumed in a way that a person directly intervenes and makes a judgment. Therefore, a model for automatically detecting fake news is required.

1. Pham, L. Transferring, Transforming, Ensembling: The Novel Formula of Identifying Fake News. In Proceedings of the 12th ACM International Conference on Web Search and Data Mining, Melbourne, Australia, 11-15 February 2019. 2. Liu, S.; Liu, S.; Ren, L. Trust or Suspect? An Empirical Ensemble Framework for Fake News Classification. In Proceedings of the 12th ACM International Conference on Web Search and Data Mining, Melbourne, Australia, 11-15 February 2019. 3. Yang, K. C.; Niven, T.; Kao, H.Y. Fake News Detection as Natural Language Inference. In Proceedings of the 12th ACM International Conference on Web Search and Data Mining, Melbourne, Australia, 11-15 February 2019. 4. Omidvar, A.; Jiang, H.; An, A. Using Neural Network for Identifying Clickbaits in Online News Media. In Annual International Symposium on Information Management and Big Data, Lima, Peru, 3-5 September 2018; Springer: Cham, Switzerland, 2018; pp. 220-232. 5. Zhou, Y. Clickbait detection in tweets using self-attentive network. arXiv 2017, arXiv:1710.05364. 6. Grigorev, A. Identifying clickbait posts on social media with an ensemble of linear models. arXiv 2017, arXiv:1710.00399. 7. Mikolov, T.; Sutskever, I.; Chen, K.; Corrado, G. S.; Dean, J. Distributed representations of words and phrases and their compositionality. In Proceedings of the 26th International Conference on Neural Information Processing Systems, Lake Tahoe, NV, USA, 5-10 December 2013; pp. 3111-3119. 8. Bojanowski, P.; Grave, E.; Joulin, A.; Mikolov, T. Enriching word vectors with subword information. Trans. Assoc. Compute. Linguist. 2017, 5, 135-146. 9. Peters, M. E.; Neumann, M.; Iyyer, M.; Gardner, M.; Clark, C.; Lee, K.; Zettlemoyer, L. Deep contextualized word representations. arXiv 2018, arXiv:1802.05365. 10. Devlin, J.; Chang, M. W.; Lee, K.; Toutanova, K. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv 2018, arXiv:1810.04805. 11. Radford, A.; Narasimhan, K.; Salimans, T.; Sutskever, I. Improving Language Understanding by Generative Pre-Training. 2018. Available online: https://s3-us-west-2.amazonaws.com/openai-assets/researchcovers/languageunsupervised/languageunderstandingpaper.pdf (accessed on 20 September 2019). 12. Schuster, M.; Paliwal, K. K. Bidirectional recurrent neural networks. IEEE Trans. Signal Process. 1997, 45, 2673-2681. 13. Wang, W.Y. “Liar, liar pants on fire”: A new benchmark dataset for fake news detection. arXiv 2017, arXiv:1705.00648. 14. Ruchansky, N.; Seo, S.; Liu, Y. Csi: A hybrid deep model for fake news detection. In Proceedings of the 2017 ACM on Conference on Information and Knowledge Management, Singapore, 6-10 November 2017; ACM: New York, NY, USA, 2017; pp. 797-806. 15. Kochkina, E.; Liakata, M.; Augenstein, I. Turing at semeval-2017 task 8: Sequential approach to rumour stance classification with branch-lstm. arXiv 2017, arXiv:1704.07221. 16. Popat, K.; Mukherjee, S.; Yates, A.; Weikum, G. DeClarE: Debunking fake news and false claims using evidence-aware deep learning. arXiv 2018, arXiv:1809.06416. 17. Yang, Y.; Zheng, L.; Zhang, J.; Cui, Q.; Li, Z.; Yu, P.S. TI-CNN: Convolutional neural networks for fake news detection. arXiv 2018, arXiv:1806.00749. 18. Rasool, T.; Butt, W. H.; Shaukat, A.; Akram, M. U. Multi-Label Fake News Detection using Multi-layered Supervised Learning. In Proceedings of the 2019 11th International Conference on Computer and Automation Engineering, Perth, Australia, 23-25 February 2019; ACM: New York, NY, USA, 2019; pp. 73-77. 19. Yang, S.; Shu, K.; Wang, S.; Gu, R.; Wu, F.; Liu, H. Unsupervised fake news detection on social media: A generative approach. In Proceedings of the 33rd AAAI Conference on Artificial Intelligence, Honolulu, HI, USA, 27 January-1 February 2019. 20. Feng, S.; Banerjee, R.; Choi, Y. Syntactic stylometry for deception detection. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics: Short Papers- Volume 2, Jeju Island, Korea, 8-14 July 2012; pp. 171-175. 21. Sean, B.; Doug, S.; Yuxi, P. Talos Targets Disinformation with Fake News Challenge Victory. 2017. Available online: https://blog.talosintelligence.com/2017/06/talos-fake-news-challenge.html (accessed on 20 September 2019). 22. De Lathauwer, L.; De Moor, B.; Vandewalle, J. A multilinear singular value decomposition. SIAM J. Matrix Anal. Appl. 2000, 21, 1253-1278. 23. Ramos, J. Using tf-idf to determine word relevance in document queries. In Proceedings of the First Instructional Conference on Machine Learning, Piscataway, NJ, USA, 3-8 December 2003; Volume 242, pp. 133-142. 24. Davis, R.; Proctor, C. Fake news, real consequences: Recruiting neural networks for the fight against fake news. 2017. Available online: https://web.stanford.edu/class/archive/cs/cs224n/cs224n.1174/reports/2761239.pdf (accessed on 20 September 2019). 25. Hanselowski, A.; PVS, A.; Schiller, B.; Caspelherr, F.; Chaudhuri, D.; Meyer, C. M.; Gurevych, I. A retrospective analysis of the fake news challenge stance detection task. arXiv 2018, arXiv:1806.05180. 26. Hochreiter, S.; Schmidhuber, J. Long short-term memory. Neural Compute. 1997, 9, 1735-1780. 27. Hermans, M.; Schrauwen, B. Training and Analysing Deep Recurrent Neural Networks. In Advances in Neural Information Processing Systems 26; Burges, C. J. C., Bottou, L., Welling, M., Ghahramani, Z., Weinberger, K. Q., Eds.; Curran Associates, Inc.: New York, NY, USA, 2013; pp. 190-198. 28. Riedel, B.; Augenstein, I.; Spithourakis, G. P.; Riedel, S. A simple but tough-to-beat baseline for the Fake News Challenge stance detection task. arXiv 2017, arXiv:1707.03264. 29. Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A. N.; Kaiser, Ł.; Polosukhin, I. Attention is all you need. In Proceedings of the 31st Conference on Neural Information Processing System, Long Beach, CA, USA, 4-9 December 2017; pp. 5998-6008. 30. Diaz, M.; Ferrer, M.A.; Impedovo, D.; Pirlo, G.; Vessio, G. Dynamically enhanced static handwriting representation for Parkinson's disease detection. Pattern Recognit. Lett. 2019, 128, 204-210. 31. Lee, J.; Yoon, W.; Kim, S.; Kim, D.; Kim, S.; So, C. H.; Kang, J. Biobert: Pre-trained biomedical language representation model for biomedical text mining. arXiv 2019, arXiv:1901.08746. 32. Liu, Y.; Lapata, M. Text Summarization with Pretrained Encoders. arXiv 2019, arXiv:1908.08345. 33. See, A.; Liu, P. J.; Manning, C.D. Get to the point: Summarization with pointer-generator networks. arXiv 2017, arXiv:1704.04368. 34. Liu, Y. Fine-tune BERT for Extractive Summarization. arXiv 2019, arXiv:1903.10318. 35. Liu, L.; Lu, Y.; Yang, M.; Qu, Q.; Zhu, J.; Li, H. Generative adversarial network for abstractive text summarization. In Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence, New Orleans, LA, USA, 2-7 February 2018. 36. Paulus, R.; Xiong, C.; Socher, R. A deep reinforced model for abstractive summarization. arXiv 2017, arXiv:1705.04304. 37. Hermann, K. M.; Kocisky, T.; Grefenstette, E.; Espeholt, L.; Kay, W.; Suleyman, M.; Blunsom, P. Teaching Machines to Read and Comprehend. In Advances in Neural Information Processing Systems 28; Cortes, C., Lawrence, N.D., Lee, D.D., Sugiyama, M., Garnett, R., Eds.; Curran Associates, Inc.: New York, NY, USA, 2015; pp. 1693-1701. 38. Yang, K.; Lee, D.; Whang, T.; Lee, S.; Lim, H. EmotionX-KU: BERT-Max based Contextual Emotion Classifier. arXiv 2019, arXiv:1906.11565. 39. Aurelio, Y. S.; de Almeida, G. M.; de Castro, C. L.; Braga, A. P. Learning from imbalanced data sets with weighted cross-entropy function. Neural Process. Lett. 2019, 1-13, doi:10.1007/s11063-018-09977-1. 40. Sudre, C. H.; Li, W.; Vercauteren, T.; Ourselin, S.; Cardoso, M. J. Generalized Dice Overlap as a Deep Learning Loss Function for Highly Unbalanced Segmentations. In Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support; Springer: New York, NY, USA, 2017; pp. 240-248. 41. Neyshabur, B.; Bhojanapalli, S.; McAllester, D.; Srebro, N. Exploring generalization in deep learning. In Proceedings of the 31st International Conference on Neural Information Processing Systems, Long Beach, CA, USA, 4-9 December 2017; pp. 5947-5956. 42. Zhang, Z.; Sabuncu, M. Generalized cross entropy loss for training deep neural networks with noisy labels. In Proceedings of the 32nd International Conference on Neural Information Processing Systems, Montreal, QC, Canada, 3-8 December 2018; pp. 8778-8788. 43. Shang, W.; Huang, H.; Zhu, H.; Lin, Y.; Qu, Y.; Wang, Z. A novel feature selection algorithm for text categorization. Expert Syst. Appl. 2007, 33, 1-5.

An object of the present invention is to provide an apparatus and method for detecting fake news based on automation.

The fake news classification apparatus according to an embodiment of the present invention pre-trains a BERT (Bidirectional Encoder Representations from Transformers model) model using news composed of at least a headline and a body text. Using a pre-learning unit, a fine-tuning unit for fine-tuning the BERT model by additionally learning a predetermined news dataset, and a detection model generated by the pre-training unit and the fine-tuning unit and a classification unit that determines whether the news to be classified is fake news.

In the case of an automated fake news detection apparatus and method according to an embodiment of the present invention, it is possible to detect whether the corresponding news is fake news based on the title and main text of the news.

In order to more fully understand the drawings recited in the Detailed Description, a detailed description of each drawing is provided.
1 is a functional block diagram of an apparatus for detecting fake news according to an embodiment of the present invention.
FIG. 2 shows a fake news detection model generated by the fake news detection apparatus shown in FIG. 1 .

Specific structural or functional descriptions of the embodiments according to the concept of the present invention disclosed in this specification are only exemplified for the purpose of explaining the embodiments according to the concept of the present invention, and the embodiments according to the concept of the present invention are It may be implemented in various forms and is not limited to the embodiments described herein.

Since the embodiments according to the concept of the present invention may have various changes and may have various forms, the embodiments will be illustrated in the drawings and described in detail herein. However, this is not intended to limit the embodiments according to the concept of the present invention to specific disclosed forms, and includes all modifications, equivalents, or substitutes included in the spirit and scope of the present invention.

Terms such as first or second may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another, for example without departing from the scope of the inventive concept, a first component may be termed a second component and similarly a second component A component may also be referred to as a first component.

When a component is referred to as being “connected” or “connected” to another component, it is understood that it may be directly connected or connected to the other component, but other components may exist in between. it should be On the other hand, when it is said that a certain element is "directly connected" or "directly connected" to another element, it should be understood that the other element does not exist in the middle. Other expressions describing the relationship between components, such as "between" and "immediately between" or "neighboring to" and "directly adjacent to", etc., should be interpreted similarly.

The terms used herein are used only to describe specific embodiments, and are not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly dictates otherwise. In this specification, terms such as "comprise" or "have" are intended to designate that a feature, number, step, operation, component, part, or combination thereof described herein exists, but one or more other features It should be understood that it does not preclude the possibility of the presence or addition of numbers, steps, operations, components, parts, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present specification. does not

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the scope of the patent application is not limited or limited by these examples. Like reference numerals in each figure indicate like elements.

1 is a functional block diagram of a fake news detection apparatus according to an embodiment of the present invention, and FIG. 2 illustrates a fake news detection model generated by the fake news detection apparatus shown in FIG. 1 .

The fake news detection device 10 (hereinafter referred to as a detection device) includes a dictionary learning unit 110 , a fine adjustment unit 120 , and a storage unit 150 . According to an embodiment, the detection apparatus 10 may further include a test unit 130 and/or a classification unit 140 . The detection device 10 receives news composed of a title or headline and a body text as input and determines whether the input news is fake news, that is, a fake news detection model (hereinafter, detection model). ) and classify the news to be classified using the generated detection model. For example, the detection apparatus 10 may generate a detection model by learning a BERT (Bidirectional Encoder Representations from Transformers model) model.

The detection device 10 may be implemented using at least one processor, where the processor is a central processing unit (CPU), a micro controller unit (MCU), and an application processor (AP). Processor), a microprocessor (Micom), an electronic control unit (ECU), and/or other electronic devices capable of processing various calculations and generating control signals, and the like. These devices may be implemented using, for example, one or more semiconductor chips and related components. According to an embodiment, the processor drives at least one application (which can be expressed as software, a program, or an app) stored in the storage unit 140 to perform a predefined operation, determination, processing and/or control operation, etc. can also be done Here, the application stored in the storage unit 150 may be directly written by a designer and input and stored in the storage unit 150, or acquired or updated through an electronic software distribution network accessible through a wired or wireless communication network. it may be

In addition, the detection device 10 may be implemented using at least one information processing device in which one or more such processors are installed, and the electronic information processing device is a desktop computer, a laptop computer, a server computer, a smart phone, a tablet PC, and a smart phone. Clocks, navigation devices, portable game machines, Head Mounted Display (HMD) devices, artificial intelligence sound reproduction devices, digital televisions, home appliances, mechanical devices and/or electronically calculate/process information and control related thereto and at least one device specially designed for energy management.

The pre-training unit 110 (which may be referred to as a pre-training unit) may pre-train the detection model, that is, the BERT model, using the first data. As a result of the pre-learning by the dictionary learning unit 110 , it is possible to improve the context comprehension of the BERT model and improve the natural language processing (NLP). Specifically, the BERT model performs pre-training using an unsupervised prediction task including a masked language model (MLM) and a next sentence predictor. MLM first understands the context and predicts words. To this end, the dictionary learning unit 110 may randomly mask some tokens with a certain probability (eg, 15%) from words or sentences input to the BERT. The input is included in a Transformer structure to predict the masked word based on the context of the surrounding words. Through this process, the BERT model can understand the context more accurately. The next sentence predictor is for identifying relationships between sentences. These tasks are important for language understanding tasks such as QA (Question Answering) or NLI (Natural Language Inference). BERT involves a binarized next sentence prediction task that combines two sentences in a corpus with the original sentence. Such a model can improve the performance of BERT in NLP tasks. The data used in the BERT model includes 800 M words from Book Corpus and 2500 M words from Wikipedia. The above data may be referred to as first data.

According to an embodiment, the pre-training unit 110 may pre-train the BERT model using the second data. For example, the second data may include a dataset (https://github.com/abisee/cnn-dailymail) of CNN (www.cnn.com) and Daily Mail (www.dailymail.co.uk). The detection performance of the BERT model can be improved through additional pre-training using the second data. The second data may mean news composed of a title and a body. In previous natural language processing (NLP) tasks, BERT's pre-training performed well. However, the data used by the BERT model is based on 2500 M words of general data obtained from Wikipedia and 800 M words from Book Corpus. Such data includes information from a vast field, but detailed information in an individual domain is insufficient. In view of this problem, in the present invention, new data is added in the pre-training stage in order to improve the ability to detect fake news. The summary data from CNN (see Non-Patent Documents 33-36) contained 90,000 documents and 380,000 queries (including 118,497 vocabularies), while the Daily Mail dataset contained 197,000 documents and 879,000 queries ( 208,045 vocabularies included). CNN documents are documents for the period from April 2007 to the end of April 2015 obtained from the CNN website, and Daily Mail documents are documents obtained from the Daily Mail website from June 2010 to the end of April 2015 is the document of

The fine tuning unit 120 (which may be referred to as a fine tuning unit) may generate a detection model by fine-tuning the BERT model. Fake news challenge stage 1 (FNC-1) data may be used in the fine-tuning stage. Here, the training data is composed of a pair of a title (headline) and a body (body text), and may include a class label of each title-body pair. Here, the classification label refers to a classification of data composed of a title-body pair, and may include first to fourth classifications. That is, the training data or the data composed of the title-body pair may include classification information of the data composed of the title-body pair. For example, the first to fourth classifications may be agree, disagree, discuss, or unrelated. 'Consistent' means that the title and the content of the text match, 'non-consistent' means that the title and text do not match, and 'discuss' means that the title and the content of the text need to be discussed. and 'irrelevant' may mean that the title and the content of the main text are not related. Here, news classified as 'non-match' and 'unrelated' may mean fake news.

In the BERT model, Weighted Cross Entropy (WCE) was used to classify news into four groups. Regarding WCE, reference can be made to Non-Patent Documents 38 to 40.

The test unit 130 may test the learned detection model using the test data. Here, the test data consists of title and body pairs, and does not include classification labels to evaluate the detection model. A total of 2587 headlines and 2587 body texts were used as training data and test data, and the data can be acquired from FNC-1 github (htttps://github.com/FakeNewsChallenge/fnc-1). have.

The classification unit 140 may classify the news to be classified by receiving the news to be classified including the title and the body, and inputting the news to be classified into the generated classification model. That is, the classification unit 140 may determine whether the news to be classified is fake news. For example, when the news to be classified is classified as 'disagree' or 'unrelated', the classification unit 140 may determine the news to be classified as fake news.

The storage unit 140 may store data used to generate the detection model, data generated during generation of the detection model, data for testing the detection model, the generated detection model, classification target news, classification results, and the like. have.

The device described above may be implemented as a hardware component, a software component, and/or a set of hardware components and software components. For example, the devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), and a PLU. It may be implemented using one or more general purpose or special purpose computers, such as a Programmable Logic Unit (Programmable Logic Unit), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications executed on the operating system. A processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For convenience of understanding, although one processing device is sometimes described as being used, one of ordinary skill in the art will recognize that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that can include For example, the processing device may include a plurality of processors or one processor and one controller. Other Processing Configurations are also possible, such as a Parallel Processor.

The software may include a computer program, code, instructions, or a combination of one or more thereof, and configure the processing device to operate as desired or independently or collectively processed You can command the device. The software and/or data may be any type of machine, component, physical device, virtual equipment, computer storage medium or device, to be interpreted by or provide instructions or data to the processing device. , or may be permanently or temporarily embodied in a transmitted signal wave (Signal Wave). The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored in one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or preferably. The program instructions recorded in the medium may be specially designed and configured for the embodiment, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magnetic such as floppy disks. - Includes hardware devices specially configured to store and execute program instructions, such as Magneto-optical Media, ROM, RAM, Flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

Although the present invention has been described with reference to the embodiment shown in the drawings, which is merely exemplary, those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom. For example, the described techniques are performed in an order different from the described method, and/or the described components of the system, structure, apparatus, circuit, etc. are combined or combined in a different form than the described method, or other components Or substituted or substituted by equivalents may achieve an appropriate result. Accordingly, the true technical protection scope of the present invention should be determined by the technical spirit of the appended claims.

10: Fake News Classification Device
110: pre-learning unit
120: fine adjustment unit
130: test unit
140: classification unit
150: storage

Claims (4)

Pre-training a BERT (Bidirectional Encoder Representations from Transformers model) model using first data including a plurality of words and second data including news consisting of a headline and body text ) pre-learning unit;
a fine tuning unit for fine-tuning the BERT model using training data, each of which includes news including a title, a body, and classification information;
a classification unit that determines whether the news to be classified is fake news by using the detection model generated by the pre-learning unit and the fine adjustment unit; and
Comprising a test unit for testing the detection model using the test data,
The classification information is information on the classification of the corresponding news among the news included in the learning data,
The classification information is any one of a first classification, a second classification, a third classification, and a fourth classification,
The first classification is a classification for news in which the title and the content of the main text match,
The second classification is a classification for news that the title and the content of the main text do not match,
The third classification is a classification for news that is judged to need discussion on the title and content of the main text,
The fourth classification is a classification for news that is unrelated to the title and content of the main text,
Fake news classification device.
delete delete According to claim 1,
The classified news consists of a title and a body,
The classification unit determines the classification target news as fake news when the classification target news is classified into a second classification or a fourth classification;
Fake news classification device.

Images