|
Authors (year) | Aim | Issues | Methods | Datasets | Findings | Shortcomings/future work |
|
Sansonetti et al. [38] | Detecting fake news and unreliable users | Figuring out whether or not a piece of social media news is trustworthy by looking for trustworthy sources | Deep learning techniques, CNN, LSTM, and neural network classifier | Twitter (user-profiles and shared news) | Recognizing reliable profile. Check for distributed material and social information’s trustworthiness. The average accuracy of 90%. | Can be used of new datasets of news and users with further reliability prediction features. |
Antoun et al. [39] | To address the risk for fake news spreading | Detecting fake news, identifying domains, and identifying bots in tweets | Bi-LSTM and voting classifier | QICC contest dataset | Superior performances and high impact features | Enhancing the detection of fake news by using elements from fact-checking sites in addition to Google searches. |
Lin et al. [40] | To properly determine the position of news | Only one inference direction was utilized for categorizing the stance, which may have resulted in some crucial info being lost | BERT language model | Fake news challenge stage 1(FNC-1) | Detect the stance more accurately | The misprediction example of the proposed model is the stance relation of claim and article is “disagree” class. |
Konkobo et al. [41] | The massive quantity of unlabeled data on social media must be dealt with | Unwilling to cope with fake news’ massive quantity of unlabeled data | CredRank algorithm, CNN | Politifact and GossipCop | Politifact’s precision is 71.10 percent, whereas Gossipcop’s accuracy is 68.07 percent | There will be improvements to a greater number of prototypes that have been suggested. We will create a multilingual dataset and look at how different languages affect news categorization. |
Huang and Chen [42] | So that fake news may be detected with greater precision | Cross-domain intractability issue, the required information is often unavailable or inadequate at the early stage | LSTM, depth LSTM, LIWC CNN, and N-gram CNN | FOR dataset, Buzzfeed corpus, SFL dataset, FND dataset satire, and political dataset | Increase the precision, optimize the weights, and look at the intractability problem cross domains | The grammatical analysis would be investigated in-depth, and the preprocessing will extract more useful info, resulting in improved precision in detecting fake news. |
Yanagi et al. [43] | So that others may provide remarks and assist classify documents to assess | Early detection of fake news | Neural network model | FakeNewsNet dataset | Achieved the best recall score | Creating remarks may assist fact-checkers in determining if something is genuine or not. |
Paixão et al. [44] | To the identification of news depending on several kinds of characteristics | Automatic detection of false news | Supervised and unsupervised learning | Fake.Br corpus | Acquired F1 scores up to 96% | Topic modeling. |
Song et al. [45] | To recognize characteristics of fake news | Maintaining the modality-specific characteristics has an impact on the model’s performances | Multichannel convolutional neural networks with residual cross-modal attention | Four real-world datasets | Learns more discriminable feature representations | — |
Ren and Zhang [46] | To learn the node description in HIN | Intentional rumors may conceal a writing style that has not been well modeled and used to match the additional and distracting multimodal info | Hierarchical graph attention network | Two real-world fake news datasets | Expandability and generalizability | Other node classification related applications. |
Ying et al. [47] | To make use of the relationship between segments at the same time | Falsification of multimedia data is a problem that is hardly handled. | Multimodal topic memory network (MTMN) | WEIBO and PHEME | Best performance | Discover innovative ways to utilize deep neural networks’ prior knowledge for detecting fake news and other valuable complementary info. |
Meel and Vishwakarma [48] | To create and falsify news stories that include both text and graphic elements to check error level analysis | As a result of its capability to have catastrophic effects by focusing on a particular kind of news, social networking became a significant issue. | Hierarchical attention network, bidirectional GRU, and ensemble learning | All Data1, fake news Sample2, and fake news detection3 | The dataset’s best precision was 95.90 percent while using fake news samples | Fake news identification utilizing multimodal data is still a challenging and unknown area that needs further study. |
Khan et al. [49] | The effectiveness of several machine learning techniques on three distinct datasets will be evaluated | Maintaining the modality-specific characteristics has an impact on the model’s performances | Pretrained language models, deep learning, and machine learning approaches | Liar4, fake or real news dataset5, and corpus | With limited datasets, pretrained algorithms like BERT and others do the greatest detecting fake news | On social media during the COVID-19 epidemic, could identify misleading and health-related fake news? |
Shim et al. [50] | To avoid the distribution of fake news | Only propagation inside a single social media may be tracked when looking at the distributors’ network. | link2vec | The fake news dataset in English and the dataset in Korean | A new source of background details for identifying false news that is successful in identifying short-form fake news | Further study is needed to determine how many top connections are ideal for maximizing the effectiveness of a fake news detecting system. |
Samadi et al. [51] | To dive into comparison research regarding utilizing various classifiers and embedding frameworks | It is not clear which contextualized embedding will provide the classifier with the most useful characteristics. | CNN, single-layer perceptron (SLP), multilayer perceptron (MLP), BERT, RoBERTa, GPT2, and funnel transformer | LIAR, ISOT, and COVID-19 | Achieve promising results without using additional features and network information increase the capacity of learning create dense contextualized embeddings for input texts | Has the ability to mine client profiles for additional information. Fake news may be detected using a variety of textual analyses in NLP. |
Mhatre and Masurkar [52] | To determine the truthfulness of the news by scraping it from the internet | Increase the precision of categorization | Methods for NLP, like naive Bayes, KNN, decision tree, logistic regression, and the passive-aggressive classifier, or the SVM ensemble learning methodology | Web-scrapped data | Fake news identification accuracy has increased; thus, the results may now be classified as trustworthy or untrustworthy | — |
|