Detecting Fake News Using Machine Learning : A Systematic Literature Review

• Machine-learning algorithms are trained on datasets and can then be used to detect fake news automatically; available families include supervised, unsupervised, and reinforcement learning.The introduction notes prior work using machine learning for detection (Wang, 2017) and reports that fake news is increasing over time.Source: §1 · p. 2
• Fake news spreads through online platforms, can damage people and organizations, and may affect public opinions and decisions, as in the 2016 US election.Early detection is difficult (Shu et al., 2017), and unchecked news can spread further as readers believe and share it.Source: §1 · p. 2
• The review examines why machine learning matters for fake-news detection, how it can be applied, and which algorithms researchers use.It frames these topics as research questions addressed by the literature review.Source: §1 · p. 2

2. Methodology

• The authors use a systematic literature review to answer their research questions, drawing on papers collected from various databases.They discuss and cite research papers in the review.Source: §2 · p. 2

Exclusion and Inclusion

• Papers were included only if they fulfilled the review’s stated inclusion criteria.The criteria are presented in Table 1; their specific contents are not provided in the supplied passages.Source: Exclusion and Inclusion · p. 3, Table 1 · p. 3

Quality Assessment

• Included papers were assessed as good quality when they discussed machine learning for fake or false news detection.The assessment was based on the research work presented in each paper.Source: Quality Assessment · p. 3

3. Research Question

• RQ1 asks why machine learning is required to detect fake news.Source: §3 · p. 3

RQ2: Which machine learning supervised classifiers can be used for detecting fake news?

• RQ2 asks which supervised machine-learning classifiers can be used to detect fake news.Source: RQ2: Which machine learning supervised… · p. 3
• RQ3 asks how machine-learning classifiers are trained to detect fake news.Source: RQ2: Which machine learning supervised… · p. 3
• The review says it will answer the research questions in its results and discussion section.Source: RQ2: Which machine learning supervised… · p. 3

4. Search Process

• The search process used Clarivate Analytics (WoS), IEEE Xplore, and Elsevier (Scopus), followed by keyword search and staged exclusion through primary-study selection.Figure 1 presents the search process model.Source: §4 · p. 4, Figure 1 · p. 4
• 26 of the 73 papers collected against the search keyword remained after title-and-abstract screening and further review against inclusion and exclusion criteria.The review first excluded papers based on titles and abstracts, then assessed the further text of the papers against the criteria.Source: §4 · p. 4

5. Result and Discussion

• Social platforms enable easy news sharing but do not verify publishers, allowing fake news to spread and potentially harm people, organizations, society, or political parties.The review notes that people may believe and reshare false stories, which can affect reputations and opinions.Source: §5 · p. 4

RQ1: Why machine learning is required to detect the fake news?

• Because people may believe and share false stories without knowing the complete story, the review presents machine learning as a way to check post contents automatically.The passages describe fake news as potentially harming individuals, organizations, businesses, society, and political parties.Source: RQ1: Why machine learning is required… · p. 5
• Logistic regression was reported as a better classifier for accuracy, while Naïve Bayes reached 74% accuracy on Facebook posts.Granik and Mesyura said the accuracy could be improved.Source: RQ1: Why machine learning is required… · pp. 5–6
• Naïve Bayes achieved 96.08% accuracy versus 90.90% for neural network and SVM; another study rated SVM and Naïve Bayes best by accuracy.The review reports these as findings from individual studies, not as an overall benchmark comparison.Source: RQ1: Why machine learning is required… · p. 5
• Linear support vector machine provided good accuracy in detecting false news; separately, Singh et al. (2017) reported that support vector machine gave the best results in detecting fake news.The passage lists K-Nearest Neighbor, Support Vector Machine, Logistic Regression, Linear Support Vector Machine, Decision tree, and Stochastic Gradient Descent as classifiers used by different researchers.Source: RQ1: Why machine learning is required… · p. 6

RQ3: How machine learning classifiers are trained for detecting fake news?

• Classifier accuracy depends on proper training; researchers used TF-IDF, N-Gram, and Bag of Words features, and removed unnecessary words and normalized word forms during preprocessing.The review also identifies imbalanced training data as a common problem and says trained classifiers are then used in experiments.Source: RQ3: How machine learning classifiers… · p. 7

• Unrestricted posting on internet and social media platforms enables fake news that can harm individuals or businesses and change opinions about political parties.Source: §6 · p. 7
• Machine-learning classifiers can be trained on a dataset and then used to detect fake news automatically.Source: §6 · p. 7
• The review discusses supervised classifiers, which require labeled training data that is not easily available, and suggests future research on unsupervised detection.Source: §6 · p. 7