Fake Reviews Detection: A Survey

With the widespread of fake online reviews, the detection of fake reviews has become a hot research issue. Despite the efforts of existing studies on fake review detection, the issues of imbalanced data and feature pruning still lack sufficient attention. To address these gaps, the present study proposes an ensemble model for the detection of fake online reviews. The model consists of four steps, and the first three steps are proposed to optimize the base classifiers: (i) Data resampling: We propose a novel way to address the data imbalance problem by combining the resampling and the grid search technique. (ii) Feature pruning: We propose an ablation study to drop unimportant features. (iii) Parameters optimization: We apply the grid search algorithm to determine suitable values of the relevant parameters for each base classifier. (iv) Classifier ensembling: We apply majority voting and stacking strategies to integrate the optimized base classifiers. The proposed data resampling method is also applied for the meta-classifier in the stacking ensemble model. This study produces advances in terms of combining different methods or algorithms into a model and the results show that the proposed ensemble model outperforms some existing techniques, thereby providing a new way to solve the data imbalance and feature pruning issues in the field of fake review detection.

Online reviews significantly influence consumer purchase decisions and business reputations on digital platforms. However, the widespread presence of fake reviews not only misguides consumers but also undermines the reliability of online reviews. Therefore, detecting fake reviews is crucial for various stakeholders and has evolved from simple visual inspection to the use of advanced machine learning models. Despite the proven effectiveness of machine learning models in fake review detection, several key research gaps remain.First, most predictive models rely on binary classification (i.e., true vs. fake), which assumes that review authenticity is a clear-cut distinction between real and fake. However, online reviews often lie on a spectrum of deception, where subtle manipulation, exaggeration, and subjective bias blur the boundary between authenticity and fabrication. Second, prior research lacks a clearly structured framework that accounts for behavioural features in predicting deceptive reviews.To address these gaps, this study develops a probabilistic modelling framework that: 1) estimates the likelihood of deception on a continuous spectrum, allowing for more flexible and accurate detection beyond a rigid true/false label; 2) systematically organizes behavioural features into three dimensions – structural (e.g., number of friends), relational (e.g., number of compliments) and activity-based (e.g., review frequency), based on Social Capital Theory; 3) incorporates store-level attributes, such as average rating, number of reviews, and consumer-generated “vibes”.This study uses a dataset of 32,964 restaurant reviews from Yelp, covering both real and fake reviews from January 2005 to November 2024. A range of probabilistic machine models – including Logistic Regression, Decision Tree, Random Forest, XGBoost, and Artificial Neural Networks – are implemented and compared using 5-fold cross-validation. XGBoost is selected as the final model based on overall superior performance across Precision, Recall, F1-Score, Accuracy, and Area Under the ROC Curve (AUC).Results show that relational and structural behavioural features are among the top predictors. Interestingly, reviewers high in social capital features – such as elite reviewers (designated by Yelp as “elite” based on their activity, consistency, and perceived credibility) – are more likely to generate fake reviews. The average probability of fake review among elite reviewers is 33.5%, compared to 12% among non-elite reviewers. At the store level, reviews associated with high-rated businesses tend to exhibit higher probabilities of being fake. Furthermore, the distribution of fakeness probability across star ratings follows a U-shaped pattern, with reviews at 1-star and 5-star exhibiting higher probabilities of deception, and a secondary peak observed at 4-star.This study contributes to the literature by developing a systematic and behaviourally grounded probabilistic model to predict fake online reviews. Practically, detecting the probabilities of fake review enhances review transparency for consumers, protects brand reputation for businesses, and support fair review policies for regulators.

Nowadays, O2O commercial platforms are playing a crucial role in our daily purchases. However, some people are trying to manipulate the online market maliciously by opinion spamming, a kind of web fraud behavior like writing fake reviews, due to fame and profits, which will harm online purchasing environment and should be detected and eliminated. Moreover, manual fake reviewers are more deceptive compared with old web spambots. Although several efficient methods were proposed in the fake review detection field, the manual fake reviewers are also evolving rapidly. They imitate to be benign users to control the velocity of review fraud actions, and deceive the detection system. Our investigation presented that geolocation factor is potential and can well reflect the distinctions between fake reviewers and benign users. In this research, we analyzed the geolocations of shops in reviews, found the distinct distribution features of those in fake reviewers and benign users, and proposed a SpamTracer model that can identify fake reviewers and benign users by exploiting an improved HMM (Hidden Markov Model). Our experiment demonstrated that SpamTracer could achieve 71% accuracy and 76% recall in the unbalanced dataset, outperforming some excellent classical approaches in the aspect of stability. Furthermore, SpamTracer can help to analyze the regularities of review fraud actions. Those regularities reflect the time and location in which online shops are likely to hire fake reviewers to increase their turnover. We also found that a small group of fake reviewers tend to work with plural shops located in a small business zone.

Product fake reviews are increasing as the trend is changing toward online sales and purchases. Fake review detection is critical and challenging for both researchers and online retailers. As new techniques are introduced to catch the fake reviewer, so are their intruding approaches. In this paper, different features are amalgamated along with sentiment score to design a model that checks the model performance under different classifiers. For this purpose, six supervised learning algorithms are utilized to build the fake review detection models, utilizing LIWC, unigrams, and sentiment score features. Results show that the amalgamation of selected features is a better approach to fake review detection, achieving an accuracy score of 88.76%, which is promising compared to similar other work.

Ontology based sentiment analysis for fake review detection

Online review plays a crucial role in many current e-commerce applications. However, fake reviews would mislead users. Therefore, detecting such reviews is an important task for safeguarding the interests of users. But the review sequence has been neglected by the former work. In this paper, we explore the issue on fake review detection in review sequence, which is crucial for implementing online anti-opinion spam. We first analyze the characteristics of fake reviews. Based on review contents and reviewer behaviors, six time sensitive features are proposed to find the fake reviews. And then, we devise two type of detection methods, the supervised and the threshold-based, for spotting the fake reviews as early as possible. Finally, we carry out intensive experiments on a real-world review set to verify the effectiveness of our methods. The experimental results show that our methods can identify the fake reviews orderly with high precision and recall.

Restaurants’ motivations to solicit fake reviews: A competition perspective

Detecting review spam is important for current e-commerce applications. However, the posted order of review has been neglected by the former work. In this paper, we explore the issue on fake review detection in review sequence, which is crucial for implementing online anti-opinion spam. We analyze the characteristics of fake reviews firstly. Based on review contents and reviewer behaviors, six time sensitive features are proposed to highlight the fake reviews. And then, we devise supervised solutions and a threshold-based solution to spot the fake reviews as early as possible. The experimental results show that our methods can identify the fake reviews orderly with high precision and recall.

Fake review detection and its elimination from the given dataset using different Natural Language Processing (NLP) techniques is important in several aspects. In this article, the fake review dataset is trained by applying two different Machine Learning (ML) models to predict the accuracy of how genuine are the reviews in a given dataset. The rate of fake reviews in E-commerce industry and even other platforms is increasing when depend on product reviews for the item found online on different websites and applications. The products of the company were trusted before making a purchase. So this fake review problem must be addressed so that these large E-commerce industries such as Flipkart, Amazon, etc. can rectify this issue so that the fake reviewers and spammers are eliminated to prevent users from losing trust on online shopping platforms. This model can be used by websites and applications with few thousands of users where it can predict the authenticity of the review based on which the website owners can take necessary action towards them. This model is developed using Naïve Bayes and random forest methods. By applying these models one can know the number of spam reviews on a website or application instantly. To counter such spammers, a sophisticated model is required in which a need to be trained on millions of reviews. In this work “amazon Yelp dataset” is used to train the models and its very small dataset is used for training on a very small scale and can be scaled to get high accuracy and flexibility.

Gaming the system: Fake online reviews v. consumer law

Mild cognitive impairment (MCI) is a clinical condition characterized by a decline in cognitive ability and progression of cognitive impairment. It is often considered a transitional stage between normal aging and Alzheimer's disease (AD). This study aimed to compare deep learning (DL) and traditional machine learning (ML) methods in predicting MCI using plasma proteomic biomarkers. A total of 239 adults were selected from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort along with a pool of 146 plasma proteomic biomarkers. We evaluated seven traditional ML models (support vector machines (SVMs), logistic regression (LR), naïve Bayes (NB), random forest (RF), k-nearest neighbor (KNN), gradient boosting machine (GBM), and extreme gradient boosting (XGBoost)) and six variations of a deep neural network (DNN) model-the DL model in the H2O package. Least Absolute Shrinkage and Selection Operator (LASSO) selected 35 proteomic biomarkers from the pool. Based on grid search, the DNN model with an activation function of "Rectifier With Dropout" with 2 layers and 32 of 35 selected proteomic biomarkers revealed the best model with the highest accuracy of 0.995 and an F1 Score of 0.996, while among seven traditional ML methods, XGBoost was the best with an accuracy of 0.986 and an F1 Score of 0.985. Several biomarkers were correlated with the APOE-ε4 genotype, polygenic hazard score (PHS), and three clinical cerebrospinal fluid biomarkers (Aβ42, tTau, and pTau). Bioinformatics analysis using Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) revealed several molecular functions and pathways associated with the selected biomarkers, including cytokine-cytokine receptor interaction, cholesterol metabolism, and regulation of lipid localization. The results showed that the DL model may represent a promising tool in the prediction of MCI. These plasma proteomic biomarkers may help with early diagnosis, prognostic risk stratification, and early treatment interventions for individuals at risk for MCI.

Online product reviews are an important driver of customers’ purchasing behavior. Fake reviews seriously mislead consumers, challenging the fairness of the online shopping environment. Although the detection of fake reviews has progressed, several problems remain. First, fake comment recognition ignores the correlation between time and the semantics of the comment texts, which is always hidden in the context of the reviews. Second, the impact of multi-source information on fake comment recognition is not considered, as it constitutes a complex, high-dimensional, heterogeneous relationship between reviewers, reviews, stores and commodities. To overcome these problems, the present paper proposes a dynamic knowledge graph-based method for fake-review detection. Based on the characteristics of online product reviews, it first extracts four types of entities using a developed neural network model called sentence vector/twin-word embedding conditioned bidirectional long short-term memory. Time series related features are then added to the knowledge graph construction process, forming dynamic graph networks. To enhance the fake-review detection, four indicators are newly defined for determining the relationships among the four types of nodes. In experimental evaluations, our method surpassed the state-of-the-art results.

This paper presents an approach to fake review detection, essentially for online hotel reviews, by combining the review-based approach and reviewer-based approach. Different Natural Language Processing techniques such as tokenization, lemmatization, vectorization, etc. are used to extract insightful features from the review text data. After text mining, the data is used to train different classification models using machine learning algorithms that detect fake reviews. After evaluating the models, a comparison is made based on the performance metrics. Furthermore, a web based user interface is created to provide a platform that combines the knowledge of the input user information with the chosen machine learning model to perform fake review detection on the input data.

• Advanced data analytics and PySpark improve fake product review detection. • Classification and clustering methods identify patterns and reduce deceptive reviews. • Visual data exploration aids anomaly detection in review characteristics. • Techniques protect consumers and foster trust in e-commerce platforms. • Study bridges theoretical research and practical spam detection applications. In e-commerce, online reviews significantly influence consumer purchasing behavior, with authenticity directly impacting business revenues and consumer trust. This study addresses the critical issue of fake product reviews (FPRs) by analyzing their effects on consumer choice and the overall reliability of online marketplaces. To improve FPR detection, we employed PySpark and advanced data science techniques to analyze a labeled dataset of user reviews, uncovering patterns and anomalies indicative of review manipulation. By integrating classification methods such as K-Nearest Neighbors (KNN), the study demonstrates how machine learning can effectively identify and mitigate the impact of FPRs, thereby enhancing the credibility of online reviews. The results contribute to ensuring fair competition, consumer protection, and the long-term integrity of digital commerce. Future research may expand this framework by incorporating additional datasets, alternative classification algorithms, and deeper linguistic or sentiment-based analyses.

Emotion classification is a research field that aims to detect the emotions in a text using machine learning methods. In traditional machine learning (TML) methods, feature engineering processes cause the loss of some meaningful information, and classification performance is negatively affected. In addition, the success of modelling using deep learning (DL) approaches depends on the sample size. More samples are needed for Turkish due to the unique characteristics of the language. However, emotion classification data sets in Turkish are quite limited. In this study, the pretrained language model approach was used to create a stronger emotion classification model for Turkish. Well-known pretrained language models were fine-tuned for this purpose. The performances of these fine-tuned models for Turkish emotion classification were comprehensively compared with the performances of TML and DL methods in experimental studies. The proposed approach provides state-of-the-art performance for Turkish emotion classification.