Deceptive Reviews Research Articles

Exploiting Burstiness in Reviews for Review Spammer Detection

Online product reviews have become an important source of user opinions. Due to profit or fame, imposters have been writing deceptive or fake reviews to promote and/or to demote some target products or services. Such imposters are called review spammers. In the past few years, several approaches have been proposed to deal with the problem. In this work, we take a different approach, which exploits the burstiness nature of reviews to identify review spammers. Bursts of reviews can be either due to sudden popularity of products or spam attacks. Reviewers and reviews appearing in a burst are often related in the sense that spammers tend to work with other spammers and genuine reviewers tend to appear together with other genuine reviewers. This paves the way for us to build a network of reviewers appearing in different bursts. We then model reviewers and their co-occurrence in bursts as a Markov Random Field (MRF), and employ the Loopy Belief Propagation (LBP) method to infer whether a reviewer is a spammer or not in the graph. We also propose several features and employ feature induced message passing in the LBP framework for network inference. We further propose a novel evaluation method to evaluate the detected spammers automatically using supervised classification of their reviews. Additionally, we employ domain experts to perform a human evaluation of the identified spammers and non-spammers. Both the classification result and human evaluation result show that the proposed method outperforms strong baselines, which demonstrate the effectiveness of the method.

Fast Detection of Deceptive Reviews by Combining the Time Series and Machine Learning

With the rapid growth of online product reviews, many users refer to others’ opinions before deciding to purchase any product. However, unfortunately, this fact has promoted the constant use of fake reviews, resulting in many wrong purchase decisions. The effective identification of deceptive reviews becomes a crucial yet challenging task in this research field. The existing supervised learning methods require a large number of labeled examples of deceptive and truthful opinions by domain experts, while the available unsupervised learning methods are inefficient because they depend on the features of reviewers to detect each fake review. Therefore, by focusing on the detection efficiency problem and the limitation of large amount of labeled examples dependence, in this paper, we proposed an effective semisupervised learning approach for detecting spam reviews. Firstly, a time series model of all the reviews of a product is constructed, and then the suspected time intervals are captured based on the burst review increases in these intervals. Secondly, a co-training two-view semisupervised learning algorithm was performed in each captured interval, in which linguistic cues, metadata, and user purchase behaviors were synthetically employed to classify the reviews and check whether they are spam ones or not. A series of numerical experiments on a real dataset acquired from Taobao.com have confirmed the effectiveness of the proposed model, not only reaping benefits in terms of time efficiency and high accuracy but also overcoming the shortcomings of supervised learning methods, which depend on large amounts of labeled examples. And a trade-off balance was obtained between accuracy and efficiency.

Darwin’s Greatest Secret Exposed: Response to Grzegorz Malec’s De Facto Fact Denying Review of My Book

Grzegorz Malec’s “There Is No Darwin’s Greatest Secret”, a review of my book Nullius in Verba: Darwin’s Greatest Secret, takes one extremely minor finding from my book and, despite his best efforts, manages to disconfirm just one of thirty examples of that minor finding. He then takes that one disconfirmed mere minor example and presents it as evidence that he has disconfirmed all the original major findings in my book. By so doing, his deceptive review goes far beyond the counter-academic deviance of mere cherry picking, it is more a case of gross misrepresentation to the point of de facto fact denial amounting to historic revisionist behaviour.

Fake Product Monitoring and Removal for Genuine Product Feedback

A customer’s decision to purchase a product or service are primarily influenced by online reviews. Customers use online reviews, which are valuable sources of information to understand the public opinion on products and/or services. Dependability on online reviews can give rise to the potential concern that violator could give deceitful reviews in order to synthetically promote or decry products and services. This practice is known as Opinion Spam, where spammers manipulate reviews by making fake, untruthful, or deceptive reviews to get profit and boost their products, and devalue a competitor’s products. In order to tackle this issue, we propose to build a fraud risk management system and removal model. This captures fraudulent transactions based on user behaviors and network, analyses them in real-time using Data Mining, and accurately predicts the suspicious users and transactions. In this system, we use two algorithms NLP and TF-IDF to differentiate between fake and genuine reviews or feedback received by the customers

WITHDRAWN: Deceptive reviewer detection by analyzing web data using HMM and similarity measures

Abstract Deceptive review on online platforms is the bitter reality in the present day. These reviews are specially written to encourage or downgrade specific services, brands, and products. There are lots of work has already been done on detection of these type of reviews. In all this work, either review text features or review metadata features to identify the review. The former provides detailed information as many features can be extracted from it, such as duplicate text writing techniques, sentiments, and common words. However, with growing technologies such as article rewriter and article spinner, these features can be easily tricked by deceptive reviewers. Moreover, there are no labeled datasets available for such classification problems. This research article solved the issue referred to above by firstly creating a labeled dataset. After developing a synonyms-based n-grams approach to extract features of review text and finally effectiveness, these features are tested against some state of art feature extracting techniques for different classifiers such as SVM and Naive Bayes classifier. The evaluation results show that the state-of-art technique significantly fails to detect deceptive reviews rewritten by software. The proposed method performs better for software rewritten deceptive review detection and reduces the length of feature vectors.

Fake Reviews Detection using Supervised Machine Learning

With the continuous evolve of E-commerce systems, online reviews are mainly considered as a crucial factor for building and maintaining a good reputation. Moreover, they have an effective role in the decision making process for end users. Usually, a positive review for a target object attracts more customers and lead to high increase in sales. Nowadays, deceptive or fake reviews are deliberately written to build virtual reputation and attracting potential customers. Thus, identifying fake reviews is a vivid and ongoing research area. Identifying fake reviews depends not only on the key features of the reviews but also on the behaviors of the reviewers. This paper proposes a machine learning approach to identify fake reviews. In addition to the features extraction process of the reviews, this paper applies several features engineering to extract various behaviors of the reviewers. The paper compares the performance of several experiments done on a real Yelp dataset of restaurants reviews with and without features extracted from users behaviors. In both cases, we compare the performance of several classifiers; KNN, Naive Bayes (NB), SVM, Logistic Regression and Random forest. Also, different language models of n-gram in particular bi-gram and tri-gram are taken into considerations during the evaluations. The results reveal that KNN(K=7) outperforms the rest of classifiers in terms of f-score achieving best f-score 82.40%. The results show that the f-score has increased by 3.80%when taking the extracted reviewers behavioral features into consideration.

Analyzing the effectiveness of semi-supervised learning approaches for opinion spam classification

Opinion spam detection is concerned with identifying fake reviews that are deliberately placed to either promote or discredit a product. Opinionated social media like product reviews are increasingly important resources for people as well as businesses in the decision-making process and can be easily manipulated by opportunistic individuals. To reduce this increasing impact of opinion spams, opinion spam detection approaches have been proposed, which adopt mostly supervised classification methods. However, in practice, the provided data is largely not labeled and therefore semi-supervised learning approaches are required instead. To this end, this study aims to analyze the effectiveness of several semi-supervised learning approaches for opinion spam classification. Four different semi-supervised methods are evaluated on a dataset of both genuine and deceptive hotel reviews. The results are compared with several traditional classification methods using the same amount of labeled data. According to this study, the self-training algorithm with Naive Bayes as the base classifier yields 93% accuracy. Results show that self-training is the only approach, out of the four tested semi-supervised models, that outperforms traditional supervised classification models when limited data is available. This study further shows that self-training can mitigate labeling efforts while retaining high model performance, which is useful for scenarios where limited data is available or retrieving labeled data is more costly.

Detection of Fake Reviews Using Group Model

Reviews of product or stores exist extensively in online e-commerce platform which is important for customers to make decisions. For economic reasons some dishonest people are employed to write fake reviews which is also called “opinion spamming” to promote or demote target products and services. Previous researches have made use of text similarity, linguistics, rating patterns, graph relationship and other behaviors for spammer detection. They mainly utilized product review list while it is difficult to find fake reviews by glancing over product reviews in time-descending order. Meanwhile there exists lots of useful information in the list of reviews of reviewers and relationships between reviewers when reviewers commonly reviewed the same stores. We propose the concept of review group and to the best of our knowledge, it’s the first time the review group concept is proposed and used. Review grouping algorithm is designed to effectively split reviews of reviewer into groups which participate in building novel grouping models to identify both positive and negative deceptive reviews. Several new features which are language independent based on group model are constructed. Additionally, we explore the collusion relationship between reviewers to build reviewer group collusion model. Evaluations show that the review group method and reviewer group collusion models can effectively improve the precision by 4%–7% compared to the baselines in fake reviews classification task especially when reviews are posted by professional review spammers.

Content analysis of fake consumer reviews by survey-based text categorization

As the influence of online consumer reviews grows, deceptive reviews are a worsening problem, betraying consumers' trust in reviews by pretending to be authentic and informative. This research identifies factors that can separate deceptive reviews from genuine ones. First, we create a novel means of detection by contrasting authentic versus fake word patterns specific to a given domain (e.g., hotel services). We use a survey on a crowdsourcing platform to obtain both genuine and deceptive reviews of hotels. We learned the word patterns from each category to discriminate genuine reviews from fake ones for positively and negatively evaluated reviews, respectively. We show that our All Terms procedure outperforms current benchmark methods in computational linguistics and marketing. Our extended analysis reveals the factors that determine fake reviews (e.g., a lack of details, present- and future-time orientation, and emotional exaggeration) and the factors influencing people's willingness to write fake reviews (including social media trust, product quality consciousness, deal proneness, hedonic and utilitarian consumption, prosocial behavior, and individualism). We also use our procedure to analyze more than 250,000 real-world hotel reviews to detect fake reviews and identify the hotel and review characteristics influencing review fakery in the industry (e.g., star rating, franchise hotel, hotel size, room price, review timing, and review rating).

A deceptive detection model based on topic, sentiment, and sentence structure information

Deceptive reviews on Web are a common phenomenon and how to detect them has a very important impact on products, services, and even business policies. In order to filter out deceptive reviews more accurately, a new model called Sentence Joint Topic Sentiment Model (SJTSM) is presented in this paper, which incorporates the sentence structure of reviews and the sentiment label information of words based on Latent Dirichlet Allocation (LDA) model to extract the review features. The proposed model employs Gibbs algorithm to estimate the maximum likelihood parameters and takes the vector of topic-sentiment distribution as the review features. Then a voting system of multiple-classifier, which takes the extracted review feature vector as its input is designed to realize the classification of deceptive review detection. The comparative experiments on different public datasets with other existing methods based on LDA model show that the new classifying system based on SJTSM model can achieve more satisfying classification results on deceptive review detection.

PV-DAE: A hybrid model for deceptive opinion spam based on neural network architectures

Opinion review is of great importance for both customers and organizations. Indeed, it helps customers in buying decisions and represents a valuable feedback for the companies, allowing them to improve their productions. However, numerous greedy companies resort to fake reviews in order to influence the customer and brighten the brand image, or to defame the one of their competitors. Various models are proposed in order to detect deceptive opinion reviews. Most of these models adopt traditional methods focusing on feature extraction and traditional classifiers. Unfortunately, these models do not capture the semantic aspect while ignoring the opinion’s context. In order to tackle this issue, we propose a new approach based on Paragraph Vector Distributed Bag of Words (PV-DBOW) and the Denoising Autoencoder (DAE). The proposed customized model provides a strong representation which is based on a global representation of the opinions while preserving their semantics. Indeed, the embedding vectors capture the semantic meaning of all words in the context of each opinion. The generated review representations are fed into a fully connected neural network in order to detect deceptive opinion spam. The obtained results concerning the deception dataset show that our model is effective and outperforms the existing state-of-the-art methodologies.

The impact of review valence and awareness of deceptive practices on consumers’ responses to online product ratings and reviews

ABSTRACT Many online retailers and some manufacturers/service providers have recently been engaging in questionable practices, where product reviews are often fabricated and/or posted without sufficient clarity and objectivity. Across an exploratory study and two main studies, we empirically examine this phenomenon and observe a pattern of effects that suggests that review valence (i.e., the average number of rating-stars a product receives) influences product attitudes and intentions, but that these outcomes are significantly impacted by the extent to which consumers are aware of potentially deceptive online review practices. Awareness of deceptive practices was found to differentially influence attitudes and intentions, depending upon whether the star-ratings were perfect (5/5 stars), highly positive (4.9/5 stars), or generally positive (4.5/5 or 4.7/5 stars). Participants’ perceptions of the e-retailer’s manipulative intent were also shown to mediate these effects, with higher perceptions of perceived manipulative intent yielding less favorable product attitudes and reduced purchase intentions.

An Attention-Based Unsupervised Adversarial Model for Movie Review Spam Detection

With the prevalence of the Internet, online reviews have become a valuable information resource for people. However, the authenticity of online reviews remains a concern, and deceptive reviews have become one of the most urgent network security problems to be solved. Review spams will mislead users into making suboptimal choices and inflict their trust in online reviews. Most existing research manually extracted features and labeled training samples, which are usually complicated and time-consuming. This paper focuses primarily on a neglected emerging domain - movie review, and develops a novel unsupervised spam detection model with an attention mechanism. By extracting the statistical features of reviews, it is revealed that users will express their sentiments on different aspects of movies in reviews. An attention mechanism is introduced in the review embedding, and the conditional generative adversarial network is exploited to learn users' review style for different genres of movies. The proposed model is evaluated on movie reviews crawled from Douban, a Chinese online community where people could express their feelings about movies. The experimental results demonstrate the superior performance of the proposed approach.

LDA_RAD: A Spam review detection method based on topic model and reviewer anomaly degree

Online shopping has become a popular activity among consumers. Reviews have become important standards for consumers to buy things. To detect the spam re-views that do not correctly response the right information of goods, a method based on topic model and reviewer anomaly degree is raised. This method divides spam reviews into content-type and deceptive spam review respectively. Firstly, the experimental data set is modelled by LDA topic, which detects the content-type spam reviews with different themes. Then the deceptive spam reviews are detected by the reviewer’s abnormality degree index. It assigns a score to each review according to extracted features and related weights. A score is finally combined with an adaptive weight calculation based on the abnormal period and the reviewer’s similarity to obtain the review score. The review with a high score is considered to be a spam review, and the review with a low score is a true review. Experiments show that this method has a certain improvement on the recognition rate of spam reviews.

A deceptive review detection framework: Combination of coarse and fine-grained features

Electronic commerce has become a popular shopping mode. To enhance their reputations, attract more customers, and finally obtain more benefits, dishonest sellers often recruit buyers or robots to post a large number of deceptive reviews to mislead users. According to the interpretability of learning results, existing methods for detecting deceptive reviews can be mainly divided into explicit feature-based mining ones and neural network-based implicit feature mining ones. The nature of these works is accurate text classification based on coarse-grained features (e.g., topic, sentence, and document) or fine-grained features (e.g., word). To take full merits of existing approaches, this paper proposes a new framework that explores a method to combine the coarse-grained features and the fine-grained features. In this framework, the coarse-grained implicit semantic features of the topic distribution are learned by the concatenation of a Latent Dirichlet Allocation (LDA) topic model and a 2-layered neural network. The fine-grained implicit semantic features from the word vectors representation of the reviews are parallelly learned by a deep learning framework. Finally, these two granular features are combined and adopted to train a Support Vector Machine (SVM) classifier for detecting whether a review is deceptive or not. To verify the effectiveness and performance of this framework, we derive three models by specifying three popular deep learning models, such as TextCNN, long short-term memory (LSTM), and Bi-directional LSTM (BiLSTM) to learn the fine-grained features. Experimental results on a mixed-domain dataset and balanced/unbalanced in-domain datasets show that all the combination models are superior to the corresponding baseline models considering single features.

Spam Detection in Online Comments Based on Feature Weight Breakdown

The user reviews posted online by the Internet users about a product plays a vital role in determining its success in the market. The reviews also influence the purchase decision of the consumers. The chances of getting cheated by fake reviews are very high because detecting spams in reviews is not an easy task either manually or automatically. Hence there is a need to evolve new techniques and methods to outperform the smartness of spammers. In this paper, we propose a Heterogeneous Feature Weight Analysis framework for extracting various features related to the review and certain parameters are calculated from these features to form a pattern for deceptive reviews. The features associated with the review are review content, review rating and user centric characteristics which are pulled out from the dataset retrieved from Amazon. This analysis has helped us to categorize reviews into normal and suspicious reviews. We have executed our algorithm in Python software and were able to achieve an accuracy of 71.6% inprediction.

DCWord: A Novel Deep Learning Approach to Deceptive Review Identification by Word Vectors

Due to the anonymous and free-for-all characteristics of online forums, it is very hard for human beings to differentiate deceptive reviews from truthful reviews. This paper proposes a deep learning approach for text representation called DCWord (Deep Context representation by Word vectors) to deceptive review identification. The basic idea is that since deceptive reviews and truthful reviews are composed by writers without and with real experience on using the online purchased goods or services, there should be different contextual information of words between them. Unlike state-of-the-art techniques in seeking best linguistic features for representation, we use word vectors to characterize contextual information of words in deceptive and truthful reviews automatically. The average-pooling strategy (called DCWord-A) and max-pooling strategy (called DCWord-M) are used to produce review vectors from word vectors. Experimental results on the Spam dataset and the Deception dataset demonstrate that the DCWord-M representation with LR (Logistic Regression) produces the best performances and outperforms state-of-the-art techniques on deceptive review identification. Moreover, the DCWord-M strategy outperforms the DCWord-A strategy in review representation for deceptive review identification. The outcome of this study provides potential implications for online review management and business intelligence of deceptive review identification.

A Method for the Detection of Fake Reviews Based on Temporal Features of Reviews and Comments

Online reviews and comments after product sales have become very important for making buying and selling decisions. Fake reviews will affect such decisions due to deceptive information, leading to financial losses for the consumers. Identification of fake reviews has thus received a great deal of attention in recent years. However, most websites have only focused on dealing with problematic reviews and comments. Amazon and Yelp would only remove possible fake reviews without questioning the sellers who could continue posting deceptive reviews for business purposes. In this paper, we propose a method for the detection of fake reviews based on review records associated with products. We first analyze the characteristics of review data using a crawled Amazon China dataset, which shows that the patterns of review records for products are similar in normal situations. In the proposed method, we first extract the review records of products to a temporal feature vector and then develop an isolation forest algorithm to detect outlier reviews by focusing on the differences between the patterns of product reviews to identify outlier reviews. We will verify the effectiveness of our method and compare it to some existing temporal outlier detection methods using the crawled Amazon China dataset. We will also study the impact caused by the parameter selection of the review records. Our work provides a new perspective of outlier review detection and our experiment demonstrates the effectiveness of our proposed method.

Incorporating feature representation into BiLSTM for deceptive review detection

Consumers are increasingly influenced by product reviews when purchasing goods or services. At the same time, deceptive reviews usually mislead users. It is inefficient and inaccurate to manually identify deceptive reviews in massive reviews. Therefore, automatically identifying deceptive reviews has become a research trend. Most of existing methods are less effective since they are lack of deeply understanding of reviews. We propose a neural network method with bidirectional long short-term memory (BiLSTM) and feature combination to learn the representation of deceptive reviews. We conduct a large amount of experiments and demonstrate the effectiveness of our proposed method. Specifically, in the mixed-domain detection experiment, the results prove that our model is effective by making comparisons with other neural network-based methods. BiLSTM gives more than 3% improvement in F1 score compared with the most advanced neural network method. Since feature selection plays an important role in this direction, we combine features to improve the performance. Then we get 87.6% F1 value which outperforms the state-of-the-art method. Moreover, in the cross-domain detection experiment, our method achieves 82.4% F1 value which is about 6% higher than the state-of-the-art method on restaurant domain, and it is also robust on doctor domain.

A Review Structure Based Ensemble Model for Deceptive Review Spam

Consumers’ purchase behavior increasingly relies on online reviews. Accordingly, there are more and more deceptive reviews which are harmful to customers. Existing methods to detect spam reviews mainly take the problem as a general text classification task, but they ignore the important features of spam reviews. In this paper, we propose a novel model, which splits a review into three parts: first sentence, middle context, and last sentence, based on the discovery that the first and last sentence express stronger emotion than the middle context. Then, the model uses four independent bidirectional long-short term memory (LSTM) models to encode the beginning, middle, end of a review and the whole review into four document representations. After that, the four representations are integrated into one document representation by a self-attention mechanism layer and an attention mechanism layer. Based on three domain datasets, the results of in-domain and mix-domain experiments show that our proposed method performs better than the compared methods.