Dissecting AI-Generated Fake Reviews: Detection and Analysis of GPT-Based Restaurant Reviews on Social Media

2023 | International Conference on Information Systems | Citations: 0

Authors: Gambetti, Alessandro; Han, Qiwei

Abstract: Recent advances in generative models such as GPT may be used to fabricate indis ... Expand

Abstract: Recent advances in generative models such as GPT may be used to fabricate indistinguishable fake customer reviews at a much lower cost, posing challenges for social media platforms to detect this kind of content. This study addresses two research questions: (1) the effective detection of AI-generated restaurant reviews generated from high-quality elite authentic reviews, and (2) the comparison of out-of-sample predicted AI-generated reviews and authentic reviews across multiple dimensions of review, user, restaurant, and content characteristics. We fine-tuned a GPT text detector to predict fake reviews, significantly outperforming existing solutions. We applied the model to predict non-elite reviews that already passed the Yelp filtering system, revealing that AI-generated reviews typically score higher ratings, users posting such content have less established Yelp reputations and AI-generated reviews are more comprehensible and less linguistically complex than human-generated reviews. Notably, machine-generated reviews are more prevalent in low-traffic restaurants in terms of customer visits. Collapse

Topics: fake review artificial intelligence application programming interface open source pandemic

Methods: machine learning survey computational algorithm sensitivity analysis ANOVA

The toxicity of our (Sim) Cities: Prevalence of dark participation in games and perceived effectiveness of reporting tools

2022 | HICSS | Citations: 0

Authors: Kowert, Rachel; Cook, Christine L

Abstract: Dark participation in games (i.e., trolling and toxic behavior) have been gainin ... Expand

Abstract: Dark participation in games (i.e., trolling and toxic behavior) have been gaining ever-increasing academic attention as a negative aspect of online gaming. Much of the literature in this area has focused on the personality and identity of the perpetrators, but this has been largely outside of the gaming context. The present study aims to explore the prevalence rates of dark participation in the online gaming community, the reporting function to punish deviant players, and the importance of dual identities (troll and gamer) in the perpetration of deviant in-game behaviors. Our results indicated that nearly all players in our sample had been victims of dark participation or witnessed in-game victimization, suggesting that it is a major problem in the community, but that many players also use the reporting function. Troll identity was predictive of these behaviors. Theoretical and practical implications are discussed. Collapse

Topics: spamming misinformation personality social identity internet technology

Methods: survey logistic regression self reported survey survey design

Fraudulent review detection model focusing on emotional expressions and explicit aspects: investigating the potential of feature engineering

2022 | Decision Support Systems | Citations: 5

Authors: Kumar, Ajay; Gopal, Ram D.; Shankar, Ravi; Tan, Kim Hua

Abstract: Reading customer reviews before purchasing items online has become a common prac ... Expand

Abstract: Reading customer reviews before purchasing items online has become a common practice; however, some companies use machine learning (ML) algorithms to generate false reviews in order to create positive brand images of their own products and negative images of competitors' offerings. Existing techniques use review content to identify fraudulent reviewers; however, spammers become more intelligent, started to learn from their mistakes, and changed their tactics in order to avoid detection techniques. Thus, investigating fraudulent ac­ counts' behaviour of generating fake negative or positive reviews for competitors or themselves and the necessity of ML classifiers to identify fraudulent reviews, is more important than ever. In this research, we present a novel feature engineering approach in which we (1) extract several “review-centric” and “reviewer-centric” features from a dataset; (2) combine the cumulative effects of features distributions into a unified model that represents overall behavior of the fraudulent reviewers; (3) investigate the role of effective data pre-processing to improve detection accuracy; and (4) develop a probabilistic approach to detect fraudulent reviewers by learning a novel M-SMOTE model over a derived balanced dataset and feature distributions, which outperforms other ML models. Our study contributes to the literature on digital platforms and fraudulent review detection with significant managerial and theoretical implications through these novel findings. Collapse

Topics: online review electronic commerce website marketing management logistics management

Methods: feature engineering machine learning computational algorithm XGBoost product review

Tragedy of the Commons - A Critical Study of Data Quality and Validity Issues in Crowd Work-Based Research

2022 | HICSS | Citations: 0

Authors: Xia, Huichuan

Abstract: Academic scholars have leveraged crowd work platforms such as MTurk to conduct r ... Expand

Abstract: Academic scholars have leveraged crowd work platforms such as MTurk to conduct research and collect data. Though prior studies have discussed data quality and validity issues in crowd work via surveys and experiments, they kind of neglected to explore the scholars’ and particularly the IRB’s ethical concerns in these respects. In this study, we interviewed 17 scholars from six disciplines and 15 IRB directors and analysts in the U.S. to fill this research gap. We identified common themes among our respondents but also discovered distinctive and even opposing views regarding the approval rate, rejection, internal and external research validity. Based on the findings, we discussed a potential Tragedy of the Commons regarding the data quality deterioration and the disciplinary differences regarding validity in crowd work-based research. Finally, we advocated that the IRB’s ethical concerns in crowd work-based research should be heard and respected further. Collapse

Topics: crowdsourcing data quality spamming privacy data protection

Methods: qualitative interview qualitative content analysis personal interview

Theories: theory of the commons

Uncovering Thematic Dimensions of Fraudulent Emails using Topic Modelling

2021 | Americas Conference on Information Systems | Citations: 0

Authors: Bera, Debalina; Kim, Dan

Abstract: Recently, research on fraudulent email detection has expanded from its purely t ... Expand

Abstract: Recently, research on fraudulent email detection has expanded from its purely technological orientation into understanding the role of human behavior in security breaches. However, an area that has been lacking is literature/theory-grounded content analysis in social engineering threat. While there exists an extensive body of anecdotal subjective literature, factors that account for phishing attack's incessant success remain largely speculative. To better understand this, we propose a dimensional framework to identify the thematic nature of phishing emails that would not only help to understand the underlining technical subterfuge but also enhance our knowledge to effectively sense contextual social engineering tricks and the nature of such attacks, from multiple angles. To validate the dimensional elements, we will employ topic modeling techniques to analyze the themes of fraudulent email attack utilizing four benchmark datasets. Based on the analysis, this study discusses theoretical contributions and practical recommendations along with further directions of the research. Collapse

Topics: email phishing phishing email spam electronic mail IT security

Methods: topic model term frequency–inverse document frequency logistic regression k-means clustering self organizing map

A Hybrid Approach of Machine Learning and Lexicons to Sentiment Analysis: Enhanced Insights from Twitter Data of Natural Disasters

2021 | Information Systems Frontiers | Citations: 7

Authors: Mendon, Shalak; Dutta, Pankaj; Behl, Abhishek; Lessmann, Stefan

Abstract: The success factor of sentimental analysis lies in identifying the most occurrin ... Expand

Abstract: The success factor of sentimental analysis lies in identifying the most occurring and relevant opinions among users relating to the particular topic. In this paper, we develop a framework to analyze users’ sentiments on Twitter on natural disasters using the data pre-processing techniques and a hybrid of machine learning, statistical modeling, and lexicon-based approach. We choose TF-IDF and K-means for sentiment classification among affinitive and hierarchical clustering. Latent Dirichlet Allocation, a pipeline of Doc2Vec and K-means used to capture themes, then perform multi-level polarity indices classification and its time series analysis. In our study, we draw insights from 243,746 tweets for Kerala’s 2018 natural disasters in India. The key findings of the study are the classification of sentiments based on similarity and polarity indices and identifying themes among the topics discussed on Twitter. We observe different sets of emotions and influencers, among others. Through this case example of Kerala floods, it shows how the government and other organizations could track the positive/negative sentiments concerning time and location; gain a better understanding of the topic of discussion trending among the public, and collaborate with crucial Twitter users/influencers to spread and figure out the gaps in the implementation of schemes in terms of design and execution. This research’s uniqueness is the streamlined and efficient combination of algorithms and techniques embedded in the framework used in achieving the above output, which can be integrated into a platform with GUI for further automation. Collapse

Topics: Twitter Python social media natural hazard Xing

Methods: term frequency–inverse document frequency Bayesian analysis computational algorithm doc2vec Gibbs sampling

Measuring the impact of spammers on e-mail and Twitter networks

2019 | International Journal of Information Management | Citations: 0

Authors: Fronzetti Colladon, Andrea; Gloor, Peter A.

Abstract: This paper investigates the research question if senders of large amounts of irr ... Expand

Abstract: This paper investigates the research question if senders of large amounts of irrelevant or unsolicited information – commonly called “spammers” – distort the network structure of social networks. Two large social networks are analyzed, the first extracted from the Twitter discourse about a big telecommunication company, and the second obtained from three years of email communication of 200 managers working for a large multinational company. This work compares network robustness and the stability of centrality and interaction metrics, as well as the use of language, after removing spammers and the most and least connected nodes. The results show that spammers do not significantly alter the structure of the information-carrying network, for most of the social indicators. The authors additionally investigate the correlation between e-mail subject line and content by tracking language sentiment, emotionality, and complexity, addressing the cases where collecting email bodies is not permitted for privacy reasons. The findings extend the research about robustness and stability of social networks metrics, after the application of graph simplification strategies. The results have practical implication for network analysts and for those company managers who rely on network analytics (applied to company emails and social media data) to support their decision-making processes. Collapse

Topics: electronic mail Twitter social network social media intranet

Methods: experiment case study network analysis machine learning method

Framings in collective action: Case of online #boycott

2019 | Americas Conference on Information Systems | Citations: 0

Authors: Prabhat, Shantanu; Motwani, Aditya; Mangurkar, Isha; Rangaswamy, Nimmi

Abstract: Online platforms are being increasingly appropriated for setting socio-politica ... Expand

Abstract: Online platforms are being increasingly appropriated for setting socio-political narratives (both offline and online). We investigate one such case study from India where a nationalist rhetoric took the form of a call to boycott an e-commerce company, Snapdeal, endorsed by Aamir Khan, an Indian celebrity. We term this a case of online collective action where Twitter users banded together to uninstall/downvote the Snapdeal app on Playstore. Through the use of "frame analysis" -a sociological device, we seek to understand the different "arguments" and "calls to action" that appear in the process of online collective action and how effective were they in terms of user engagement. We investigate these through a deductive coding process. We identify six major logical fallacies and three primary calls to action strategies and present their engagement metrics. We also discuss the role of frames in setting political narratives online and offer learnings for designing online civic media tools. Collapse

Topics: Twitter digital distribution mobile application market mobile application social media

Methods: parametric test ANOVA case study descriptive statistic qualitative content analysis

Seller's creditworthiness in the online service market: A study from the control perspective

2019 | Decision Support Systems | Citations: 2

Authors: Zheng, Haichao; Xu, Bo; Lin, Zhangxi

Abstract: Information asymmetry and the community-based nature of online marketplaces may ... Expand

Abstract: Information asymmetry and the community-based nature of online marketplaces may lead to opportunistic behaviors by sellers. This study investigates seller opportunism in the online service market from the control perspective. By analyzing data on online transactions and trade disputes from the largest crowdsourcing platform in China, Zhubajie, this study found that a seller's opportunistic behaviors, such as the non-delivery or late delivery of the final solution, are related to social and formal control in the market. Specifically, social control, i.e., the number of seller fans and members the seller follows, negatively affects opportunistic behaviors by the seller; the seller's reputation score negatively affects opportunistic behaviors, but the effect is weakened when social control is high. The seller's rank in the market is positively related to opportunistic behaviors. This paper contributes to research on online marketplaces by exploring seller creditworthiness from the control perspective. It provides implications for online marketplaces for controlling seller opportunism and enhancing creditworthiness in the market. Collapse

Topics: crowdsourcing electronic market spamming crowdfunding social capital

Methods: regression analysis method robustness check post-hoc analysis statistical hypothesis test logistic regression

Theories: control theory

Detection of Spammers in Twitter marketing: A Hybrid Approach Using Social Media Analytics and Bio Inspired Computing

2018 | Information Systems Frontiers | Citations: 6

Authors: Aswani, Reema; Kar, Arpan Kumar; Vigneswara Ilavarasan, P.

Abstract: Customer engagement is drastically improved through Web 2.0 technologies, especi ... Expand

Abstract: Customer engagement is drastically improved through Web 2.0 technologies, especially social media platforms like Twitter. These platforms are often used by organizations for marketing, of which creation of numerous spam profiles for content promotion is common. The present paper proposes a hybrid approach for identifying the spam profiles by combining social media analytics and bio inspired computing. It adopts a modified K-Means integrated Levy flight Firefly Algorithm (LFA) with chaotic maps as an extension to Firefly Algorithm (FA) for spam detection in Twitter marketing. A total of 18,44,701 tweets have been analyzed from 14,235 Twitter profiles on 13 statistically significant factors derived from social media analytics. A Fuzzy C-Means Clustering approach is further used to identify the overlapping users in two clusters of spammers and non-spammers. Six variants of K-Means integrated FA including chaotic maps and levy flights are tested. The findings indicate that FA with chaos for tuning attractiveness coefficient using Gauss Map converges to a working solution the fastest. Further, LFA with chaos for tuning the absorption coefficient using sinusoidal map outperforms the rest of the approaches in terms of accuracy. Collapse

Topics: Twitter email spam social media social media analytics user-generated content

Methods: computational algorithm cluster analysis topic model literature study mixed method

Theories: chaos theory