This paper presents a study on detecting deceptive opinion spam, which is fictitious opinions designed to appear authentic to deceive readers. The authors develop and compare three approaches to detect deceptive opinions: (1) text categorization using n-gram features, (2) psycholinguistic deception detection based on psychological effects of lying, and (3) genre identification based on the difference between imaginative and informative writing. They create a large-scale, publicly available dataset of 800 reviews, including 400 truthful and 400 gold-standard deceptive reviews. Using this dataset, they train classifiers and find that n-gram-based text categorization techniques outperform psycholinguistic and genre-based approaches. A combined classifier using both n-gram and psychological deception features achieves nearly 90% accuracy. The study also reveals that deceptive opinions are more likely to use superlatives and focus on external aspects of the hotel, while truthful opinions are more specific about spatial configurations. The authors also find that human judges perform at-chance in detecting deceptive opinions, highlighting the need for automated detection methods. Theoretical contributions include the importance of considering context and motivation in deception detection, and the difficulties liars face in encoding spatial information. The study shows that deceptive opinion spam detection is well beyond the capabilities of most human judges, and that automated methods based on n-gram features and psycholinguistic cues can achieve high accuracy. The authors also suggest that future work should consider both contextual and motivational factors in deception detection.This paper presents a study on detecting deceptive opinion spam, which is fictitious opinions designed to appear authentic to deceive readers. The authors develop and compare three approaches to detect deceptive opinions: (1) text categorization using n-gram features, (2) psycholinguistic deception detection based on psychological effects of lying, and (3) genre identification based on the difference between imaginative and informative writing. They create a large-scale, publicly available dataset of 800 reviews, including 400 truthful and 400 gold-standard deceptive reviews. Using this dataset, they train classifiers and find that n-gram-based text categorization techniques outperform psycholinguistic and genre-based approaches. A combined classifier using both n-gram and psychological deception features achieves nearly 90% accuracy. The study also reveals that deceptive opinions are more likely to use superlatives and focus on external aspects of the hotel, while truthful opinions are more specific about spatial configurations. The authors also find that human judges perform at-chance in detecting deceptive opinions, highlighting the need for automated detection methods. Theoretical contributions include the importance of considering context and motivation in deception detection, and the difficulties liars face in encoding spatial information. The study shows that deceptive opinion spam detection is well beyond the capabilities of most human judges, and that automated methods based on n-gram features and psycholinguistic cues can achieve high accuracy. The authors also suggest that future work should consider both contextual and motivational factors in deception detection.