Deep Deception: The story of the spycop network, by the women who uncovered the shocking truth

Product Information

f) While the theory on deception detection strongly relates it to the subject’s emotional state, most studies did not approach the problem under this perspective, rather modeling the features from behavioral changes; LSTM model performances were measured by accuracy in 5 [ 38, 50, 53, 65, 81] out the 9 studies. Those accuracies range from 0.7487 to 1.0, with a mean at 0.8886 ± 0.0965. In two cases [ 106, 110] the measure was F1-score, as 0.6390 and 0.6562. In one study [ 101] the performance was reported as Area Under the Curve, which measured 0.6650, and in the other [ 82] the Unweighted Average Recall measured 0.7471. Our main goal is to comprehensively understand of the state of research regarding deception detection with Machine Learning. To do so, we surveyed, studied, and selected a collection of 81 documents out of 648 retrieved from four scientific databases. We report our findings in both quantitative and qualitative fashions.

g) Machiavellianism is a psychological trait that can change the interpretation of detection cues, but authors did not exploited it; It has severely affected our ability to trust other people, or to form intimate relationships again. You can’t compensate for that,” she says, noting that even the state compensation system does not see the world from a woman’s perspective, being more inclined to focus on loss of earnings. There are several Machine Learning algorithms based on different theoretical frameworks and strategies [ 19], such as Decision Trees [ 28], Naïve Bayes [ 29], Support Vector Machines [ 30], K-Means [ 31], Random Forests [ 32] and Neural Networks [ 33]. We consider this theme important because it divides the studies into two distinct groups: one based on English and another based on other languages. Verbal cues depend heavily on language aspects. Thus, most of the knowledge found in English-based studies needs to be adapted or tested for other languages. Statistical details can be found in section 2 (Language analysis) in S6 File (Statistical Analysis Notebook). For this study, we define both “deceiving” and”lying” as the intentional act of making the interlocutor believe in something the deceiver considers false [ 1]; it is a conscious and deliberated act, perpetrated by the deceiver [ 2]. However, a false information believed to be true by the emitter is not considered deceptive.j) Vocal cues were almost exclusively provided by OpenSMILE and in general are reported as a highly discriminant feature set; On 5thApril, CWJ are co-hosting a book launch of ‘Deep Deception’, hosted by Samira Ahmed where I shall join the authors of this important book to explore the dark and seedy scandal of undercover policing that impacted so profoundly on so many women’s lives. All the metadata was extracted directly from the selected corpus and no value was, by any means, inferred or interpreted. Sometimes, the total number of features was summed when the text didn’t present it, but all the primitive values were there. Such metadata describes the source of training data, training strategy, Machine Learning methods, dataset sizes, predictors exploited, cues complexity, modality cardinality, performance levels, and performance metrics. Actors specialize in displaying fake emotions, and the face plays an essential role in this context. Would they be able to mislead an already trained Machine Learning Deception Detector? Ekman talks about how to detect false emotions [ 2], but not one study included that in their research.

The variation of size, source, and features of the data consumed is so high that it’s impossible compare works results. Although a multitude of distinct approaches had been tested with several performance levels for over a decade, the field still seems to be at initial stages of development. Three studies experimented on psychological features. One consumed NEO-FFI (Neuroticism-Extraversion-Openness Five-Factor Inventory) scores along with demographic and vocal cues [ 105]. NEO-FFI is a five-factor personality model based on an empirically developed taxonomy of personality traits. This model measures five personality components: Openness to experience, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. n) Neural networks, Support Vector Machines, Logistic Regression, K-Nearest Neighbor and Decision Trees were the most exploited Machine Learning algorithms;

This finding works as a reasonable explanation for the variety of experiment results. While there are several reliable deception clues, exceptions exist because they may suffer from certain interferences, particularly the so-called Othello error [ 2]. The Othello error occurs when lie-catchers confuse emotions and motivations. The emotion is present, but it does not originate from deception.