Article

An Ai Method to Score Celebrity Visual Potential from Human Faces

Authors:
To read the full-text of this research, you can request a copy directly from the authors.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... Emotion analysis is conducted using image processing techniques, with studies directly focusing on the human face. In a study by [14], they attempted to calculate the charisma score of the human face. They aimed to determine the visual attractiveness of celebrities by using both celebrities and noncelebrities. ...
... A similar previous study by [14] attempted to identify the features that distinguish the facial characteristics of celebrities. They constructed their datasets from facial images of both celebrities and non-celebrities. ...
Article
Full-text available
This study analyzes a social phenomenon using technical methods to uncover the underlying reasons for the researched phenomenon. This study investigates the authenticity of facial expressions and emotional cues of well-known film actors in Turkish comedy cinema films. For the study, 480 video data samples related to the actors were collected from the social media platform YouTube. The videos were categorized into smile, surprise, and anger, with 120 samples analyzed for each category. The Kaggle database containing facial expressions of smile, surprise, and anger from regular individuals was utilized to compare the images. The Local Binary Pattern (LBP) feature extraction technique was employed to extract features from the images. Machine learning models were then constructed using the extracted features. Based on the classification results, the accuracy values were 99.37% for the smile category, 97.19% for the surprise category, and 97.81% for the anger category. The analysis results show that the emotional expressions of film actors and normal individuals differ. This study aims to develop a unique perspective by highlighting the distinctive characteristics of renowned actors through their emotional expressions.
... Other publications directly predict relevant Marketing Outcomes from images. Examples include return rate prediction (Dzyabura et al. 2019), inference of design aesthetics (Burnap et al. 2021), or other subjective variables such as number of likes, or visual potential of being a celebrity (Feng et al. 2022) (e.g., for selecting suitable endorsers for campaign collaboration). Lastly, articles on Content & Context identify product content on images or classify the product usage or consumption context. ...
... Next to the task areas outlined in Figure 1, we find no consensus among marketing researchers on optimal method choice. Publications typically contain a few qualitative arguments for using a specific method, and seldom a comparison of different convolutional neural networks (e.g., Feng et al. 2022 Zhang and Luo 2022), too. We therefore investigate the performance of these methodological alternatives and compare them to additional methods that are less common in marketing. ...
Article
Full-text available
Smartphones have made it nearly effortless to share images of branded experiences. This research classifies social media brand imagery and studies user response. Aside from packshots (standalone product images), two types of brand-related selfie images appear online: consumer selfies (featuring brands and consumers’ faces) and an emerging phenomenon the authors term “brand selfies” (invisible consumers holding a branded product). The authors use convolutional neural networks to identify these archetypes and train language models to infer social media response to more than a quarter-million brand-image posts (185 brands on Twitter and Instagram). They find that consumer-selfie images receive more sender engagement (i.e., likes and comments), whereas brand selfies result in more brand engagement, expressed by purchase intentions. These results cast doubt on whether conventional social media metrics are appropriate indicators of brand engagement. Results for display ads are consistent with this observation, with higher click-through rates for brand selfies than for consumer selfies. A controlled lab experiment suggests that self-reference is driving the differential response to selfie images. Collectively, these results demonstrate how (interpretable) machine learning helps extract marketing-relevant information from unstructured multimedia content and that selfie images are a matter of perspective in terms of actual brand engagement.
Article
Full-text available
Previous research on non-facial features demonstrated that masculinity and femininity correlated highly with perceived competence and warmth, respectively. Several studies focused on dimorphic facial cues and found an association between masculine faces and competence. However, there’s no study exploring the association between facial dimorphism and social judgment both using explicit and implicit experimental paradigms, i.e. Triad Classification Task, Implicit Associate Task. This study examined the association of masculinity/femininity and competence/warmth via explicit and implicit measures in three experiments. The results showed that participants saw feminine/masculine faces as more consistent with warmth/competence for both male and female faces. Besides, it was found that the above associations were more obvious in female participants. The current studies extended research of effects of dimorphic facial cue in social judgment and provided direct evidence linking facial masculinity with perceived competence, and facial femininity with perceived warmth.
Article
Full-text available
There is ample evidence that morphological and social cues in a human face provide signals of human personality and behaviour. Previous studies have discovered associations between the features of artificial composite facial images and attributions of personality traits by human experts. We present new findings demonstrating the statistically significant prediction of a wider set of personality features (all the Big Five personality traits) for both men and women using real-life static facial images. Volunteer participants (N = 12,447) provided their face photographs (31,367 images) and completed a self-report measure of the Big Five traits. We trained a cascade of artificial neural networks (ANNs) on a large labelled dataset to predict self-reported Big Five scores. The highest correlations between observed and predicted personality scores were found for conscientiousness (0.360 for men and 0.335 for women) and the mean effect size was 0.243, exceeding the results obtained in prior studies using ‘selfies’. The findings strongly support the possibility of predicting multidimensional personality profiles from static facial images using ANNs trained on large labelled datasets. Future research could investigate the relative contribution of morphological features of the face and other characteristics of facial images to predicting personality.
Article
Full-text available
There is ample evidence that a human face provides signals of human personality and behaviour. Previous studies have found associations between the features of artificial composite facial images and attributions of personality traits by human experts. We present new findings demonstrating the statistically significant prediction of a wider set of personality features (all the Big Five personality traits) for both men and women using real-life static facial images. Volunteer participants (N = 12,447) provided their face photographs (31,367 images) and completed a self-report measure of the Big Five traits. We trained a cascade of artificial neural networks (ANNs) on a large labelled dataset to predict self-reported Big Five scores. The highest correlations were found for conscientiousness (.360 for men and .335 for women), exceeding the results obtained in prior studies. The findings provide strong support for the hypothesis that it is possible to predict multidimensional personality profiles from static facial images using ANNs trained on large labelled datasets.
Article
Full-text available
Sarcasm can indirectly express aggression, though it might be more cognitively demanding to produce. We expected positive correlations between trait anger and self-reported, but not objective (response choice task) sarcasm use. Results revealed associations between trait anger and self-reported sarcasm use only. Regression analysis showed trait anger to be a significant, independent predictor of self-reported sarcasm use. The results are discussed with reference to expanding the understanding of individual differences in nonliteral language use.
Article
Full-text available
Tree-based machine learning models such as random forests, decision trees and gradient boosted trees are popular nonlinear predictive models, yet comparatively little attention has been paid to explaining their predictions. Here we improve the interpretability of tree-based models through three main contributions. (1) A polynomial time algorithm to compute optimal explanations based on game theory. (2) A new type of explanation that directly measures local feature interaction effects. (3) A new set of tools for understanding global model structure based on combining many local explanations of each prediction. We apply these tools to three medical machine learning problems and show how combining many high-quality local explanations allows us to represent global structure while retaining local faithfulness to the original model. These tools enable us to (1) identify high-magnitude but low-frequency nonlinear mortality risk factors in the US population, (2) highlight distinct population subgroups with shared risk characteristics, (3) identify nonlinear interaction effects among risk factors for chronic kidney disease and (4) monitor a machine learning model deployed in a hospital by identifying which features are degrading the model’s performance over time. Given the popularity of tree-based machine learning models, these improvements to their interpretability have implications across a broad set of domains. Tree-based machine learning models are widely used in domains such as healthcare, finance and public services. The authors present an explanation method for trees that enables the computation of optimal local explanations for individual predictions, and demonstrate their method on three medical datasets.
Article
Full-text available
The authors examine how gender identity and local–global identity—two important consumer identities and segmentation bases—affect consumer price sensitivity. These identities may be associated with similar behavioral norms such that a female identity and a local identity are associated with the norms of small groups and intimate relationships, whereas a male identity and a global identity are associated with the norms of large groups and broad relationships. Thus, a female identity and a global (vs. local) identity, as is the case with a male identity and a local (vs. global) identity, results in identity incongruence (vs. congruence). The authors argue that identity incongruence depletes cognitive resources and induces affective processing, which in turn activates a sacrifice mindset and results in lower price sensitivity. Five studies with different research designs, divergent measures of key constructs, and diverse samples support the theories and identify boundary conditions of the focal relationship. The findings expand the literature on multiple consumer identities and price sensitivity and provide useful guidelines for global companies’ pricing strategies.
Article
Full-text available
This study aimed to investigate the emotional impact of sarcasm. Previous research in this area has mainly required participants to answer questions based on written materials, and results have been mixed. With the aim of instead examining the emotional impact of sarcasm when used in a more conversational setting, the current study utilised animated video clips as stimuli. In each clip, one individual answered general knowledge questions while the other provided feedback that could be delivered either literally or sarcastically, and either complimented or criticised the question answerer’s performance. Participants rated the feedback based on (1) the anticipated emotional impact on the recipient, (2) how the speaker intended the recipient to feel, and (3) whether the speaker intended to be humorous. Results overall supported the Tinge Hypothesis (Dews & Winner, 1995), showing that sarcastic criticism was rated as less negative than literal criticism, and sarcastic compliments (also termed “ironic praise”) were rated as less positive than literal compliments, when judged from both the perspective of the recipient and of the speaker. The speaker was also perceived to be intending to be more humorous when sarcastic feedback was given.
Article
Full-text available
Typically, before and after surgical correction faces are assessed on still images by surgeons, orthodontists, the patients, and family members. We hypothesized that judgment of faces in motion and by naïve raters may closer reflect the impact on patients’ real life, and the treatment impact on e.g. career chances. Therefore we assessed faces from dysgnathic patients (Class II, III and Laterognathia) on video clips. Class I faces served as anchor and controls. Each patient’s face was assessed twice before and after treatment in changing sequence, by 155 naïve raters with similar age to the patients. The raters provided independent estimates on aesthetic trait pairs like ugly /beautiful, and personality trait pairs like dominant /flexible. Furthermore the perception of attractiveness, intelligence, health, the persons’ erotic aura, faithfulness, and five additional items were rated. We estimated the significance of the perceived treatment related differences and the respective effect size by general linear models for repeated measures. The obtained results were comparable to our previous rating on still images. There was an overall trend, that faces in video clips are rated along common stereotypes to a lesser extent than photographs. We observed significant class differences and treatment related changes of most aesthetic traits (e.g. beauty, attractiveness), these were comparable to intelligence, erotic aura and to some extend healthy appearance. While some personality traits (e.g. faithfulness) did not differ between the classes and between baseline and after treatment, we found that the intervention significantly and effectively altered the perception of the personality trait self-confidence. The effect size was highest in Class III patients, smallest in Class II patients, and in between for patients with Laterognathia. All dysgnathic patients benefitted from orthognathic surgery. We conclude that motion can mitigate marked stereotypes but does not entirely offset the mostly negative perception of dysgnathic faces.
Article
Full-text available
We show that faces contain much more information about sexual orientation than can be perceived or interpreted by the human brain. We used deep neural networks to extract features from 35,326 facial images. These features were entered into a logistic regression aimed at classifying sexual orientation. Given a single facial image, a classifier could correctly distinguish between gay and heterosexual men in 81% of cases, and in 71% of cases for women. Human judges achieved much lower accuracy: 61% for men and 54% for women. The accuracy of the algorithm increased to 91% and 83%, respectively, given five facial images per person. Facial features employed by the classifier included both fixed (e.g., nose shape) and transient facial features (e.g., grooming style). Consistent with the prenatal hormone theory of sexual orientation, gay men and women tended to have gender-atypical facial morphology, expression, and grooming styles. Prediction models aimed at gender alone allowed for detecting gay males with 57% accuracy and gay females with 58% accuracy. Those findings advance our understanding of the origins of sexual orientation and the limits of human perception. Additionally, given that companies and governments are increasingly using computer vision algorithms to detect people’s intimate traits, our findings expose a threat to the privacy and safety of gay men and women.
Article
Full-text available
Automatically learned quality assessment for images has recently become a hot topic due to its usefulness in a wide variety of applications such as evaluating image capture pipelines, storage techniques and sharing media. Despite the subjective nature of this problem, most existing methods only predict the mean opinion score provided by datasets such as AVA [1] and TID2013 [2]. Our approach differs from others in that we predict the distribution of human opinion scores using a convolutional neural network. Our architecture also has the advantage of being significantly simpler than other methods with comparable performance. Our proposed approach relies on the success (and retraining) of proven, state-of-the-art deep object recognition networks. Our resulting network can be used to not only score images reliably and with high correlation to human perception, but also to assist with adaptation and optimization of photo editing/enhancement algorithms in a photographic pipeline. All this is done without need of a "golden" reference image, consequently allowing for single-image, semantic- and perceptually-aware, no-reference quality assessment.
Article
Full-text available
Significance Claims about how reported emotional experiences are geometrically organized within a semantic space have shaped the study of emotion. Using statistical methods to analyze reports of emotional states elicited by 2,185 emotionally evocative short videos with richly varying situational content, we uncovered 27 varieties of reported emotional experience. Reported experience is better captured by categories such as “amusement” than by ratings of widely measured affective dimensions such as valence and arousal. Although categories are found to organize dimensional appraisals in a coherent and powerful fashion, many categories are linked by smooth gradients, contrary to discrete theories. Our results comprise an approximation of a geometric structure of reported emotional experience.
Conference Paper
Full-text available
Understanding why a model makes a certain prediction can be as crucial as the prediction's accuracy in many applications. However, the highest accuracy for large modern datasets is often achieved by complex models that even experts struggle to interpret, such as ensemble or deep learning models, creating a tension between accuracy and interpretability. In response, various methods have recently been proposed to help users interpret the predictions of complex models, but it is often unclear how these methods are related and when one method is preferable over another. To address this problem, we present a unified framework for interpreting predictions, SHAP (SHapley Additive exPlanations). SHAP assigns each feature an importance value for a particular prediction. Its novel components include: (1) the identification of a new class of additive feature importance measures, and (2) theoretical results showing there is a unique solution in this class with a set of desirable properties. The new class unifies six existing methods, notable because several recent methods in the class lack the proposed desirable properties. Based on insights from this unification, we present new methods that show improved computational performance and/or better consistency with human intuition than previous approaches.
Article
Full-text available
The purported "black box"' nature of neural networks is a barrier to adoption in applications where interpretability is essential. Here we present DeepLIFT (Deep Learning Important FeaTures), a method for decomposing the output prediction of a neural network on a specific input by backpropagating the contributions of all neurons in the network to every feature of the input. DeepLIFT compares the activation of each neuron to its 'reference activation' and assigns contribution scores according to the difference. By optionally giving separate consideration to positive and negative contributions, DeepLIFT can also reveal dependencies which are missed by other approaches. Scores can be computed efficiently in a single backward pass. We apply DeepLIFT to models trained on MNIST and simulated genomic data, and show significant advantages over gradient-based methods. A detailed video tutorial on the method is at http://goo.gl/qKb7pL and code is at http://goo.gl/RM8jvH.
Article
Full-text available
In this work, we introduce a novel interpretation of residual networks showing they are exponential ensembles. This observation is supported by a large-scale lesion study that demonstrates they behave just like ensembles at test time. Subsequently, we perform an analysis showing these ensembles mostly consist of networks that are each relatively shallow. For example, contrary to our expectations, most of the gradient in a residual network with 110 layers comes from an ensemble of very short networks, i.e., only 10-34 layers deep. This suggests that in addition to describing neural networks in terms of width and depth, there is a third dimension: multiplicity, the size of the implicit ensemble. Ultimately, residual networks do not resolve the vanishing gradient problem by preserving gradient flow throughout the entire depth of the network - rather, they avoid the problem simply by ensembling many short networks together. This insight reveals that depth is still an open research question and invites the exploration of the related notion of multiplicity.
Article
Full-text available
We take historical stock of charisma, tracing its origins and how it has been conceptualized in the sociological and organizational sciences literatures. Although charisma has been intensely studied, the concept is still not well understood and much of the research undertaken cannot inform policy. We show that the major obstacles to advancing our understanding of charisma have included issues with its definition, its confusion with transformational leadership, the use of questionnaire measures, and that it has not been studied using correctly specified causal models. To help spawn a new genre of research on charisma, we use signaling theory to provide a general definition of charisma, and make suggestions about how charisma should be conceptualized, operationalized, and modeled. We also describe trends and patterns in articles we reviewed, using cocitation as well as bibliometric analyses, and discuss the practical implications of our findings.
Article
Full-text available
Most publicly available image quality databases have been created under highly controlled conditions by introducing graded simulated distortions onto high-quality photographs. However, images captured using typical real-world mobile camera devices are usually afflicted by complex mixtures of multiple distortions, which are not necessarily well-modeled by the synthetic distortions found in existing databases. The originators of existing legacy databases usually conducted human psychometric studies to obtain statistically meaningful sets of human opinion scores on images in a stringently controlled visual environment, resulting in small data collections relative to other kinds of image analysis databases. Towards overcoming these limitations, we designed and created a new database that we call the LIVE In the Wild Image Quality Challenge Database, which contains widely diverse authentic image distortions on a large number of images captured using a representative variety of modern mobile devices. We also designed and implemented a new online crowdsourcing system, which we have used to conduct a very large-scale, multi-month image quality assessment subjective study. Our database consists of over 350000 opinion scores on 1162 images evaluated by over 7000 unique human observers. Despite the lack of control over the experimental environments of the numerous study participants, we demonstrate excellent internal consistency of the subjective dataset. We also evaluate several top-performing blind Image Quality Assessment algorithms on it and present insights on how mixtures of distortions challenge both end users as well as automatic perceptual quality prediction models.
Conference Paper
Full-text available
In this paper we tackle the estimation of apparent age in still face images with deep learning. Our convolutional neural networks (CNNs) use the VGG-16 architecture and are pretrained on ImageNet for image classification. In addition, due to the limited number of apparent age annotated images, we explore the benefit of finetuning over crawled Internet face images with available age. We crawled 0.5 million images of celebrities from IMDB and Wikipedia that we make public. This is the largest public dataset for age prediction to date. We pose the age regression problem as a deep classification problem followed by a softmax expected value refinement and show improvements over direct regression training of CNNs. Our proposed method, Deep EXpectation (DEX) of apparent age, first detects the face in the test image and then extracts the CNN predictions from an ensemble of 20 networks on the cropped face. The CNNs of DEX were finetuned on the crawled images and then on the provided images with apparent age annotations. DEX does not use explicit facial landmarks. Our DEX is the winner (1st place) of the ChaLearn LAP 2015 challenge on apparent age estimation with 115 registered teams, significantly outperforming the human reference.
Article
Full-text available
Understanding and interpreting classification decisions of automated image classification systems is of high value in many applications as it allows to verify the reasoning of the system and provides additional information to the human expert. Although machine learning methods are solving very successfully a plethora of tasks, they have in most cases the disadvantage of acting as a black box, not providing any information about what made them arrive at a particular decision. This work proposes a general solution to the problem of understanding classification decisions by pixel-wise decomposition of non- linear classifiers. We introduce a methodology that allows to visualize the contributions of single pixels to predictions for kernel-based classifiers over Bag of Words features and for multilayered neural networks. These pixel contributions can be visualized as heatmaps and are provided to a human expert who can intuitively not only verify the validity of the classification decision, but also focus further analysis on regions of potential interest. We evaluate our method for classifiers trained on PASCAL VOC 2009 images, synthetic image data containing geometric shapes, the MNIST handwritten digits data set and for the pre-trained ImageNet model available as part of the Caffe open source package.
Article
Full-text available
This article reports the results of a study examining the relationships among strategic charismatic leadership, organizational performance, and environmental uncertainty with primary data from a sample of 128 CEOs of major U.S. corporations. Drawing on 770 surveys from top management team members in these companies, objective stock market and accounting data, and an objective measure of environmental uncertainty, we found that organizational performance was associated with subsequent perceptions of CEO charisma but that perceptions of CEO charisma were not associated with subsequent organizational performance, even after we incorporated the potential moderating effect of environmental uncertainty.
Article
Full-text available
This article shows that the joint effect of tie strength and image-impairment concern on negative word-of-mouth (NWOM) transmission is different for males and females and argues that this effect occurs because of differences in their relative concern for self versus others. For males, there was not a significant interaction between image-impairment concern and tie strength on NWOM transmission likelihood. In contrast, for females the effect of image-impairment concern on NWOM transmission likelihood was stronger for weak ties than for strong ties. The robustness of the findings were tested in two additional studies by directly manipulating relative concern for self versus others and by employing an indirect proxy: interdependent and independent self-construal. Self- versus other-focused thoughts mediated the joint effect on NWOM transmission.
Article
Full-text available
Researchers studying a range of psychological phenomena (e.g., theory of mind, emotion, stereotyping and prejudice, interpersonal attraction, etc.) sometimes employ photographs of people as stimuli. In this paper, we introduce the Chicago Face Database, a free resource consisting of 158 high-resolution, standardized photographs of Black and White males and females between the ages of 18 and 40 years and extensive data about these targets. In Study 1, we report pre-testing of these faces, which includes both subjective norming data and objective physical measurements of the images included in the database. In Study 2 we surveyed psychology researchers to assess the suitability of these targets for research purposes and explored factors that were associated with researchers' judgments of suitability. Instructions are outlined for those interested in obtaining access to the stimulus set and accompanying ratings and measures.
Article
Full-text available
Predicting face attributes from web images is challenging due to background clutters and face variations. A novel deep learning framework is proposed for face attribute prediction in the wild. It cascades two CNNs (LNet and ANet) for face localization and attribute prediction respectively. These nets are trained in a cascade manner with attribute labels, but pre-trained differently. LNet is pre-trained with massive general object categories, while ANet is pre-trained with massive face identities. This framework not only outperforms state-of-the-art with large margin, but also reveals multiple valuable facts on learning face representation as below. (1) It shows how LNet and ANet can be improved by different pre-training strategies. (2) It reveals that although filters of LNet are fine-tuned by attribute labels, their response maps over the entire image have strong indication of face's location. This fact enables training LNet for face localization with only attribute tags, but without face bounding boxes (which are required by all detection works). With a novel fast feed-forward scheme, the cascade of LNet and ANet can localize faces and recognize attributes in images with arbitrary sizes in real time. (3) It also demonstrates that the high-level hidden neurons of ANet automatically discover semantic concepts after pre-training, and such concepts are significantly enriched after fine-tuning. Each attribute can be well explained by a sparse linear combination of these concepts. By analyzing such combinations, attributes show clear grouping patterns, which could be well interpreted semantically.
Article
The inherent nature of social media content poses serious challenges to practical applications of sentiment analysis. We present VADER, a simple rule-based model for general sentiment analysis, and compare its effectiveness to eleven typical state-of-practice benchmarks including LIWC, ANEW, the General Inquirer, SentiWordNet, and machine learning oriented techniques relying on Naive Bayes, Maximum Entropy, and Support Vector Machine (SVM) algorithms. Using a combination of qualitative and quantitative methods, we first construct and empirically validate a gold-standard list of lexical features (along with their associated sentiment intensity measures) which are specifically attuned to sentiment in microblog-like contexts. We then combine these lexical features with consideration for five general rules that embody grammatical and syntactical conventions for expressing and emphasizing sentiment intensity. Interestingly, using our parsimonious rule-based model to assess the sentiment of tweets, we find that VADER outperforms individual human raters (F1 Classification Accuracy = 0.96 and 0.84, respectively), and generalizes more favorably across contexts than any of our benchmarks.
Conference Paper
Face recognition constitutes a relatively a popular area which has emerged from the rulers of the social media to top universities in the world. Those frontiers and rule makers recently designed deep learning based custom face recognition models. A modern face recognition pipeline consists of four common stages: detecting, alignment, representation and verification. However, face recognition studies mainly mention the representation stage of a pipeline. In this paper, first of all a review face recognition has been done and then the description of the developed lightweight hybrid high performance face recognition framework has been made. Its hybrid feature enables to switch face recognition models among state-of-the-art ones.
Article
Are social media posts with pictures more popular than those without? Why do pictures with certain characteristics induce higher engagement than some other pictures? Using data sets of social media posts about major airlines and sport utility vehicle brands collected from Twitter and Instagram, the authors empirically examine the influence of image content on social media engagement. After accounting for selection bias on the inclusion of image content, the authors find a significant and robust positive mere presence effect of image content on user engagement in both product categories on Twitter. They also find that high-quality and professionally shot pictures consistently lead to higher engagement on both platforms for both product categories. However, the effect of colorfulness varies by product category, while the presence of human face and image–text fit can induce higher user engagement on Twitter but not on Instagram. These findings shed light on how to improve social media engagement using image content.
Article
Research Summary We demonstrate how a novel synthesis of three methods—(a) unsupervised topic modeling of text data to generate new measures of textual variance, (b) sentiment analysis of text data, and (c) supervised ML coding of facial images with a cutting‐edge convolutional neural network algorithm—can shed light on questions related to CEO oral communication. With videos and corresponding transcripts of interviews with emerging market CEOs, we use this synthesis of methods to discover five distinct communication styles that incorporate both verbal and nonverbal aspects of communication. Our data comprises interviews that represent unedited expressions and content, making them especially suitable as data sources for the measurement of an individual's communication style. We then perform a proof‐of‐concept analysis, correlating CEO communication styles to M&A outcomes, highlighting the value of combining text and videographic data to define styles. We also discuss the benefits of using our methods versus current research methods. Managerial Summary CEOs spend most of their time communicating to investors, customers, and partners with the aim of influencing these various stakeholders. To what extent though does their effectiveness as leaders depend on a mixture of what they say and how they say it? We use cutting‐edge machine learning approaches to measure a CEO's communication style, which can give clues about the major strategic decisions a CEO's firm must make. With a collection of video interviews with 61 organizational leaders from emerging markets, we use textual analysis and facial image expression recognition to code whether CEOs are “excitable,” “stern,” “dramatic,” “rambling,” and “melancholy” in their communication styles. As a proof‐of‐concept, we also show that CEOs who were more dramatic in expressing themselves were also less likely to oversee major acquisitions. Therefore, not only can CEO communication styles help predict a firm's ability to grow, adapt to change, and reallocate existing assets, styles can also be coded more intuitively by using our new method, representing a vast improvement over previous methods in both accessibility and interpretability.
Article
Consumer-to-consumer brand message sharing is pivotal for effective social media marketing. Even as companies join social media conversations and generate millions of brand messages, it remains unclear what, how, and when brand messages stand out and prompt sharing by consumers. With a conceptual extension of speech act theory, this study offers a granular assessment of brands’ message intentions (i.e., assertive, expressive, or directive) and the effects on consumer sharing. A text mining study of more than two years of Facebook posts and Twitter tweets by well-known consumer brands empirically demonstrates the impacts of distinct message intentions on consumers’ message sharing. Specifically, the use of rhetorical styles (alliteration and repetitions) and cross-message compositions enhance consumer message sharing. As a further extension, an image-based study demonstrates that the presence of visuals, or so-called image acts, increases the ability to account for message sharing. The findings explicate brand message sharing by consumers and thus offer guidance to content managers for developing more effective conversational strategies in social media marketing.
Article
Facial beauty prediction (FBP) is a significant visual recognition problem to make assessment of facial attractiveness that is consistent to human perception. To tackle this problem, various data-driven models, especially state-of-the-art deep learning techniques, were introduced, and benchmark dataset become one of the essential elements to achieve FBP. Previous works have formulated the recognition of facial beauty as a specific supervised learning problem of classification, regression or ranking, which indicates that FBP is intrinsically a computation problem with multiple paradigms. However, most of FBP benchmark datasets were built under specific computation constrains, which limits the performance and flexibility of the computational model trained on the dataset. In this paper, we argue that FBP is a multi-paradigm computation problem, and propose a new diverse benchmark dataset, called SCUT-FBP5500, to achieve multi-paradigm facial beauty prediction. The SCUT-FBP5500 dataset has totally 5500 frontal faces with diverse properties (male/female, Asian/Caucasian, ages) and diverse labels (face landmarks, beauty scores within [1,~5], beauty score distribution), which allows different computational models with different FBP paradigms, such as appearance-based/shape-based facial beauty classification/regression model for male/female of Asian/Caucasian. We evaluated the SCUT-FBP5500 dataset for FBP using different combinations of feature and predictor, and various deep learning methods. The results indicates the improvement of FBP and the potential applications based on the SCUT-FBP5500.
Article
This study adopted a two (author: algorithm or journalist) by two (notification of author: real or inverse) between subject design to investigate how the public and journalists perceive the quality of algorithms-written articles compared with human journalist’s work. Findings showed that both the public and journalists’ evaluations were varied by the manipulation of author notification. That is, the public gave higher scores to the algorithm’s work when it was notified as the real author, but they gave lower scores to the algorithm’s work when the author was notified as a journalist. It confirmed the public’s negative attitude toward journalists’ credibility and craving for new information and communication technology (ICT) products/services in Korea. Based on journalists’ resistance to change and innovation and the theory of prejudice, it was expected that journalists would be favorable to another journalist’s work and unfavorable to an algorithm’s work. However, contrary to the hypothetical expectation, journalists also gave higher scores to an algorithm’s work and lower scores to a journalist’s work. Implications relating to the intrusion of algorithm-written articles into journalism were discussed.
Conference Paper
Selfies are now a global phenomenon. This massive number of self-portrait images taken and shared on social media is revolutionizing the way people introduce themselves and the circle of their friends to the world. While taking photos of oneself can be seen simply as recording personal memories, the urge to share them with other people adds an exclusive sensation to the selfies. Due to the Big Data nature of selfies, it is nearly impossible to analyze them manually. In this paper, we provide, to the best of our knowledge, the first selfie dataset for research purposes with more than 46,000 images. We address interesting questions about selfies, including how appearance of certain objects, concepts and attributes influences the popularity of selfies. We also study the correlation between popularity and sentiment in selfie images. In a nutshell, from a large scale dataset, we automatically infer what makes a selfie a good selfie. We believe that this research creates new opportunities for social, psychological and behavioral scientists to study selfies from a large scale point of view, a perspective that best fits the nature of the selfie phenomenon.
Conference Paper
Tree boosting is a highly effective and widely used machine learning method. In this paper, we describe a scalable end-to-end tree boosting system called XGBoost, which is used widely by data scientists to achieve state-of-the-art results on many machine learning challenges. We propose a novel sparsity-aware algorithm for sparse data and weighted quantile sketch for approximate tree learning. More importantly, we provide insights on cache access patterns, data compression and sharding to build a scalable tree boosting system. By combining these insights, XGBoost scales beyond billions of examples using far fewer resources than existing systems.
Article
Article
n everyday language, charisma has been used to refer to a broad spectrum of human relations, from romantic love to admiration for certain extraordinary abilities of public figures. Historically, the term has been used to describe “a gift of divine grace,” an attribute of magical magnetism and the power of a religious or secular leader as seen by that person’s followers. Over time, the term became detached from its religious origin although the importance of this source is still acknowledged in present-day research on the term. The present, structural analysis observes the knowledge growth in charisma research, as well as disciplinary shifts that the term charisma has undergone over time. The results of the analyses show that the focus of charisma studies has shifted from the emotional (religious) to the socio-political arena and from there to the management domain. As the amount of knowledge on charisma grew, research on this subject evolved into a mixture of emotional (psychology) and rational (business and management) research issues. In the new millennium, though, charisma studies have been characterized mostly by the management/business domain. It is evident from this study that although charisma is a contextual phenomenon, it still permits multidisciplinary treatment. In addition to bibliometric data on the current state of research on the subject, several indicators of future research directions and interest are presented. © 2015, Springer Science+Business Media New York.
Article
Human faces are used extensively in print advertisements. In prior literature, researchers have studied spokespersons in general, but few have studied faces explicitly. This paper aims to answer three questions that are important to both researchers and practitioners: (1) Do faces affect how a viewer reacts to an advertisement on the metrics that advertisers care about? (2) If faces do have an effect, is it large enough to warrant careful selection of faces when constructing print advertisements? (3) If faces do have an effect and the effect is large, what facial features elicit such differential reactions on these metrics, and are such reactions different across individuals and/or product categories? Relying on the eigenface method, a holistic approach widely used in the computer science field for face recognition, we conducted an empirical study to answer these three questions. The results show that different faces do have an effect on people's attitude toward the advertisement, attitude toward the brand, and purchase intention and that the effect is nontrivial. Multiple segments were identified and substantial differences were found among people's reactions to the faces in the ads across those segments. We also found that the effect of faces interacts with product categories and is mediated by various facial traits such as attractiveness, trustworthiness, and competence. Implications and directions for future research are discussed.
Article
Langlois and her colleagues reported in this journal that composite faces are more attractive than the component faces used to create them, and conjectured that averageness is attractive (Langlois & Roggman, 1990; Langlois, Roggman, & Musselman, 1994). However, extremes may also be attractive (Perrett, May, & Yoshikawa, 1994). We investigated the effect of averageness (proximity to a norm or average face) on attractiveness using a computerized caricature generator to vary averageness. Attractiveness increased with averageness (Experiment 1) and was negatively correlated with distinctiveness, a subjective measure of the converse of averageness (Experiments 1 and 2). Extremes (caricatures) were not attractive. Line-drawing composites, which avoid some of the problems associated with gray-level composites, were significantly more attractive and less distinctive (more average) than individual faces (Experiment 2). These results support the claim that averageness is attractive.
Article
We argue in this paper that in an age of complexity, change, large enterprises, and nation states, leaders are more important than ever. However, their effectiveness depends on their personality and charisma and not solely on their control over bureaucratic structures. We used a study of U.S. presidents to test a general model of leader effectiveness that includes leader personality characteristics, charisma, crises, age of the institution headed by the leader, and leader effectiveness. Age of the presidency accounted for approximately 20 percent of the variance in presidential needs for power, achievement, and affiliation. Presidential needs and a measure of leader self-restraint in using power, the age of the presidency, and crises accounted for 24 percent of the variance in presidential charisma. Age of the presidency, crises, needs, and charisma together predicted from 25 percent to 66 percent of the variance in five measures of presidential performance. Our study demonstrates that personality and charisma do make a difference.
Article
In a longitudinal, randomized field experiment, we tested the impact of transformational leadership, enhanced by training, on follower development and performance. Experimental group leaders received transformational leadership training, and control group leaders, eclectic leadership training. The sample included 54 military leaders, their 90 direct followers, and 724 indirect followers. Results indicated the leaders in the experimental group had a more positive impact on direct followers' development and on indirect followers' performance than did the leaders in the control group.
Article
We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 dif-ferent classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make train-ing faster, we used non-saturating neurons and a very efficient GPU implemen-tation of the convolution operation. To reduce overfitting in the fully-connected layers we employed a recently-developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.