Cristina Segalin

Conversationally-inspired stylometric features for authorship attribution in instant messaging

Authorship attribution (AA) aims at recognizing automatically the author of a given text sample. Traditionally applied to literary texts, AA faces now the new challenge of recognizing the identity of people involved in chat conversations. These share many aspects with spoken conversations, but AA approaches did not take it into account so far. Hence, this paper tries to fill the gap and proposes two novelties that improve the effectiveness of traditional AA approaches for this type of data: the first is to adopt features inspired by Conversation Analysis (in particular for turn-taking), the second is to extract the features from individual turns rather than from entire conversations. The experiments have been performed over a corpus of dyadic chat conversations (77 individuals in total). The performance in identifying the persons involved in each exchange, measured in terms of area under the Cumulative Match Characteristic curve, is 89.5%.

Faved! Biometrics: Tell Me Which Image You Like and I'll Tell You Who You Are

This paper builds upon the belief that every human being has a built-in image aesthetic evaluation system. This sort of personal aesthetics mostly follows certain aesthetic rules widely studied in image aesthetics (e.g., rules of thirds, colorfulness, etc.), though it likely contains some innate, unique preferences. This paper is a proof of concept of this intuition, presenting personal aesthetics as a novel behavioral biometrical trait. In our scenario, personal aesthetics activate when an individual is presented with a set of photos he may like or dislike. The goal is to distill and encode the uniqueness of his visual preferences into a compact template. To this aim, we extract a pool of low- and high-level state-of-the-art image features from a set of Flickr images preferred by a user, feeding them successively into a LASSO regressor. LASSO highlights the most discriminant cues for the individual, allowing authentication and recognition tasks. The results are surprising given only 1 image as test. We can match the user identity against a gallery of 200 individuals definitely much better than chance. Using 20 images (all preferred by a single user) as a biometrical trait, we reach an AUC of 96%, considering the cumulative matching characteristic curve. Extensive experiments also support the interpretability of our approach, effectively modeling what is the “what we like” that distinguishes us from others.

The Pictures We Like Are Our Image: Continuous Mapping of Favorite Pictures into Self-Assessed and Attributed Personality Traits

Flickr allows its users to tag the pictures they like as “favorite”. As a result, many users of the popular photo-sharing platform produce galleries of favorite pictures. This article proposes new approaches, based on Computational Aesthetics, capable to infer the personality traits of Flickr users from the galleries above. In particular, the approaches map low-level features extracted from the pictures into numerical scores corresponding to the Big-Five Traits, both self-assessed and attributed. The experiments were performed over 60,000 pictures tagged as favorite by 300 users (the PsychoFlickr Corpus). The results show that it is possible to predict beyond chance both self-assessed and attributed traits. In line with the state-of-the-art of Personality Computing, these latter are predicted with higher effectiveness (correlation up to 0.68 between actual and predicted traits).

Reading between the turns: Statistical modeling for identity recognition and verification in chats

Identity safekeeping has recently become an important problem for the social web: as a case study, we focus here on instant messaging platforms, proposing novel soft-biometric cues for user recognition and verification. Specifically, we design a set of features encoding effectively how a person converses: since chats are crossbreeds of written text and face-to-face verbal communication, the features inherit equally from textual authorship attribution and conversational analysis of speech. Importantly, our cues ignore completely the semantics of the chat, relying solely on non-verbal aspects, taking care of possible privacy and ethical issues. We apply our approach on a novel dataset of 94 different individuals, whose chat conversations have been recorded for an average period of five months; recognition rate, intended as normalized AUC on CMC curve, is 95.73%, while verification rate amounts to 95.66%, as normalized AUC on ROC curve.

Social profiling through image understanding

Linking OCEAN personality traits and preferred images in the Flickr social network.Classification of personality traits by novel image features, designed by CNN.Interpretation of visual features by an ad-hoc deconvolution strategy.Online demo application. The role of images in the last ten years has changed radically due to the advent of social networks: from media objects mainly used to communicate visual information, images have become personal, associated with the people that create or interact with them (for example, giving a like). Therefore, in the same way that a post reveals something of its author, so now the images associated to a person may embed some of her individual characteristics, such as her personality traits. In this paper, we explore this new level of image understanding with the ultimate goal of relating a set of image preferences to personality traits by using a deep learning framework. In particular, our problem focuses on inferring both self-assessed (how the personality traits of a person can be guessed from her preferred image) and attributed traits (what impressions in terms of personality traits these images trigger in unacquainted people), learning a sort of wisdom of the crowds. Our characterization of each image is locked within the layers of a CNN, allowing us to discover more entangled attributes (aesthetic patterns and semantic information) and to better generalize the patterns that identify a trait. The experimental results show that the proposed method outperforms state-of-the-art results and captures what visually characterizes a certain trait: using a deconvolution strategy we found a clear distinction of features, patterns and content between low and high values in a given trait.

What your Facebook Profile Picture Reveals about your Personality

People spend considerable effort managing the impressions they give others. Social psychologists have shown that people manage these impressions differently depending upon their personality. Facebook and other social media provide a new forum for this fundamental process; hence, understanding peoples behaviour on social media could provide interesting insights on their personality. In this paper we investigate automatic personality recognition from Facebook profile pictures. We analyze the effectiveness of four families of visual features and we discuss some human interpretable patterns that explain the personality traits of the individuals. For example, extroverts and agreeable individuals tend to have warm colored pictures and to exhibit many faces in their portraits, mirroring their inclination to socialize; while neurotic ones have a prevalence of pictures of indoor places. Then, we propose a classification approach to automatically recognize personality traits from these visual features. Finally, we compare the performance of our classification approach to the one obtained by human raters and we show that computer-based classifications are significantly more accurate than averaged human-based classifications for Extraversion and Neuroticism.

Biometrics on visual preferences: A “pump and distill” regression approach

We present a statistical behavioural biometric approach for recognizing people by their aesthetic preferences, using colour images. In the enrollment phase, a model is learnt for each user, using a training set of preferred images. In the recognition/authentication phase, such model is tested with an unseen set of pictures preferred by a probe subject. The approach is dubbed “pump and distill”, since the training set of each user is pumped by bagging, producing a set of image ensembles. In the distill step, each ensemble is reduced into a set of surrogates, that is, aggregates of images sharing a similar visual content. Finally, LASSO regression is performed on these surrogates; the resulting regressor, employed as a classifier, takes test images belonging to a single user, predicting his identity. The approach improves the state-of-the-art on recognition and authentication tasks in average, on a dataset of 40000 Flickr images and 200 users. In practice, given a pool of 20 preferred images of a user, the approach recognizes his identity with an accuracy of 92%, and sets an authentication accuracy of 91% in terms of normalized Area Under the Curve of the CMC and ROC curve, respectively.

Statistical Analysis of Visual Attentional Patterns for Video Surveillance

We show that the way people observe video sequences, other than what they observe, is important for the understanding and the prediction of human activities. In this study, we consider 36 surveillance videos, organized in four categories (confront, nothing, fight, play): the videos are observed by 19 people, ten of them are experienced operators and the other nine are novices, and the gaze trajectories of both populations are recorded by an eye tracking device. Due to the proved superior ability of experienced operators in predicting violence in surveillance footage, our aim is to distinguish the two classes of people, highlighting in which respect expert operators differ from novices. Extracting spatio-temporal features from the eye tracking data, and training standard machine learning classifiers, we are able to discriminate the two groups of subjects with an average accuracy of 80.26%. The idea is that expert operators are more focused on few regions of the scene, sampling them with high frequency and low predictability. This can be thought as a first step toward the advanced automated analysis of video surveillance footage, where machines imitate as best as possible the attentive mechanisms of humans.

Recognizing People by Their Personal Aesthetics: A Statistical Multi-level Approach

This paper presents a study on personal aesthetics, a recent soft biometrics application where the goal is to recognize people by considering the images they like. Here we propose a multi-level approach, where each level is intended as a low-dimensional space where the images preferred by a user can be projected, and similar images are mapped nearby, namely a Counting Grid. Multiple levels are generated by adopting Counting Grids at different resolutions, corresponding to analyze images at different grains. Each level is then associated to an exemplar Support Vector Machine, which separates the images of an individual from the rest of the users. Putting together multiple levels gives a battery of classifiers whose performances are very good: on a dataset of 200 users, and 40 K images, using 5 preferred images as biometric template gives 97 % of probability of guessing the correct user; as for the verification capability, the equal error rate is 0.11. The approach has also been tested with diverse comparative methods and different features, showing that color image properties are crucial to encode the personal aesthetics, and that high-level information (as the objects within the images) could be very effective, but current methods are not robust enough to catch it.

The expressivity of turn-taking: Understanding children pragmatics by hybrid classifiers

We analyze the effect of children age on pragmatic skills, i.e. on the way children manage the conversation dynamics. In particular, we focus exclusively on the turn-taking (who talks when and how much), reducing conversations as sequences of simple speech/silence periods. Employing a hybrid (generative + discriminative) classification framework, we demonstrate that such a simple signature is very informative, allowing to separate 22 “pre-School” conversations (between 3-4 years old children) and 24 “School” conversations (between 6-8 years old children) subjects, with 78% of accuracy. The framework exploits Steady Conversational Periods and Observed Influence Models as feature extractors, plus LASSO regression as feature selector and classifier. The generative nature of our method permits, as byproduct, to identify the pragmatic skills that better discriminate the two groups: no-tably, scholar children tend to have more frequent periods of sustained conversation, in a statistically significant way.