This dissertation hypothesizes that the key to automated veracity assessment of data from open sources is the careful estimation and extraction of relevant features and indicators. These features and indicators provide added value to a quantifiable veracity assessment, either directly or indirectly. The importance and usefulness of a veracity assessment largely depend on the specific situation and reason for which it is being conducted. Factors such as the recipient of the veracity assessment, the scope of the assessment, and the metrics used to measure accuracy and performance, all play a role in determining the value and perceived quality of the assessment.

Five peer-reviewed publications; two journal articles, two conference articles, and one workshop article, are included in this compilation thesis.

The main contributions of the work presented in this dissertation are: i) a compilation of challenges with manual methods of veracity assessment, ii) a road map for addressing the identified challenges, iii) identification of the state-of-the-art and gap analysis of veracity assessment of open-source data, iv) exploration of indicators such as topic geo-location tracking over time and stance classification, and v) evaluation of various feature types, model transferability, and style obfuscation attacks and the impact on accuracy for automated veracity assessment of a type of deception: fake reviews.

Denna avhandling har som hypotes att nyckeln till automatiserad trovärdighetsbedömning av data från öppna källor ligger i det noggranna urvalet och estimeringen av relevanta särdrag och indikatorer. Dessa särdrag och indikatorer ger ett direkt eller indirekt mervärde till en kvantifierbar trovärdighetsbedömning. Betydelsen och användbarheten av en trovärdighetsbedömning beror till stor del på den specifika kontexten och anledningen till att den genomförs. Faktorer som mottagaren av trovärdighetsbedömningen, omfattningen av bedömningen och de mått som används för att mäta noggrannhet och prestanda, spelar alla in för att bestämma värdet och den upplevda kvalitén på bedömningen.

Fem referentgranskade publikationer ingår i denna sammanläggningsavhandling; två tidskriftsartiklar, två konferensartiklar och en workshopartikel.

De huvudsakliga bidragen från arbetet som presenteras i denna avhandling är: i) en sammanställning av utmaningar relaterade till manuella metoder för trovärdighetsbedömning, ii) en plan för att ta itu med de identifierade utmaningarna, iii) identifiering av forskningsfronten och en gapanalys av trovärdighetsbedömning av data från öppna källor, iv) studie av indikatorer såsom geolokalisering av ämnen och spårning av dem över tid samt klassificering av individers reaktioner i inlägg på sociala medier, och v) en utvärdering av särdragstyper som påverkar noggrannheten för automatisk trovärdighetsbedömning applicerat på en typ av bedrägeri: falska recensioner.

Remote photoplethysmography (rPPG) is a technique that aims to remotely estimate the heart rate of an individual using an RGB camera. Although several studies use the rPPG methodology, it is usually applied in a laboratory in a controlled environment, where both the camera and the subject are static, and the illumination is ideal for the task. However, applying rPPG in a real-life scenario is much more demanding, since dynamic illumination issues arise. The work presented in this paper introduces a framework to estimate the heart rate of an individual in real-time using an RGB camera in a situation characterized by dynamic illumination. Such situations occur, for example, when either the camera or the subject is moving, and/or the face visibility is limited. The framework uses a face detection program to extract regions of interest on an individual's face. These regions are combined and constitute the input to a convolutional neural network, which is trained to estimate the heart rate in real-time. The method is evaluated on three publicly available datasets, and an in-house dataset specifically collected for the purpose of this study, that includes motions and dynamic illumination. The method shows good performance on all four datasets, outperforming other methods.

Fake news, malicious rumors, fabricated reviews, generated images and videos, are today spread at an unprecedented rate, making the task of manually assessing data veracity for decision-making purposes a daunting task. Hence, it is urgent to explore possibilities to perform automatic veracity assessment. In this work we review the literature in search for methods and techniques representing state of the art with regard to computerized veracity assessment. We study what others have done within the area of veracity assessment, especially targeted towards social media and open source data, to understand research trends and determine needs for future research. The most common veracity assessment method among the studied set of papers is to perform text analysis using supervised learning. Regarding methods for machine learning much has happened in the last couple of years related to the advancements made in deep learning. However, very few papers make use of these advancements. Also, the papers in general tend to have a narrow scope, as they focus on solving a small task with only one type of data from one main source. The overall veracity assessment problem is complex, requiring a combination of data sources, data types, indicators, and methods. Only a few papers take on such a broad scope, thus, demonstrating the relative immaturity of the veracity assessment domain.

The temptation to influence and sway public opinion most certainly increases with the growth of open online forums where anyone anonymously can express their views and opinions. Since online review sites are a popular venue for opinion influencing attacks, there is a need to automatically identify deceptive posts. The main focus of this work is on automatic identification of deceptive reviews, both positive and negative biased. With this objective, we build a deceptive review SVM based classification model and explore the performance impact of using different feature types (TF-IDF, word2vec, PCFG). Moreover, we study the transferability of trained classification models applied to review data sets of other types of products, and, the classifier robustness, i.e., the accuracy impact, against attacks by stylometry obfuscation trough machine translation. Our findings show that i) we achieve an accuracy of over 90% using different feature types, ii) the trained classification models do not perform well when applied on other data sets containing reviews of different products, and iii) machine translation only slightly impacts the results and can not be used as a viable attack method. 

For the competition SemEval-2017 we investigated the possibility of performing stance classification (support, deny, query or comment) for messages in Twitter conversation threads related to rumours. Stance classification is interesting since it can provide a basis for rumour veracity assessment. Our ensemble classification approach of combining convolutional neural networks with both automatic rule mining and manually written rules achieved a final accuracy of 74.9% on the competition's test data set for Task 8A. To improve classification we also experimented with data relabeling and using the grammatical structure of the tweet contents for classification.

Tracking how discussion topics evolve in social media and where these topics are discussed geographically over time has the potential to provide useful information for many different purposes. In crisis management, knowing a specific topic’s current geographical location could provide vital information to where, or even which, resources should be allocated. This paper describes an attempt to track online discussions geographically over time. A distributed geo-aware streaming latent Dirichlet allocation model was developed for the purpose of recognizing topics’ locations in unstructured text. To evaluate the model it has been implemented and used for automatic discovery and geographical tracking of election topics during parts of the 2016 American presidential primary elections. It was shown that the locations correlated with the actual election locations, and that the model provides a better geolocation classification compared to using a keyword-based approach.

Intelligence analysis is dependent on veracity assessment of Open Source Information (OSINF) which includes assessment of the reliability of sources and credibility of information. Traditionally, OSINF veracity assessment is done by intelligence analysts manually, but the large volumes, high velocity, and variety make it infeasible to continue doing so, and calls for automation. Based on meetings, interviews and questionnaires with military personnel, analysis of related work and state of the art, we identify the challenges and propose an approach and a corresponding framework for automated veracity assessment of OSINF. The framework provides a basis for new tools which will give the intelligence analysts the ability to automatically or semi-automatically assess veracity of larger amounts of data in a shorter amount of time. Instead of spending their time working with irrelevant, ambiguous, contradicting, biased, or plain wrong data, they can spend more time on analysis.