Site Agnostic Approach to Early Detection of Cyberbullying on Social Media Networks.

The rise in the use of social media networks has increased the prevalence of cyberbullying, and time is paramount to reduce the negative effects that derive from those behaviours on any social media platform. This paper aims to study the early detection problem from a general perspective by carrying out experiments over two independent datasets (Instagram and Vine), exclusively using users' comments. We used textual information from comments over baseline early detection models (fixed, threshold, and dual models) to apply three different methods of improving early detection. First, we evaluated the performance of Doc2Vec features. Finally, we also presented multiple instance learning (MIL) on early detection models and we assessed its performance. We applied timeawareprecision (TaP) as an early detection metric to asses the performance of the presented methods. We conclude that the inclusion of Doc2Vec features improves the performance of baseline early detection models by up to 79.6%. Moreover, multiple instance learning shows an important positive effect for the Vine dataset, where smaller post sizes and less use of the English language are present, with a further improvement of up to 13%, but no significant enhancement is shown for the Instagram dataset.

Annotated Dataset for Anomaly Detection in a Data Center with IoT Sensors.

The relative simplicity of IoT networks extends service vulnerabilities and possibilities to different network failures exhibiting system weaknesses. Therefore, having a dataset with a sufficient number of samples, labeled and with a systematic analysis, is essential in order to understand how these networks behave and detect traffic anomalies. This work presents DAD: a complete and labeled IoT dataset containing a reproduction of certain real-world behaviors as seen from the network. To approximate the dataset to a real environment, the data were obtained from a physical data center, with temperature sensors based on NFC smart passive sensor technology. Having carried out different approaches, performing mathematical modeling using time series was finally chosen. The virtual infrastructure necessary for the creation of the dataset is formed by five virtual machines, a MQTT broker and four client nodes, each of them with four sensors of the refrigeration units connected to the internal IoT network. DAD presents a seven day network activity with three types of anomalies: duplication, interception and modification on the MQTT message, spread over 5 days. Finally, a feature description is performed, so it can be used for the application of the various techniques of prediction or automatic classification.

High Order Profile Expansion to tackle the new user problem on recommender systems.

Collaborative Filtering algorithms provide users with recommendations based on their opinions, that is, on the ratings given by the user for some items. They are the most popular and widely implemented algorithms in Recommender Systems, especially in e-commerce, considering their good results. However, when the information is extremely sparse, independently of the domain nature, they do not present such good results. In particular, it is difficult to offer recommendations which are accurate enough to a user who has just arrived to a system or who has rated few items. This is the well-known new user problem, a type of cold-start. Profile Expansion techniques had been already presented as a method to alleviate this situation. These techniques increase the size of the user profile, by obtaining information about user tastes in distinct ways. Therefore, recommender algorithms have more information at their disposal, and results improve. In this paper, we present the High Order Profile Expansion techniques, which combine in different ways the Profile Expansion methods. The results show 110% improvement in precision over the algorithm without Profile Expansion, and 10% improvement over Profile Expansion techniques.

Early Detection of Depression: Social Network Analysis and Random Forest Techniques.

BACKGROUND: Major depressive disorder (MDD) or depression is among the most prevalent psychiatric disorders, affecting more than 300 million people globally. Early detection is critical for rapid intervention, which can potentially reduce the escalation of the disorder. OBJECTIVE: This study used data from social media networks to explore various methods of early detection of MDDs based on machine learning. We performed a thorough analysis of the dataset to characterize the subjects' behavior based on different aspects of their writings: textual spreading, time gap, and time span. METHODS: We proposed 2 different approaches based on machine learning singleton and dual. The former uses 1 random forest (RF) classifier with 2 threshold functions, whereas the latter uses 2 independent RF classifiers, one to detect depressed subjects and another to identify nondepressed individuals. In both cases, features are defined from textual, semantic, and writing similarities. RESULTS: The evaluation follows a time-aware approach that rewards early detections and penalizes late detections. The results show how a dual model performs significantly better than the singleton model and is able to improve current state-of-the-art detection models by more than 10%. CONCLUSIONS: Given the results, we consider that this study can help in the development of new solutions to deal with the early detection of depression on social networks.

Twitter: a good place to detect health conditions.

With the proliferation of social networks and blogs, the Internet is increasingly being used to disseminate personal health information rather than just as a source of information. In this paper we exploit the wealth of user-generated data, available through the micro-blogging service Twitter, to estimate and track the incidence of health conditions in society. The method is based on two stages: we start by extracting possibly relevant tweets using a set of specially crafted regular expressions, and then classify these initial messages using machine learning methods. Furthermore, we selected relevant features to improve the results and the execution times. To test the method, we considered four health states or conditions, namely flu, depression, pregnancy and eating disorders, and two locations, Portugal and Spain. We present the results obtained and demonstrate that the detection results and the performance of the method are improved after feature selection. The results are promising, with areas under the receiver operating characteristic curve between 0.7 and 0.9, and f-measure values around 0.8 and 0.9. This fact indicates that such approach provides a feasible solution for measuring and tracking the evolution of health states within the society.