Despoina Chatzakou

Mean Birds: Detecting Aggression and Bullying on Twitter

In recent years, bullying and aggression against social media users have grown significantly, causing serious consequences to victims of all demographics. Nowadays, cyberbullying affects more than half of young social media users worldwide, suffering from prolonged and/or coordinated digital harassment. Also, tools and technologies geared to understand and mitigate it are scarce and mostly ineffective. In this paper, we present a principled and scalable approach to detect bullying and aggressive behavior on Twitter. We propose a robust methodology for extracting text, user, and network-based attributes, studying the properties of bullies and aggressors, and what features distinguish them from regular users. We find that bullies post less, participate in fewer online communities, and are less popular than normal users. Aggressors are relatively popular and tend to include more negativity in their posts. We evaluate our methodology using a corpus of 1.6M tweets posted over 3 months, and show that machine learning classification algorithms can accurately detect users exhibiting bullying and aggressive behavior, with over 90% AUC.

Retweeting Activity on Twitter: Signs of Deception

Given the re-broadcasts (i.e. retweets) of posts in Twitter, how can we spot fake from genuine user reactions? What will be the tell-tale sign — the connectivity of retweeters, their relative timing, or something else? High retweet activity indicates influential users, and can be monetized. Hence, there are strong incentives for fraudulent users to artificially boost their retweets’ volume. Here, we explore the identification of fraudulent and genuine retweet threads. Our main contributions are: (a) the discovery of patterns that fraudulent activity seems to follow (the “triangles ” and “homogeneity ” patterns, the formation of micro-clusters in appropriate feature spaces); and (b) “RTGen ”, a realistic generator that mimics the behaviors of both honest and fraudulent users. We present experiments on a dataset of more than 6 million retweets crawled from Twitter.

ND-SYNC: Detecting Synchronized Fraud Activities

Given the retweeting activity for the posts of several Twitter users, how can we distinguish organic activity from spammy retweets by paid followers to boost a post’s appearance of popularity? More generally, given groups of observations, can we spot strange groups? Our main intuition is that organic behavior has more variability, while fraudulent behavior, like retweets by botnet members, is more synchronized. We refer to the detection of such synchronized observations as the Synchonization Fraud problem, and we study a specific instance of it, Retweet Fraud Detection, manifested in Twitter. Here, we propose: (A) ND-Sync, an efficient method for detecting group fraud, and (B) a set of carefully designed features for characterizing retweet threads. ND-Sync is effective in spotting retweet fraudsters, robust to different types of abnormal activity, and adaptable as it can easily incorporate additional features. Our method achieves a 97% accuracy on a real dataset of 12 million retweets crawled from Twitter.

Community Detection in Social Media by Leveraging Interactions and Intensities

Communities’ identification in topic-focused social media users interaction networks can offer improved understanding of different opinions and interest expressed on a topic. In this paper we present a community detection approach for user interaction networks which exploits both their structural properties and intensity patterns. The proposed approach builds on existing graph clustering methods that identify both communities of nodes, as well as outliers. The importance of incorporating interactions’ intensity in the community detection algorithm is initially investigated by a benchmarking process on synthetic graphs. By applying the proposed approach on a topic-focused dataset of Twitter users’ interactions, we reveal communities with different features which are further analyzed to reveal and summarize the given topic’s impact on social media users.

Requirements and architecture design principles for a smart city experiment with sensor and social networks integration

Smart city infrastructures offer unique testbeds ground for innovative experimentation and services offering. Sensors networks in cities with integrated social networks activities can improve people-centric services, while improving infrastructures setting. This work summarizes the principles and priorities chosen in a smart city experiment, entitled SEN2SOC which bridges sensor measurements and social networks interactions for supporting smart city services. SEN2SOC prioritizes requirements along particular categories which cover data collection, users sensing along with applications implementation and architectural concerns. These requirements are correlated with the suggested components in an architecture which is flexible enough in order to permit various activities control flow in terms of data preprocessing, conditions detection, statistical analysis as well as applications development and social data mining.

Harvesting Opinions and Emotions from Social Media Textual Resources

Harvesting sentiments from social media textual resources can reveal insightful information. The understanding and modeling of such resources are key requirements for accurately capturing the conveyed sentiments. Here, the authors consider multiple approaches, with an emphasis on detecting sentiments in Web 2.0 textual resources.

MultiSpot: Spotting Sentiments with Semantic Aware Multilevel Cascaded Analysis

Given a textual resource (e.g. post, review, comment), how can we spot the expressed sentiment? What will be the core information to be used for accurately capturing sentiment given a number of textual resources? Here, we introduce an approach for extracting and aggregating information from different text-levels, namely words and sentences, in an effort to improve the capturing of documents’ sentiments in relation to the state of the art approaches. Our main contributions are: (a) the proposal of two semantic aware approaches for enhancing the cascaded phase of a sentiment analysis process; and (b) MultiSpot, a multilevel sentiment analysis approach which combines word and sentence level features. We present experiments on two real-world datasets containing movie reviews.

Detecting Aggressors and Bullies on Twitter

Online social networks constitute an integral part of peoples every day social activity and the existence of aggressive and bullying phenomena in such spaces is inevitable. In this work, we analyze user behavior on Twitter in an effort to detect cyberbullies and cuber-aggressors by considering specific attributes of their online activity using machine learning classifiers.

Detecting variation of emotions in online activities

Abstract Online text sources form evolving large scale data repositories out of which valuable knowledge about human emotions can be derived. Beyond the primary emotions which refer to the global emotional signals, deeper understanding of a wider spectrum of emotions is important to detect online public views and attitudes. The present work is motivated by the need to test and provide a system that categorizes emotion in online activities. Such a system can be beneficial for online services, companies recommendations, and social support communities. The main contributions of this work are to: (a) detect primary emotions, social ones, and those that characterize general affective states from online text sources, (b) compare and validate different emotional analysis processes to highlight those that are most efficient, and (c) provide a proof of concept case study to monitor and validate online activity, both explicitly and implicitly. The proposed approaches are tested on three datasets collected from different sources, i.e., news agencies, Twitter, and Facebook, and on different languages, i.e., English and Greek. Study results demonstrate that the methodologies at hand succeed to detect a wider spectrum of emotions out of text sources.

User communities evolution in microblogs: A public awareness barometer for real world events

In social media, users’ interactions are affected by real-world events which influence emergence and shifts of opinions and topics. Interactions around an event-related topic can be captured in a weighted network, while identification of connectivity and intensity patterns can improve understanding of users’ interest on the topic. Community detection is studied here as a means to reveal groups of social media users with common interaction patterns in such networks. The proposed community detection approach identifies communities exploiting both structural properties and intensity patterns, while dynamics of communities’ evolution around an event are revealed based on an iterative community detection and mapping scheme. We investigate the importance of considering interactions’ intensity for community detection via a benchmarking process on synthetic graphs and propose a generic framework for: i) modeling user interactions, ii) identifying static and evolving communities around events, iii) extracting quantitative and qualitative measurements from the communities’ timeline, iv) leveraging measurements to understand the events’ impact. Two real-world case studies based on Twitter interactions demonstrate the framework’s potential for capturing and interpreting associations among communities and events.