Aspect terms extraction (ATE), a key subtask for aspect-based sentiment analysis, opinion summarization, and topic modeling aims at extracting grammatical elements (nouns, phrases, and adjectives) from user reviews that reveal the discussed features of the entity under review. These aspect terms are usually the targets of the opinions expressed. Identifying them requires tackling substantial linguistic challenges but, due to the multiple commercial and social applications, significant research effort has been invested in efficiently mining aspects. Recent advances in ATE address methods that exploit a sentence or a word-level encoding of a user review as a solution. This article proposes a novel and effective word- and sentence-level encoding framework, which utilizes a neural network architecture that learns to extract aspect terms. The main advantage of our approach is that it can extract explicit and implicit aspects (i.e., aspects that are not directly mentioned in the user-generated text). We evaluate our method on four widely used datasets where we prove its efficiency against state-of-the-art alternative approaches.

We define and solve the problem of event detection and delineation as a task of identifying events and decomposing them to their major sub-events, with a description and a timeline. We propose DeLi, an algorithm that focuses on providing such an understanding of events and sub-events. DeLi, to the best of our knowledge, is the first method that addresses the problem in a generic stream of text, and in an online fashion. Extensive evaluation on social streaming data demonstrates that, by combining the structure of a social network with content attributes, our method outperforms the state-of-the-art techniques.

Social networks have become the de facto online resource for people to share, comment on and be informed about events pertinent to their interests and livelihood, ranging from road traffic or an illness to concerts and earthquakes, to economics and politics. This has been the driving force behind research endeavors that analyse such data. In this paper, we focus on how Content Networks can help us identify events effectively. Content Networks incorporate both structural and content-related information of a social network in a unified way, at the same time, bringing together two disparate lines of research: graph-based and content-based event discovery in social media. We model interactions of two types of nodes, users and content, and introduce an algorithm that builds heterogeneous, dynamic graphs, in addition to revealing content links in the network's structure. By linking similar content nodes and tracking connected components over time, we can effectively identify different types of events. Our evaluation on social media streaming data suggests that our approach outperforms state-of-the-art techniques, while showcasing the significance of hidden links to the quality of the results.

In this paper we tackle the recently proposed problem of hidden streams. In many situations, the data stream that we are interested in, is not directly accessible. Instead, part of the data can be accessed only through applying filters (e.g. keyword filtering). In fact this is the case of the most discussed social stream today, Twitter. The problem in this case is how to retrieve as many relevant documents as possible by applying the most appropriate set of filters to the original stream and, at the same time, respect a number of constrains (e.g. maximum number of filters that can be applied). In this work we introduce a search approach on a dynamic filter space. We utilize heterogeneous filters (not only keywords) making no assumptions about the attributes of the individual filters. We advance current research by considering realistically hard constraints based on real-world scenarios that require tracking of multiple dynamic topics. We demonstrate the effectiveness of our approaches on a set of topics of static and dynamic nature. The development of the approach was motivated by a real application. Our system is deployed in Dublin City's Traffic Management Center and allows the city officers to analyze large sources of heterogeneous data and identify events related to traffic as well as emergencies.

Online Social Networks (OSNs) constitute one of the most important communication channels and are widely utilized as news sources. Information spreads widely and rapidly in OSNs through the word-of-mouth effect. However, it is not uncommon for misinformation to propagate in the network. Misinformation dissemination may lead to undesirable effects, especially in cases where the non-credible information concerns emergency events. Therefore, it is essential to timely limit the propagation of misinformation. Towards this goal, we suggest a novel propagation model, namely the Dynamic Linear Threshold (DLT) model, that effectively captures the way contradictory information, i.e., misinformation and credible information, propagates in the network. The DLT model considers the probability of a user alternating between competing beliefs, assisting in either the propagation of misinformation or credible news. Based on the DLT model, we formulate an optimization problem that aims in identifying the most appropriate subset of users to limit the spread of misinformation by initiating the propagation of credible information. Through extensive experimental evaluation we demonstrate that our approach outperforms its competitors.

Users in social networks use hashtags for various reasons, some of them being serving search purposes, gaining attention or popularity or starting new conversation - thus, creating viral memes. In this paper we address the problem of classifying these hashtags in different categories, based on whether they represent a real life event or a social network generated meme. We compute a set of language-agnostic features to aid the classification of hashtags into events and memes and we provide an extensive study of the behavior that characterizes memes and events. We focus on Twitter social network, we apply our methods on a big dataset and reveal interesting characteristics of the two classes of hashtags.

Voting advice applications (VAAs) are online tools that have become increasingly popular and purportedly aid users in deciding which party/candidate to vote for during an election. In this paper we present an innovation to current VAA design which is based on the introduction of a social network element. We refer to this new type of online tool as a social voting advice application (SVAA). SVAAs extend VAAs by providing (a) community-based recommendations, (b) comparison of users' political opinions, and (c) a channel of user communication. In addition, SVAAs enriched with data mining modules, can operate as citizen sensors recording the sentiment of the electorate on issues and candidates. Drawing on VAA datasets generated by the Preference Matcher research consortium, we evaluate the results of the first VAA-Choose4Greece-which incorporated social voting features and was launched during the landmark Greek national elections of 2012. We demonstrate how an SVAA can provide community based features and, at the same time, serve as a citizen sensor. Evaluation of the proposed techniques is realized on a series of datasets collected from various VAAs, including Choose4Greece. The collection is made available online in order to promote research in the field.

A key challenge for Grid and Cloud infrastructures is to make their services easily accessible and attractive to end-users. In this paper we introduce tagging capabilities to the Miner soft system, a powerful tool for software search and discovery in order to help end-users locate application software suitable to their needs. Miner soft is now able to predict and automatically assign tags to software resources it indexes. In order to achieve this, we model the problem of tag prediction as a multi-label classification problem. Using data extracted from production-quality Grid and Cloud computing infrastructures, we evaluate an important number of multi-label classifiers and discuss which one and with what settings is the most appropriate for use in the particular problem.

A simple yet effective multilabel learning method, called label powerset (LP), considers each distinct combination of labels that exist in the training set as a different class value of a single-label classification task. The computational efficiency and predictive performance of LP is challenged by application domains with large number of labels and training examples. In these cases, the number of classes may become very large and at the same time many classes are associated with very few training examples. To deal with these problems, this paper proposes breaking the initial set of labels into a number of small random subsets, called labelsets and employing LP to train a corresponding classifier. The labelsets can be either disjoint or overlapping depending on which of two strategies is used to construct them. The proposed method is called RAkEL (RAndom k labELsets), where k is a parameter that specifies the size of the subsets. Empirical evidence indicates that RAkEL manages to improve substantially over LP, especially in domains with large number of labels and exhibits competitive performance against other high-performing multilabel learning methods.

This paper proposes a general framework for classifying data streams by exploiting incremental clustering in order to dynamically build and update an ensemble of incremental classifiers. To achieve this, a transformation function that maps batches of examples into a new conceptual feature space is proposed. The clustering algorithm is then applied in order to group different concepts and identify recurring contexts. The ensemble is produced by maintaining an classifier for every concept discovered in the streamThe full version of this paper as well as the datasets used for evaluation can be found at: http://mlkd.csd.auth.gr/concept_drift.html