Streams of objects that are associated with one or more labels at the same time appear in many applications. However, stream classification of multi-label data is largely unexplored. Existing approaches try to tackle the problem by transferring traditional single-label stream classification practices to the multi-label domain. Nevertheless, they fail to consider some of the unique properties of the problem such as within and between class imbalance and multiple concept drift. To deal with these challenges, this paper proposes a novel multilabel stream classification approach that employs two windows for each label, one for positive and one for negative examples. Instance-sharing is exploited for space efficiency, while a time-efficient instantiation based on the k-Nearest Neighbor algorithm is also proposed. Finally, a batch-incremental thresholding technique is proposed to further deal with the class imbalance problem. Results of an empirical comparison against two other methods on three real world datasets are in favor of the proposed approach.

/pdf/dealing-with-concept-drift-and-class-imbalance-in-multi-3hz3oi0utp.pdf

Dealing with concept drift and class imbalance in multi-label stream classification

Big data trend has enforced the data-centric systems to have continuous fast data streams. In recent years, real-time analytics on stream data has formed into a new research field, which aims to answer queries about what-is-happening-now with a negligible delay. The real challenge with real-time stream data processing is that it is impossible to store instances of data, and therefore online analytical algorithms are utilized. To perform real-time analytics, pre-processing of data should be performed in a way that only a short summary of stream is stored in main memory. In addition, due to high speed of arrival, average processing time for each instance of data should be in such a way that incoming instances are not lost without being captured. Lastly, the learner needs to provide high analytical accuracy measures. Sentinel is a distributed system written in Java that aims to solve this challenge by enforcing both the processing and learning process to be done in distributed form. Sentinel is built on top of Apache Storm, a distributed computing platform. Sentinels learner, Vertical Hoeffding Tree, is a parallel decision tree-learning algorithm based on the VFDT, with ability of enabling parallel classification in distributed environments. Sentinel also uses SpaceSaving to keep a summary of the data stream and stores its summary in a synopsis data structure. Application of Sentinel on Twitter Public Stream API is shown and the results are discussed.

Distributed Real-Time Sentiment Analysis for Big Data Social Streams

The development of new techniques and the emergence of new high- throughput tools have led to a new information revolution. The amount and the diversity of the information that need to be stored and processed have led to the adoption of data integration systems in order to deal with information extraction from disparate sources. The mediation between traditional databases and ontologies has been recognized as a cornerstone issue in bringing in legacy data with formal semantic meaning. However, our knowledge evolves due to the rapid scientific development, so ontologies and schemata need to change in order to capture and accommodate such an evolution. When ontologies change, these changes should somehow be rendered and used by the pre-existing data integration systems, a problem that most of the integration systems seem to ignore. In this paper, we review existing approaches for ontology/schema evolution and examine their applicability in a state-of-the-art, ontology-based data integration setting. Then, we show that changes in schemata differ significantly from changes in ontologies. This strengthens our position that current state of the art systems are not adequate for ontology-based data integration. So, we give the requirements for an ideal data integration system that will enable and exploit ontology evolution.

Ontology and Schema Evolution in Data Integration: Review and Assessment

Creating global schema over a set of heterogeneous and distributed database entails dealing with a different data model or different structure. The most important issue in schema integration is how to identify correspondences among schema element names. Element names may be given a different meaning at different schemas. For example, the noun Area based on the WorldNet dictionary has five word-senses (location, square feet, country, region and sphere), so how is it ensured that the concept name Area in different schema is used in the same meaning. In this paper, a new approach called (ICASE) Identify Correspondences among Schema Elements, ICASE takes the challenge to solve the problems of naming conflict which includes homonym conflict and synonym conflict. Homonym conflict arises as a result of using the same name to represent different elements while the synonym conflict means the presence of two elements in different schemas, represented by different names. WorldNet dictionary helps ICASE to represent the intended meaning of the element name since fully automatic schema integration is still impossible due to the huge amount of semantic diversity in designing databases. A friendly user interface is built to allow the user to interact with the system and to automate the system.

Semi-Automated Schema Integration (Icase): A Tool To Identify And Resolve Naming Conflicts

Applications such as Facebook, Twitter and Foursquare have brought the mass adoption of personal short messages, distributed in (soft) real-time on the Internet to a large number of users. These messages are complemented with rich contextual information such as the identity, time and location of the person sending the message (e.g., Foursquare has millions of users sharing their location on a regular basis, with almost 1 million updates per day). Such contextual messages raise serious concerns in terms of scalability and delivery delay; this results not only from their huge number but also because the set of user recipients changes for each message (as their interests continuously change), preventing the use of well-known solutions such as pub-sub and multicast trees. This leads to the use of non-scalable broadcast based solutions or point-to-point messaging. We propose Radiator, a middleware to assist application programmers implementing efficient context propagation mechanisms within their applications. Based on each user’s current context, Radiator continuously adapts each message propagation path and delivery delay, making an efficient use of network bandwidth, arguably the biggest bottleneck in the deployment of large-scale context propagation systems. Our experimental results demonstrate a 20x reduction on consumed bandwidth without affecting the real-time usefulness of the propagated messages.

/pdf/radiator-efficient-message-propagation-in-context-aware-4axkwezxlk.pdf

Patricia G. Selinger

Papers

Proceedings of the 29th international conference on Very large data bases - Volume 29