The 25th Australasian Database Conference (ADC 2014)

The ADC-14 program-at-a-glance is available. Detailed program below:

Monday 14 July - SEMINARS
Location: St Leo’s College, The University of Queensland

Text data, broadly including all kinds of natural language text produced by humans (e.g., web pages, social media, email messages, news articles, government documents, and scientific literature), have been growing dramatically recently. This creates great opportunities for applying computational methods to mine large amounts of text data to discover all kinds of useful knowledge, especially knowledge about people's opinions, preferences, and behavior. Due to the difficulty in precisely understanding natural language by computers, scalable text mining algorithms tend to be based on statistical analysis and probabilistic reasoning. In this tutorial, I will systematically review the major statistical methods developed for mining text data, with a focus on covering probabilistic topic models for mining topics and topical patterns in text data, and statistical methods for integrating and analyzing scattered online opinions.

Recently, the popularity of crowdsourcing has brought a new opportunity to engage human intelligence into various data analysis tasks. Compared with computer systems, crowds are good at handling items with human-intrinsic values or features. Existing approaches develop sophisticated methods by utilizing the crowd as a new type of processor, a.k.a. HPU (Human Processing Unit). As a consequence, tasks executed on HPU are called HPU-based tasks. Now we are in the Big Data Era, a nature question arises: How about crowdsourcing over Big Data, are we there yet? In this talk, I will first briefly review the history of crowdsourcing and discuss the key issues related to crowdsourcing. Then, I will demonstrate the power of crowdsourcing in solving the well-known and very hard data integration problem, schema matching, and discuss how to migrate the power of crowdsourcing to a social media platform whose users can serve as a huge reservoir of workers. Finally, I will highlight some research challenges about crowdsourcing over Big Data.

The real applications that need graph processing techniques to handle a large graph can be found from many real applications including online social networks, biological networks, ontology, transportation networks, etc. In this talk, we will discuss some selected research topics on graph mining and graph query processing over large graphs. For graph mining, we will focus on ranking nodes in a large graph. We will discuss ranking over trust networks, random-walk domination, and diversified ranking. For ranking nodes over trust network, we discuss how to take the trust score into consideration while ranking. For the random-walk domination, we discuss the techniques for handling item-placement in online social networks and ads-placement in advertisement networks. For diversified ranking, we discuss how to find top-k nodes that match the user query and are very different from each other. For graph query processing, we will discuss top-k structural diversity search, finding the maximal cliques in massive networks, and I/O efficient computing techniques that make a large directed graph small and simple. The other related topics may be also addressed in this talk.

Tuesday 15 July – CONFERENCE DAY 1
Location: St Leo’s College, The University of Queensland

Data ecosystems involve the coexistence of one or more data collections, typically databases, and their surrounding applications for data entry and retrieval. For decades, both data and ecosystem management have failed to address significant, costly and labor-consuming challenges which involve (a) the departure from databases focusing on alphanumeric data only, (b) their inability to be integrated and provide transparent access and composition facilities for heterogeneous data, (c) their static querying nature, which is deprived of personal, context-aware or interactive characteristics, (d) the enforcement of DBMS operation over monolithic servers, and, (e) the complete indifference to problems of evolution and adaptation over time. In this talk we address issues around the methodologies, the theoretical and modeling foundations as well as the algorithmic techniques and the necessary software architectures that will facilitate the personalization, integration, and evolution management facilities for data ecosystems that operate over a decentralized infrastructure for a large variety of data types.

RESEARCH PAPER PRESENTATIONS

Wednesday 16 July – CONFERENCE DAY 2
Location: St Leo’s College, The University of Queensland

Data integration is a challenging task due to the large numbers of autonomous data sources, which necessitates the development of techniques to reason about the costs and benefits of acquiring and integrating data. Too many sources can result in a huge integration cost, and low quality sources can be detrimental to the benefit of integration. In this talk, Dr Srivastava presents the problem of source selection, that is, identifying the subset of sources before integration that maximize the profit (benefit - cost) of integration, for static and dynamic sources. To address this problem, he proposes techniques that, inspired by the marginalism principle in economic theory, integrate a source only if its marginal benefit is higher than its marginal cost. He quantifies the integration benefit in terms of the quality of the integrated data, which is characterized using a set of data quality metrics, including coverage, freshness and accuracy, and develop statistical models for estimating these metrics. Although source selection is NP-complete, he shows that for many practical cases solutions to this problem can be found in polynomial time with approximation guarantees. Finally, he empirically establishes the effectiveness and scalability of his techniques on real-world and synthetic data.

RESEARCH PAPER PRESENTATIONS