Electronic Thesis/Dissertation


Deep Web Data Analytics Open Access

Downloadable Content

Download PDF

A large portion of data available on the web is present in the so called ''Deep Web''. The deep web consists of private or hidden databases that lie behind form-like query interfaces that allow users to browse these databases in a controlled manner. While hidden database interfaces are normally designed to allow users to execute search queries, for certain applications it is also useful to perform data analytics over such databases. Data analytics challenges toward online social network is one of most popular topic on this area. In the first party of my research. We targeted challenges by data analytics techniques that can be performed only using the public interfaces of the databases while respecting the data access limitations (e.g., query rate limits) imposed by the data owners on a general view. We developed System HYDRA (Hidden Database Research and Analytics) which enables fast sampling and data analytics over a hidden web database that provides nothing but a form-like web search interface as its only access channel. Broadly, it consists of three major components: (1) SAMPLE-GEN which produces samples according to a given sampling distribution (2) SAMPLE-EVAL that evaluates samples produced by SAMPLE-GEN and also generates estimations for a given aggregate query and (3) TIMBR that enables fast and easy construction of a wrapper that models both input and output interface of the web database thereby translating supported search queries to HTTP requests and retrieving top-k query answers from HTTP responses. As another part of my research, we will target challenges on the existing Markov Chain Monte Carlo methods such as random walks based sampling algorithms on websites that having graph browsing interfaces(e.g., online social networks). The problem with such an approach, however, is the large amount of queries often required (i.e., a long ''burn-in time'') for a random walk to reach a desired (stationary) sampling distribution. To reduce the ''burn-in'' time, we introduce the idea of a ''Cross-Community Random Walk'' algorithm leverage the community affiliation information. By increasing the weight of the edge that across different community, our random walk can go through all different community. We demonstrated the superiority of ''Cross-Community Random Walk'' over traditional simple random walks through theoretical analysis and extensive experiments over real world online social networks.

Author Language Keyword Date created Type of Work Rights statement GW Unit Degree Advisor Committee Member(s) Persistent URL