Raymond K. Pon

Tracking multiple topics for finding interesting articles

We introduce multiple topic tracking (MTT) for iScore to better recommend news articles for users with multiple interests and to address changes in user interests over time. As an extension of the basic Rocchio algorithm, traditional topic detection and tracking, and single-pass clustering, MTT maintains multiple interest profiles to identify interesting articles for a specific user given user-feedback. Focusing on only interesting topics enables iScore to discard useless profiles to address changes in user interests and to achieve a balance between resource consumption and classification accuracy. Also by relating a topics interestingness to an article.s interestingness, iScore is able to achieve higher quality results than traditional methods such as the Rocchio algorithm. We identify several operating parameters that work well for MTT. Using the same parameters, we show that MTT alone yields high quality results for recommending interesting articles from several corpora. The inclusion of MTT improves iScores performance by 9% in recommending news articles from the Yahoo! News RSS feeds and the TREC11 adaptive filter article collection. And through a small user study, we show that iScore can still perform well when only provided with little user feedback.

TimeLine and visualization of multiple-data sets and the visualization querying challenge

Data in its raw form can potentially contain valuable information, but much of that value is lost if it cannot be presented to a user in a way that is useful and meaningful. Data visualization techniques offer a solution to this issue. Such methods are especially useful in spatial data domains such as medical scan data and geophysical data. However, to properly see trends in data or to relate data from multiple sources, multiple-data set visualization techniques must be used. In research with the time-line paradigm, we have integrated multiple streaming data sources into a single visual interface. Data visualization takes place on several levels, from the visualization of query results in a time-line fashion to using multiple visualization techniques to view, analyze, and compare the data from the results. A significant contribution of this research effort is the extension and combination of existing research efforts into the visualization of multiple-data sets to create new and more flexible techniques. We specifically address visualization issues regarding clarity, speed, and interactivity. The developed visualization tools have also led recently to the visualization querying paradigm and challenge highlighted herein.

Data quality inference

In the field of sensor networks, data integration and collaboration, and intelligence gathering efforts, information on the quality of data sources are important but are often not available. We describe a technique to rank data sources by observing and comparing their behavior (i.e., the data produced by data sources) to rank. Intuitively, our measure characterizes data sources that agree with accurate or high-quality data sources as likely accurate. Furthermore, our measure includes a temporal component that takes into account a data sources past accuracy in evaluating its current accuracy. Initial experimental results based on simulation data to support our hypothesis demonstrate high precision and recall on identifying the most accurate data sources.

Online selection of parameters in the rocchio algorithm for identifying interesting news articles

We show that users have different reading behavior when evaluating the interestingness of articles, calling for different parameter configurations for information retrieval algorithms for different users. Better recommendation results can be made if parameters for common information retrieval algorithms, such as the Rocchio algorithm, are learned dynamically instead of being statically fixed a priori. By dynamically learning good parameter configurations, Rocchio can adapt to differences in user behavior among users. We show that by adaptively learning online the parameters of a simple retrieval algorithm, similar recommendation performance can be achieved as more complex algorithms or algorithms that require extensive fine-tuning. Also we have also shon that online parameter learning can yield 10% better results than best performing filter from the TREC11 adaptive filter task.

Performance-oriented privacy-preserving data integration

Current solutions to integrating private data with public data have provided useful privacy metrics, such as relative information gain, that can be used to evaluate alternative approaches. Unfortunately, they have not addressed critical performance issues, especially when the public database is very large. The use of hashes and noise yields better performance than existing techniques, while still making it difficult for unauthorized entities to distinguish which data items truly exist in the private database. As we show here, the uncertainty introduced by collisions caused by hashing and the injection of noise can be leveraged to perform a privacy-preserving relational join operation between a massive public table and a relatively smaller private one.

Image stack stream viewing and access

Abstract Growing amounts of multimedia data (alphanumeric, image, sound and video) are being captured with the increasing deployment of sensors. Data streams from sensors are being increasingly broadcasted via the Internet. We review the image stack stream model or view of data as a convenient basis for both querying and visualizing the situation and changes through time of phenomena buried in the multitude of such variety of data streams. We outline briefly the requirements with motivating queries, and the recommended extensions to the latest object database management developments, particularly the ODMG standard, to support this. The image stack view should be provided whether or not data streams are stored in a DBMS. We present a common user interfaces over the heterogeneity of data stream sources to define, access and view stack streams, illustrating it with a testbed of Internet stream sources in the geophysical domain. A systems architecture and design is outlined. Image co-registration is an important element and we provide a brief on our approach. Performance and scalability are addressed.