Prasanna Kumar Desikan

Incremental page rank computation on evolving graphs

Link Analysis has been a popular and widely used Web mining technique, especially in the area of Web search. Various ranking schemes based on link analysis have been proposed, of which the PageRank metric has gained the most popularity with the success of Google. Over the last few years, there has been significant work in improving the relevance model of PageRank to address issues such as personalization and topic relevance. In addition, a variety of ideas have been proposed to address the computational aspects of PageRank, both in terms of efficient I/O computations and matrix computations involved in computing the PageRank score. The key challenge has been to perform computation on very large Web graphs. In this paper, we propose a method to incrementally compute PageRank for a large graph that is evolving. We note that although the Web graph evolves over time, its rate of change is rather slow. When compared to its size. We exploit the underlying principle of first order markov model on which PageRank is based, to incrementally compute PageRank for the evolving Web graph. Our experimental results show significant speed up in computational cost, the computation involves only the (small) portion of Web graph that has undergone change. Our approach is quite general, and can be used to incrementally compute (on evolving graphs) any metric that satisfies the first order Markov property.

WICER: A Weighted Inter-Cluster Edge Ranking for Clustered Graphs

Several algorithms based on link analysis have been developed to measure the importance of nodes on a graph such as pages on the World Wide Web. PageRank and HITS are the most popular ranking algorithms to rank the nodes of any directed graph. But, both these algorithms assign equal importance to all the edges and nodes, ignoring the semantically rich information from nodes and edges. Therefore, in the case of a graph containing natural clusters, these algorithms do not differentiate between inter-cluster edges and intra-cluster edges. Based on this parameter, we propose a weighted inter-cluster edge ranking for clustered graphs that weighs edges (based on whether it is an inter-cluster or an intra-cluster edge) and nodes (based on the number of clusters it connects). We introduce a parameter /spl alpha/ which can be adjusted depending on the bias desired in a clustered graph. Our experiments were two fold. We implemented our algorithm to a relationship set representing legal entities and documents and the results indicate the significance of the weighted edge approach. We also generated biased and random walks to quantitatively study the performance.

Mining temporally changing web usage graphs

Web mining has been explored to a vast degree and different techniques have been proposed for a variety of applications that include Web Search, Web Classification, Web Personalization etc. Most research on Web mining has been from a ‘data-centric’ point of view. The focus has been primarily on developing measures and applications based on data collected from content, structure and usage of Web until a particular time instance. In this project we examine another dimension of Web Mining, namely temporal dimension. Web data has been evolving over time, reflecting the ongoing trends. These changes in data in the temporal dimension reveal new kind of information. This information has not captured the attention of the Web mining research community to a large extent. In this paper, we highlight the significance of studying the evolving nature of the Web graphs. We have classified the approach to such problems at three levels of analysis: single node, sub-graphs and whole graphs. We provide a framework to approach problems in this kind of analysis and identify interesting problems at each level. Our experiments verify the significance of such an analysis and also point to future directions in this area. The approach we take is generic and can be applied to other domains, where data can be modeled as a graph, such as network intrusion detection or social networks.

Web Mining For Self-directed E-learning

Self-directed e-learning focuses on the independent learner, one who engages in education at his own pace, free from curricular obligation. A number of tools, some purposefully and others serendipitously, have become key enablers of this learning paradigm. For example, tools such a Google Scholar, CiteSeer Research Index, etc. make it possible to do literature search without stepping out of one’s room. Due to the same technologies which helped make self-directed e-learning possible in the first place, these tools are in danger of delivering diminishing returns as micro-learning, lifelong education, and continuous education become the norm in our Information Age. Web Mining, however, may potentially offer a solution to this issue. In this chapter, we investigate specific examples of selfdirected e-learning and how their functionality and utility can be improved through the use of Web Mining technology, techniques, and practices. Our work demonstrates the usefulness of Web Mining as it applies to self-directed elearning and the need to map implicit relationships in learner behaviour, usage, and context.

Divide and conquer approach for efficient pagerank computation

PageRank is a popular ranking metric for large graphs such as theWorld Wide Web. Current research techniques for improving computational efficiency of PageRank have focussed on improving the I/O cost, convergence and parallelizing the computation process. In this paper, we propose a divide and conquer strategy for efficient computation of PageRank. The strategy is different from contemporary improvements in that itcan be combined with any existing enhancements to PageRank, giving way to an entire class of more efficient algorithms. Wepresent a novel graph-partitioning technique for dividing thegraph into subgraphs, on which computation can be performed independently. This approach has two significant benefits. Firstly, since the approach focuses on work-reduction, it can be combined with any existing enhancements to PageRank. Secondly, the proposed approach leads naturally into developing an incremental approach for computation of such ranking metrics given that these large graphs evolve over a period of time. The partitioning technique is both lossless and independent of the type (variant) ofPageRank computation algorithm used. The experimental results for a static single graph (graph at a single time instance) as well as for the incremental computation in case of evolving graphs, illustrate the utility of our novel partitioning approach. The proposed approach can also be applied for the computation of anyother metric based on first order Markov chain model.

USER: user-sensitive expert recommendations for knowledge-dense environments

Traditional recommender systems tend to focus on e-commerce applications, recommending products to users from a large catalog of available items. The goal has been to increase sales by tapping into the users interests by utilizing information from various data sources to make relevant recommendations. Education, government, and policy websites face parallel challenges, except the product is information and their users may not be aware of what is relevant and what isnt. Given a large, knowledge-dense website and a nonexpert user searching for information, making relevant recommendations becomes a significant challenge. This paper addresses the problem of providing recommendations to non-experts, helping them understand what they need to know, as opposed to what is popular among other users. The approach is usersensitive in that it adopts a ‘model of learning whereby the users context is dynamically interpreted as they browse and then leveraging that information to improve our recommendations.

Early Prediction of Potentially Preventable Events in Ambulatory Care Sensitive Admissions from Clinical Data

Ambulatory care sensitive conditions (ACSCs) are characterized as health conditions for which good outpatient care can potentially prevent the need for hospitalization, or for which early intervention can prevent complications or more severe disease. Currently, there are 16 identified ACSCs within the US health system: diabetes short-term complication, perforated appendix, diabetes long-term complication, pediatric asthma, chronic obstructive pulmonary disease, pediatric gastroenteritis, hypertension, congestive heart failure, low birth weight rate, dehydration, bacterial pneumonia, urinary tract infection, angina admission without procedure, uncontrolled diabetes, adult asthma, and lower-extremity amputation among patients with diabetes. Potentially preventable acute health events (PPEs) for such diagnosis codes represent a straightforward opportunity for reducing medical costs while concomitantly improving quality of care. While claims data have previously been used to predict future health outcomes of patients, we report here a novel approach, using data mining techniques, towards supplementing such data with patients electronic health records (EHR) to develop a clinical decision support system that satisfactorily predicts the onset of PPEs in a large population of patients.