We consider the problem of identifying the defectives from a population of items via anon-adaptive group testingframework with a random pooling-matrix design. We analyze the sufficient number of tests needed forapproximate set identification, i.e., for identifyingalmostall the defective and non-defective items with highconfidence. To this end, we view the group testing problem as a function learning problem in the probably approximately correct (PAC) framework. We derive sufficiency bounds on the number of tests for popular binary group testing recovery algorithms, namely, combinatorial orthogonal matching pursuit under Bernoulli and near-constant row-weight test designs, and definite defectives under a Bernoulli test design. We compare the derived bounds with the existing ones in the literature for exact recovery both theoretically and using simulations. Finally, we contrast the three cases under consideration in terms of the sufficient testing ratesurfaceand the sufficient number of testscontoursacross the range of the approximation and confidence levels.

In this paper, we derive sufficiency results on the number of group tests required, in a non-adaptive random pooling matrix setting, to find almost all the defective and non-defective items with high confidence, via two popular algorithms in the group testing literature, namely CoMa and DD. To this end, we propose viewing the group testing problem as an online function learning problem and develop our analysis using the probably approximately correct (PAC) framework. We compare the derived bounds with existing bounds literature for exact recovery both theoretically and using simulations. We also illustrate the savings in the number of tests required for approximate defective set recovery compared to exact recovery.

Emails constitute the bulk of all official communications in any organization. Email repositories are tacit store-houses of knowledge about people, projects and processes. Mining one's own email repository can also provide interesting and valuable insights about his or her engagements and contacts along different dimensions. In this paper, we propose an email analytics framework that combines text-mining, network analysis and data analytics principles to mine email repositories for useful insights. While individuals are more attuned to looking at emails as individual items along with a history that is embedded in the trail, mining the whole collection can also lead to knowledge-discovery about similarities and dissimilarities of different engagements. This in turn can lead to valuable information like comparative status reports on various projects or deeper insights about why certain projects succeed while others don't. Given the volumes, diversity and noisy nature of e-mails, it becomes impossible for human beings to comprehend the impact of all of it unless the task is automated and approached in a structured fashion. We show that combination of text and network analytics along with temporal reasoning can provide valuable insights about task-states, actionable items, recommendations and forecasts. These insights can be exploited very effectively for project-management tasks like automated identification of bottlenecks or their causes, elimination of inefficiencies, early-warnings and suggestions about proactive measures to avoid problems. It is possible to extend the framework quite easily to analyze multiple email repositories of different users, though this work does not address the privacy or security concerns that might exist.

Text analytics on consumer-generated content has gained significant momentum over last few years. A wide-range of text mining techniques has been proposed which can provide interesting insights about the text content. But, the challenge still exists in consuming the extracted information in form of actionable intelligence. Identifying actionable intelligence is difficult due to differences in consumer and business languages. Since feedbacks rarely talks of a single problem, determining the problems is also challenging. We propose a framework to address some of these challenges. Organizational websites or standard domain-ontologies are rich repositories of domain knowledge. The proposed method utilizes this knowledge to learn a discriminative classifier model for a domain using Fisher's discriminant metric. The consumer feedbacks are classified to different business categories using the learnt model. The output is further fed into a fuzzy reasoning unit where every feedback is assigned confidence values for each category. Initial experiments show that the proposed framework is capable of handling text feedbacks containing customer complaints in various domains.