Penn Interactive Media Colloquium

Philadelphia, Pennsylvania
Penn Interactive Media Colloquium
Spring 2010 Schedule

12 p.m. - 1:30 p.m.
JMHH Room F60
Lunch will be served

April 15, 2010
"What are people saying about my product"
Lyle Ungar
Associate Professor, School of Engineering and Applied Sciences
Saurabh Goorha
Solutions Strategist
Dow Jones & Co.

Social and mainstream media are rich sources of information about what people think about products and companies. We present two case studies that show how this information can be extracted and exploited. 1) We analyze product comparisons on discussion boards and determine which products consumers compare, what attributes they compare the products on, and which products they prefer on each dimension. 2) We process over 100,000 news articles, blog posts and tweets a day and extract new things being said about products of interest. The second case study demonstrates a good ability to rapidly pinpoint emerging subjects that are of potential interest to, for example, brand managers, in spite of the vast number of "uninteresting" mentions of the products.

March 18, 2010 - In Collaboration with The Annenberg School's Internet & Media Policy Workshop Series
"Inference in Large Social Networks"
Shawndra Hill
Assistant Professor Operations and Information Management

Researchers and practitioners increasingly are gaining access to data on explicit social networks. For example, telecommunications and technology firms record data on consumer networks (via phone calls, emails, voice-over-IP, instant messaging), and social-network portal sites such as MySpace, Friendster and Facebook record consumer-generated data on social networks and online advertising firms are now linking these explicit networks to clicks and purchases. Inference for fraud detection, marketing, and other tasks may be improved with learned models that take social networks into account and with collective inference, which allows inferences about nodes in the network to affect each other. However, the aforementioned social network graphs can be huge, comprising millions to billions of nodes and one or two orders of magnitude more links. This paper studies the application of collective inference to improve prediction over a massive graph. Faced initially with a social network comprising hundreds of millions of nodes and a few billion edges, our goal is: to produce an approximate consumer network that is orders of magnitude smaller, but still facilitates improved performance via collective inference. We assess whether collective inference can improve learned targeted-marketing models for a social network of consumers of telecommunication services and other products advertised online. Prior work has shown improvement to the learning of targeting models by including social-neighborhood information—in particular, information on existing customers in the immediate social network of a potential target. However, the improvement was restricted to the “network neighbors”, those targets linked to a prior customer thought to be good candidates for the new service.   Collective inference techniques may extend the predictive influence of existing customers beyond their immediate neighborhoods. For the present work, our motivating conjecture has been that this influence can improve prediction for consumers who are not strongly connected to existing customers. Our results show that this is indeed the case: collective inference on the approximate network enables significantly improved predictive performance for non-network-neighbor consumers, and for consumers who have few links to existing customers. In the talk, we motivate our approach, describe our sampling method, present results on applying our approach to a large real-world target marketing campaigns in two industries, and finally discuss our findings.

February 18, 2010
"Nonlinear Systems and Design"
Peter Lloyd Jones, Jenny Sabin and Andrew Lucia

Sabin+Jones LabStudio
School of Design and the Institute for Medicine & Engineering

Scale-free analysis of complex and dynamic datasets using tools developed within LabStudio: How might we interpret complex, dynamic, yet heterogeneous systems, and effectively describe their nature? Architects and cell biologists are each interested in revealing the logical nature of systemic rules and relationships, particularly as they relate to time and space. Technology has afforded us with an extraordinary ability to generate information, yet this has resulted in an ever-increasing inability to organize, visualize and model diverse datasets and processes. Given this, it is becoming even more challenging to interpret and model complexity using existing approaches. With increasing amounts of data now being generated, there is a growing demand for more sophisticated computational tools that are capable of extracting and analyzing specific temporo-spatial relationships. To address this problem, Andrew Lucia, LabStudio Research Associate, will present his latest approaches for the analysis and visualization of large biological datasets that remain indecipherable using existing means.


Contact Information

  • Rebecca Alig, WIMI Associate Director
    Phone: 215-746-4160
© 2019
Quick, easy and affordable online event registration and event management software for all event sizes.