Ph.D. Dissertation Defense
Computer Science and Electrical Engineering
University of Maryland, Baltimore County

Finding Story Chains and Creating Story Maps in Newswire Articles

Xianshu Zhu

10:00-12:00pm Monday 25 November 2013, ITE 325B

There are huge amounts of news articles about events published on the Internet everyday. The flood of information on the Internet can easily swamp people, which seems to produce more pain than gain. While there are some excellent search engines, such as Google, Yahoo and Bing, to help us retrieve information by simply providing keywords, the problem of information overload makes it hard to understand the evolution of a news story. Conventional search engines display unstructured search results, which are ranked by relevance using keyword-based ranking methods and other more complicated ranking algorithms. However, when it comes to searching for a story (a sequence of events), none of the ranking algorithms above can organize the search results by evolution of the story. Limitations of unstructured search results include: (1) Lack of the big picture on complex stories. In general, news articles tend to describe the news story from different perspectives. For complex news stories, users can spend significant time looking through unstructured search results without being able to see the big picture of the story. For instance, Hurricane Katrina struck New Orleans on August 23, 2005. By typing “Hurricane Katrina” in Google, people can get much information about the event and its impact on the economy, health, and government policies, etc. However, people may feel desperate to sort the information to form a story chain that tells how, for example, Hurricane Katrina has impacted government policies. (2) Hard to find hidden relationships between two events: The connections between news events are sometimes extremely complicated and implicit. It is hard for users to discover the connections without thorough investigation of the search results.

In this dissertation, we seek to extend the capability of existing search engines to output coherent story chains and story maps (a map that demonstrates various perspectives on news events), rather than loosely connected pieces of information. By this means, people can obtain a better understanding of the news story, capture the big picture of the news story quickly, and discover hidden relationships between news events. First of all, algorithms for finding story chains have the following two advantages: (1) they can find out how two events are correlated by finding a chain of events that coherently connect them together. Such story chains will help people discover hidden relationship between two events. (2) they allow users to search by complex queries such as “how is event A related to event B”, which does not work well on conventional keyword-based search engines. Secondly, creating story maps by finding different perspectives on a news story and grouping news articles by the perspectives can help users better capture the big picture of the story and give them suggestions on what directions they can further pursue. From a functionality point of view, the story map is similar to the table of content of a book which gives users a high-level overview of the story and guides them during news reading process.

The specific contributions of this dissertation are: (1) Develop various algorithms to find story chains, including: (a) random walk based story chain algorithm; (b) co-clustering based story chain algorithm which further improves the story chains by grouping semantically close words together and propagating the relevance of word nodes to document nodes; (c) finding story chains by extracting multi-dimensional event profiles from unstructured news articles, which aims to better capture relationships among news events. This algorithm significantly improves the quality of the story chains. (2) Develop an algorithm to create story maps which uses Wikipedia as the knowledge base. News articles are represented in the form of bag-of-aspects instead of bag-of-words. Bag-of-aspects representation allows users to search news articles through different aspects of a news event but not through simple keywords matching.

Committee: Drs. Tim Oates (chair), Tim Finin, Charles Nicholas, Sergei Nirenburg and Doug Oard

Computer Science and Electrical Engineering
Quantum Computing Seminar

Four Quantum Algorithms

Samuel Lomonaco, CSEE, UMBC

2:30-3:00 Tuesday, 19 November 2013, ITE 325b

In this lecture, we discuss four quantum algorithms, i.e., Deutsch’s algorithm, Simon’s algorithm, Shor’s algorithm, and Grover’s algorithm. No prior knowledge of quantum mechanics will be assumed. These talks are based on four invited lectures (slides) given at Oxford University for the UMBC audience.

Samuel J. Lomonaco is a professor at the Department of Computer Science and Electrical Engineering of the University of Maryland Baltimore County. He is internationally known for his many contributions in mathematics and in computer science. His research interests span a wide range of subjects from knot theory, algebraic and differential topology to algebraic coding theory, quantum computation, and symbolic computation. In quantum cryptography, he has shown how quantum information theory can be used to gain a better understanding of eavesdropping with quantum entanglement. In quantum computation, he has shown how Lie groups can be used to solve problems arising in the study of quantum entanglement. In 2000 Professor Lomonoco organized the first American Mathematical Society short course on quantum computation.

Organizer: Prof. Samuel Lomonaco, Sorry, you need javascript to view this email address.

Center for Hybrid Multicore Productivity Research
Distinguished Computational Science Lecture Series

Human Computing Capacity and Future Human Development

Professor Bezalel Gavish
Information Technology and Operations Management
Southern Methodist University, Dallas TX

2:30pm Monday, 18 November 2013, ITE 325B, UMBC

This talk introduces bounds on future computers’ processing capacity and analyzes the possibilities for their realization in the long run. The analysis shows the existence of hard limits on the progress in processing capacity, which in turn generates bounds on future computing capacity. The results show that it is unlikely that some of the predictions on future computing capabilities will ever be achieved. The capacity bounds stem from fundamental physical limitations, which generate the relatively tight bounds. Different bounds have been developed that will be reached much faster than expected when compared to using simple traditional forecasting methods. This will be discussed in the lecture.

Assuming that computational activities like decision making, processing, vision, control, auditory and sensing activities of human beings require energy, the above energy based results generate upper bounds on the computational capacity (in the broadest sense) of human beings. The results are architecture independent and have direct impact on research on models of the brain and provide bounds on the cognitive abilities of human beings. A byproduct of this line of research is providing some new conjectures on the past and future physical development of the human species.

Professor Bezalel Gavish holds the Eugene J. and Ruth F. Constantin Distinguished Chair at Southern Methodist University in Dallas, Texas. He was the Chairman of the Information Technology and Operations Management department at the Cox business School. Professor Gavish is the founding Chairman of the International Conference on Telecommunications Systems Management and the International Conference on Networking and Electronic Commerce. He is the Editor-in-Chief of two top ranked research oriented journals, the Telecommunication Systems Journal, and of the Electronic Commerce Research Journal; serves as an Editorial board member of the Wireless Networks journal, Networks, Annals of Information Systems; was Telecommunications Departmental Editor for the Operations Research journal and Department Editor of Distributed Systems in ORSA Journal on Computing; and serves on the editorial boards of Computers and Operations Research, Annals of Mathematics of Artificial Intelligence, INFOR, Mathematics of Industrial Systems, Combinatorial Optimization: Theory and Practice, and Pesquisa Operacional. Prof. Gavish has published over 100 papers in his areas of expertise. He received the Ph.D. (1975) in operations research from the Technion, Israel Institute of Technology.

Computer Science and Electrical Engineering
University of Maryland, Baltimore County

Ph.D. Dissertation Defense

Private Packet Filtering Searching for Sensitive Indicators
without Revealing the Indicators in Collaborative Environments

Michael John Oehler

10:30-12:30 Thursday, 21 November 2013, ITE 325

Private Packet Filtering (PPF) is a new capability that preserves the confidentiality of sensitive attack indicators, and retrieves network packets that match those indicators without revealing the matching packets. The capability is achieved through the definition of a high-level language, the definition of a conjunction operator that expands the breadth of the language, a simulation of the document detection and recovery rates of the output buffer, and through a description of applicable system facets. Fundamentally, PPF uses a private search mechanism that in turn relies on the (partial) homomorphic property of the Paillier cryptosystem. PPF is intended for use in a collaborative environment involving a cyber defender and a partner: The defender has access to a set of sensitive indicators, and is willing to share some of those indicators with the partner. The partner has access to network data, and is willing to share that access. Neither is willing to provide full access. Using the language, the defender creates an encrypted form of the sensitive indicators, and passes the encrypted indicators to the partner. The partner then uses the encrypted indicators to filter packets, and returns an encrypted packet capture file. The partner does not decrypt the indicators and cannot identify which packets matched. The defender decrypts, reassembles the matching packets, gains situational awareness, and notifies the partner of any malicious activity. In this sense, the defender reveals only the observed indicator and retains control of all other indicators. PPF allows both parties to gain situational awareness of malicious activity, and to retain control without exposing every indicator or all network data.

Committee: Dhananjay Phatak (chair), Michael Collins, Josiah Dykstra, Russell Fink, John Pinkston and Alan Sherman

UMBC Information Systems Department
Distinguished Lecture for the Fall

Collective Intelligence

Dr. Ophir Frieder

Georgetown University

12:00-1:00pm Wednesday, 13 November, ITE 456

Collective Intelligence is group intelligence generated by the collaboration of many individuals. However, such intelligence is only as powerful as one’s ability to digest it. Thus, after describing two recent efforts, the first focusing on early disease detection using microblogs and the second focusing on collaborative tag labeling. Potentially, I will likewise describe an older effort that effectively integrates information and comment on its potential for the future.

Ophir Frieder holds the Robert L. McDevitt, K.S.G., K.C.H.S. and Catherine H. McDevitt L.C.H.S. Chair in Computer Science and Information Processing and is Chair of the Department of Computer Science at Georgetown University. He is also Professor of Biostatistics, Bioinformatics and Biomathematics in the Georgetown University Medical Center. He is a Fellow of the AAAS, ACM, and IEEE.

UMBC Information Systems Department

Topical Search in Twitter

Dr. Niloy Ganguly

Indian Institute of Technology Kharagpur

1:00pm Tuesday, 5 November 2013, ITE 459

Twitter is now a popular platform for discovering real-time news on various topics. We are developing methodologies to improve topical search in Twitter, specifically search for topical experts and popular content on specific topics. Utilizing social annotations provided by the Twitter population through the Lists feature, we have developed the following:

• A novel who-is-who system for Twitter, which gives the topical attributes of a specified user. The list-based methodology gives accurate and comprehensive topical attributes for millions of popular Twitter users.
• A search system for topical experts in Twitter. Comparison of our system with the expert search service offered by Twitter shows that the List-based method provide better results for a large number of topical queries.
• A novel topical search system which, given a topic, identifies and clusters the content (tweets, hashtags) being discussed by the community of experts on that topic. Our methodology gives relevant and trustworthy content for a wide range of topics. To the best our knowledge, this is the first systematic attempt to utilize social annotations to provide topical search in Twitter.

Niloy Ganguly is an associate professor in the department of computer science and engineering, Indian Institute of Technology Kharagpur. He received his PhD from Bengal Engineering and Science University, Calcutta, India and his Bachelors in Computer Science and Engineering from IIT Kharagpur. He has been a post doctoral fellow in Technical University of Dresden, Germany. He focuses on investigating several different aspects on online-social networks. He has worked on designing recommendation system based on community structures on various web-social networks like Twitter and Delicious. He has also simultaneously worked on various theoretical issues related to dynamical large networks often termed as complex networks. Specifically he has looked into problems related to percolation, evolution of networks as well as flow of information over these networks. He has been collaborating with various national and international universities and research lab including Duke University, TU Dresden, Germany, MPI PKS and MPI SWS, Germany, Microsoft Lab, India etc. He currently publishes in various top ranking international journals and conferences including CCS, PODC, ICDM, ACL, WWW, INFOCOM, SIGIR, Euro Physics Letters, Physical Review E, ACM and IEEE Transactions, etc.

Computer Science and Electrical Engineering
Quantum Computing Seminar

The Strange World of Quantum Computing

Samuel Lomonaco, CSEE, UMBC

2:30-3:00 Tuesday, 5 November 2013, ITE 325b

This talk will give an introductory overview of quantum computing in an intuitive and conceptual fashion. No prior knowledge of quantum mechanics will be assumed. This is the first of a series of talks based on the four invited lectures given at Oxford. PowerPoint slides can be found online here.

Samuel J. Lomonaco is a professor at the Department of Computer Science and Electrical Engineering of the University of Maryland Baltimore County. He is internationally known for his many contributions in mathematics and in computer science. His research interests span a wide range of subjects from knot theory, algebraic and differential topology to algebraic coding theory, quantum computation, and symbolic computation. In quantum cryptography, he has shown how quantum information theory can be used to gain a better understanding of eavesdropping with quantum entanglement. In quantum computation, he has shown how Lie groups can be used to solve problems arising in the study of quantum entanglement. In 2000 Professor Lomonoco organized the first American Mathematical Society short course on quantum computation.

Organizer: Prof. Samuel Lomonaco, Sorry, you need javascript to view this email address.

Computer Science and Electrical Engineering
University of Maryland, Baltimore County

Deriving Insights from Data:
Peek at Research Challenges for some Industry Verticals

Dr. Manish Gupta
Vice President and Director
Xerox Research Center India

3:00pm Monday, 4 November, 2013, ITE 325b, UMBC

We are entering an era that will usher dramatic changes in several industries via exploitation of data. With the proliferation of data from sensors that are becoming ubiquitous and increasing digitization of data that used to be in non-electronic form, there are opportunities to completely transform the way the world is run. We present examples of such opportunities in the Financial, Healthcare, Education and Infrastructure domains. We also describe unique challenges like scale and heterogeneity in growth markets like India, which often require different approaches to solving these problems. Diving deeper into the healthcare industry, we present preliminary work that shows the applicability of remote sensing and data analytics not only to measure body vitals such as temperature and heart rate, but also to diagnose diseases such as breast cancer and atrial fibrillation (a form of cardiac arrhythmia) in the future. As more of the patients’ medical history gets captured in electronic health record systems, there is a further promise of applying real-time predictive analytics (based on accurate models for diagnosis and treatment of various diseases, utilizing the latest medical literature) to assist doctors in practicing personalized, evidence-based medicine. We describe outstanding challenges, including data privacy, machine learning over heterogeneous forms of data, and financial incentives design, which we believe must be addressed to enable transformational impact.

Dr. Manish Gupta is Vice President at Xerox Corporation and Director of Xerox Research Centre in India. Previously, Manish has served as Managing Director, Technology Division at Goldman Sachs in India, and has held various leadership positions with IBM, including that of Director, IBM Research – India and Chief Technologist, IBM India/South Asia. From 2001 to 2006, he served as a Senior Manager at the IBM T.J. Watson Research Center in Yorktown Heights, New York, where he led the team developing system software for the Blue Gene/L supercomputer. IBM was awarded a National Medal of Technology and Innovation for the invention of Blue Gene by US President Barack Obama in 2009. Manish earned a B.Tech. in Computer Science from IIT Delhi in 1987, an M.S. from the Ohio State University in 1988 and a Ph.D. from the University of Illinois at Urbana Champaign in 1992. He has co-authored over 75 papers, with more than 5,000 citations in Google Scholar (with an h-index of 41) in the areas of high-performance computing, compilers, and virtual machine optimizations, and has been granted more than 15 US patents. While at IBM, Manish received an Outstanding Innovation Award, two Outstanding Technical Achievement Awards and the Lou Gerstner Team Award for Client Excellence. Manish is an ACM Fellow.

Computer Science and Electrical Engineering
University of Maryland, Baltimore County

Acoustic-Tactile Rendering of Visual
Information for the Visually Impaired

Thrasyvoulos N. Pappas

Electrical Engineering and Computer Science
Northwestern University

2:30pm Monday, 11 November 2013, ITE 325B, UMBC

After a brief overview of research in the Department of Electrical Engineering and Computer Science at Northwestern University, we will focus on one particular research problem, the use of hearing and touch for conveying graphical and pictorial information to visually impaired (VI) people. This problem combines visual, acoustic, and tactile signal analysis with and understanding of human perception and interface design. The main idea is that the user actively explores a two-dimensional layout consisting of one or more objects with the finger on a touch screen. The objects are displayed via sounds and raised-dot tactile patterns. The finger acts as a pointing device and provides kinesthetic feedback. The touch screen is partitioned into regions, each representing an element of a visual scene or graphical display. A key element of our research is the use of spatial sound to facilitate active exploration of the layout. We use the head-related transfer function (HRTF) for rendering sound directionality and variations of sound intensity and tempo for rendering proximity. Our research has addressed object shape and size perception, as well as the of a 2-D layout of simple objects with identical size and shape. We have also considered the rendering of a simple scene layout consisting of objects in a linear arrangement, each with a distinct tapping sound, which we compare to a “virtual cane.” Subjective experiments with visually-blocked subjects demonstrate the effectiveness of the proposed approaches. Our research findings are also expected to have an impact in other applications where vision cannot be used, e.g., for GPS navigation while driving, fire-fighter operations in thick smoke, and military missions conducted under the cover of darkness.

Thrasos Pappas received the Ph.D. degree in electrical engineering and computer science from MIT in 1987. From 1987 until 1999, he was a Member of the Technical Staff at Bell Laboratories, Murray Hill, NJ. He joined the EECS Department at Northwestern in 1999. His research interests are in human perception and electronic media, and in particular, image and video quality and compression, image and video analysis, content-based retrieval, model-based halftoning, and tactile and multimodal interfaces. Prof. Pappas is a Fellow of the IEEE and SPIE. He has served as editor-in-chief of the IEEE Transactions on Image Processing (2010-12), elected member of the Board of Governors of the Signal Processing Society of IEEE (2004-07), chair of the IEEE Image and Multidimensional Signal Processing Technical Committee (2002-03), and technical program co-chair of ICIP-01 and ICIP-09. Since 1997 he has been co-chair of the SPIE/IS&T Conference on Human Vision and Electronic Imaging.

Host: Janet C. Rutledge, Ph.D.

PhD Defense

Visualizing Sequential Patterns in Large

Datasets Using Levels of Abstraction

Dana Wortman

11am – 2pm, Friday, 1 November 2013, ITE 325b

Student retention and success are important topics in all academic fields and institutions. Faculty seek to understand which topics, theories, or skills defeat students or require strengthening to promote success. Programs seek to understand how to better sequence courses to ensure students are prepared for requisite future courses. Institutions seek to understand how to intervene to promote retention and improve graduation rates. Unfortunately, most statistics gathered by Institutional Research efforts are limited to failure rates, enrollment rates, and graduation rates and do not often explore individual student performance. While these are often further analyzed by various student demographic attributes such as race and gender, these statistical methods alone are insufficient to understand student performance over time and sequential patterns of enrollment or success and failure. This research presents a method using multiple levels of abstraction to visualize performance patterns over time.

To visualize student enrollment and performance patterns, several issues must be addressed including sequential versus concurrent enrollment, spatial layout of course events, and performance over time. Another challenge addressed by this work is that of presenting sequences within the context of the entire program. To address these issues, multiple abstractions are combined in a multi-layered visualization that presents a high-level overview of students enrollment and performance patterns while retaining detailed information regarding individual student progress and performance as they advance through their courses.

The aggregated view represents the lowest level of abstraction, student enrollment and performance are aggregated into a graph structure, presenting patterns of movement throughout the program at the individual course level. The clustered view represents mined sequential patterns of enrollment and performance, illustrating common sequences. The directed view represents the highest level of abstraction and uses two visual elements, heat maps and a vector field, to illustrate overall performance in individual events and movement through the program. Results from multiple cohorts can then be superimposed on the same visualization to enable easy comparisons between patterns. Together, these abstractions provide a focus+context view of student performance, retaining outliers and emphasizing common patterns to illuminate dominant and unique patterns between cohorts of students.

This approach can help educators better understand student progress through the program, performance in individual courses, or student-selected course sequencing and this information can be used to address deficiencies in preparation, skills, or prerequisites. To demonstrate the appropriateness of this approach, performance and enrollment patterns are explored in the Computer Science program at the University of Maryland, Baltimore County. Specifically, this work examines the Gateway policy that requires students to earn a B or higher in the first two required programming courses before progressing with the hopes of validating the existing Gateway but also exploring other possible Gateway courses. Other issues explored within the Computer Science program include race, gender, math placement, and high school scores with the goal of attracting and retaining a more diverse group of students.

Committee: Penny Rheingans (chair), Marie desJardins, Marc Olano, Tim Finin and Diane Lee