Publications by topic

My primary research interest is in applying machine learning and data mining techniques to practical problems, particularly those with applications to personalization that preserve privacy and trust. One aspect of my dissertation research is combining individual preference with transportation network and spatial-temporal knowledge to learn individual activity patterns for traveler personalization. Earlier work focused on ways personalization techniques such as recommender systems could be made more secure to malicious bias and thus more trustworthy while still remaining open systems. Below are my publications in these areas. (Superseded papers are not listed.)

Contents:

• Learning Activity Patterns of Individuals
• Secure recommender systems
• Miscellaneous

Learning Activity Patterns of Individuals

Learning Activity Patterns of Individuals (Ph.D. Dissertation, 2010) (updated 10/29/2009)

Algorithms and techniques for quickly learning activity patterns of individuals. The focus of this study is leveraging transferable aspects of travel behavior and patterns to reduce learning time, while also creating a richer model of the individual traveler. This research effort identifies algorithms and techniques needed to address the problem of learning and predicting the activity needs of an individual for anticipating their associated travel demands with little input required from the traveler. A major component of this work is the theoretical aspect of making better time series projections of discrete sets despite missing data. The goal of this work is to enable intelligent travel applications by providing insight into an individual's future travel plans and scheduling preferences. A major component of this effort is to provide this insight without compromising user privacy.

New! Urban Travel Route and Activity Choice Survey (UTRACS): An Internet-Based Prompted Recall Activity Travel Survey using GPS Data (To appear in Proceedings of 89th Annual Meeting of the Transportation Research Board, 2010)

Presents the results of an internet-based prompted recall activity-travel survey using GPS data collection combined with a short activity preplanning and scheduling survey. Results reinforce findings that GPS surveys can passively capture correct travel information as well as valuable data on activities and schedule planning, thus reducing participant burden.

New! Urban Travel Route and Activity Choice Survey (UTRACS): An Internet-Based Prompted Recall Activity Travel Survey using GPS Data (tentatively accepted for publication in Transportation Research Record, 2010)

Presents the results of an internet-based prompted recall activity-travel survey using GPS data collection combined with a short activity preplanning and scheduling survey. Results reinforce findings that GPS surveys can passively capture correct travel information as well as valuable data on activities and schedule planning, thus reducing participant burden.

New! Attribute Constrained Rules For Partially Labeled Sequence Completion (Advances in Data Mining - Applications and Theoretical Aspects, 2009)

Introduces a new constraint technique for mining a special form of sequential association rules specifically for more accurate predictions when data sources suffer from missing data. The attribute constrained rules introduced in this work are shown to significantly improve rule performance and robustness over traditional association rules mining when even large amounts of sequence data are missing.

An Automated GPS-Based Prompted Recall Survey With Learning Algorithms (Transportation Letters: The International Journal of Transportation Research, 2009)

This paper documents recent developments in the field of GPS travel surveying and ways in which GPS has been incorporated into or even replaced traditional household travel survey methods. A new household activity survey is presented which uses automated data reduction methods to determine activity and travel locations based on a series of heuristics developed from land-use data and travel characteristics.

Mining Sequential Association Rules for Traveler Context Prediction (Proceedings of the First International Workshop on Computational Transportation Science, 2008)

A technique is introduced based on sequential data mining for predicting multiple aspects of an individual's next activity using a combination of user history and their similarity to other travelers. The proposed technique is empirically shown to perform better than more traditional approaches to this problem.

Secure recommender systems

Defending recommender systems: detection of profile injection attacks (Service Oriented Computing and Applications, 2007)

A number of attributes are identified that distinguish characteristics present in attack profiles in general, as well as an attribute generation approach for detecting profiles based on reverse engineered attack models. Three well-known classification algorithms are then used to demonstrate the combined benefit of these attributes and the impact the selection of classifier has with respect to improving the robustness of the recommender system.

Toward trustworthy recommender systems: An analysis of attack models and algorithm robustness (ACM Transactions on Internet Technology, 2007)

Extensive review of some of the major issues in building secure recommender systems, concentrating in particular on the modeling of attacks and their impact on various recommendation algorithms. Our study shows that both user-based and item-based algorithms are highly vulnerable to specific attack models, but that hybrid algorithms may provide a higher degree of robustness. Using our formal characterization of attack models, we also introduce a novel classification-based approach for detecting attack profiles and evaluate its effectiveness in neutralizing attacks.

Classification features for attack detection in collaborative recommender systems (KDD '06: Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, 2006)

Different attributes derived from user profiles are studied for their utility in attack detection. We show that a machine learning classification approach that includes attributes derived from attack models is more successful than more generalized detection algorithms previously studied.

Analysis and Detection of Segment-Focused Attacks Against Collaborative Recommendation (Advances in Web Mining and Web Usage Analysis, 2006)

An extended examination of the segment attack, where a subset of users with similar tastes can be targeted in a highly successful against both user-based and item-based collaborative filtering. We also introduce a detection model that can significantly decrease the impact of this attack.

The Impact of Attack Profile Classification on the Robustness of Collaborative Recommendation (Proceedings of the 2006 WebKDD Workshop, 2006)

Detailed analysis of the information gain associated with several detection attributes across the dimensions of attack type and profile size. Combined effectiveness is evaluated at improving the robustness of user based recommender systems.

Detection of Obfuscated Attacks in Collaborative Recommender Systems (Proceedings of the ECAI'06 Workshop on Recommender Systems, 2006)

Several techniques are proposed that an attacker might use to obfuscate their attack to avoid detection. We show that these obfuscated versions can be nearly as effective as the reverse-engineered models yet harder to detect, and discuss alternate approaches to reducing the effectiveness of such attacks.

Securing Collaborative Filtering Against Malicious Attacks Through Anomaly Detection (Proceedings of the 4th Workshop on Intelligent Techniques for Web Personalization (ITWP'06), 2006)

In this paper we examine approaches for detecting suspicious rating trends based on statistical anomaly detection and the effects of rating distribution on detection performance.

Profile Injection Attack Detection for Securing Collaborative Recommender Systems (Masters Thesis, 2006)

A technique for detecting and reducing the impact of attack profiles on recommender systems is proposed and evaluated against the dimensions of attack type, attack intent, filler size, and attack size. The weaknesses of such a supervised classification scheme are experimentally explored and techniques that can be combined with this approach to address these vulnerabilities are discussed.

Detecting Profile Injection Attacks in Collaborative Recommender Systems (Proceedings of the 8th IEEE Conference on E-Commerce Technology (CEC'06), 2006)

A classification approach to the problem of detecting and responding to profile injection attacks is introduced. The proposed technique is shown to significantly reduce the effectiveness of the most powerful attack models previously studied.

Evaluation of Profile Injection Attacks In Collaborative Recommender Systems (DePaul CTI Research Symposium / Midwest Software Engineering Conference (CTIRS/MSEC 2006), 2006)

A probabilistic latent semantic analysis and a supervised classification approach are introduced for reducing the impact of the segment attack and other traditional attack models.

Segment-Based Injection Attacks against Collaborative Filtering Recommender Systems (Proceedings of the 2005 International Conference on Data Mining (ICDM'05), 2005)

In this paper, we examine “segmented” attacks which concentrate on a targeted set of users with similar tastes, biasing the system’s responses to these users. We show that such attacks are both pragmatically reasonable and also highly effective against both user-based and item-based algorithms.

Collaborative Recommendation Vulnerability to Focused Bias Injection Attacks (Proceedings of the Workshop on Privacy and Security Aspects of Data Mining, 2005)

This work demonstrates attacks that concentrate on a targeted set of users with similar tastes, such as the segment attack, can be highly effective against both user-based and item-based collaborative filtering even with little knowledge of the specific system being targeted.

Effective Attack Models for Shilling Item-Based Collaborative Filtering Systems (Proceedings of the 2005 WebKDD Workshop, 2005)

Introduced an attack model, segment attack, that was highly effective at introducing bias in item-based collaborative filtering. This work demonstrated that even item-based recommenders, which were previously thought secure, compared to user-based systems, were still susceptible to bias profile injection attacks.

Miscellaneous

Genetically Evolving Optimal Neural Networks (Neural Networks and Expert Systems, 2007)

This paper examines genetic algorithms research that has focused on addressing the challenges of neural network configuration. Specific focus is paid to techniques that have been used to encode the problem space in order to optimize neural network configurations. A summary of the results as well as areas for future research in this field are also discussed.