Nonparametric Deconvolution Models

Decomposition models decompose observations into constituent parts by representing observations as a product between group representations and factor features. We are working on deconvolution models, which similarly decompose, or deconvolve, observations into constituent parts, but also capture group-specific (or local) fluctuations in factor features.

Algorithmic Confounding in Recommendation Systems

Recommendation systems occupy an expanding role in everyday decision making, from choice of movies and household goods to consequential medical and legal decisions. The data used to train and test these systems is algorithmically confounded in that it is the result of a feedback loop between human choices and an existing algorithmic recommendation system. We are currently exploring the impact of algorithmic confounding in this context.

How Algorithmic Confounding in Recommendation Systems Increases Homogeneity and Decreases Utility
arXiv preprint.

Historical Event Detection

Historians and political scientists commonly read large quantities of text to construct an accurate picture of significant events. Our methods help historians identify possible events from the texts of historical communication.

Detecting and Characterizing Events
EMNLP, 2016.

Capsule project page (includes code) or Capsule code directly on GitHub

Visualization demo



Capsule slides (generalized from multiple invited talks)

Who, What, When, Where, and Why? A Computational Approach to Understanding Historical Events Using State Department Cables
Text As Data, 2015. (slides)

Social Poisson Factorization

The downside to most algorithmic recommendations is that, for some people, part of the appeal of reading, watching, or consuming other media is in creating shared experiences with friends. We incorporate the ratings of friends (and not just friends' general preferences) in providing personalized recommendations.

A Probabilistic Model for Using Social Networks in Personalized Item Recommendation
RecSys, 2015. (slides)

SPF project page (includes code) or SPF code directly on GitHub



A Probabilistic Model for Using Social Networks in Personalized Item Recommendation
Invited Talk at BYU, 2015.

Poisson Trust Factorization for Incorporating Social Networks into Personalized Item Recommendation
NIPS Workshop: What Difference Does Personalization Make?, 2013.

Poisson Trust Factorization for Incorporating Social Networks into Personalized Item Recommendation
General Exam presentation, 2013.

Recommendations for Groups

We performed a large-scale study of television viewing habits, focusing on how individuals adapt their preferences when consuming content with others. We constructed a simple model for estimating how individual preferences are combined in group settings.

A Large-scale Exploration of Group Viewing Patterns
TVX (ACM International Conference on Interactive Experiences for Tekevision and Online Video), 2014.

Mining Large-scale TV Group Viewing Patterns for Group Recommendation
Microsoft Reseach Technical Report, 2013.

Visualizing Topic Models

Topic modeling is a machine learning method that learns underlying themes in a collection of documents, which can be used to summarize and organize the documents. We have created a method for visualizing topic models, allowing users to explore a corpus by navigating between high level topic descriptions and individual documents, hopefully deepening their understanding of the corpus.

Visualizing Topic Models
International AAAI Conference on Social Media and Weblogs, 2012.

Original TMVE code

Online TMV code

Wikipedia Demo