Projects
Personalized News Recommendation
Java
The news personalization problem is solved using a multi-armed bandit problem. The aim of this project is to study the previous methods and improve this approach by using the context of each task to improve the personalization. We are developing an intelligent explore/exploit strategy to decide which news to show the user while the system is still learning their interests. The website can be found here.
Team members: Saloni Potdar and Li Zhou
Advisors: Prof. Emma Brunskill, Eric Nyberg(Capstone advisor)
Twitter Analytics
Java, AWS, Hadoop, MySQL
Built a robust REST web service on the AWS cloud to analyze 250GBs of twitter data.
Team members: Saloni Potdar and Navneet Rao
Expert Identification in MOOCs
Python, Java, MALLET, LIWC
Studying infuence in social networks is an important topic that has attracted the attention of a variety of researchers in different domains. We wish to identify the student leaders in MOOCs and we define leaders as important individuals who remarkably influence the language usage and its propagation process. We are exploring variious methods like influence propagation through language accommodation, topic models, syntactic and lexical features.
Team members: Saloni Potdar, Shane Moon and Lara Martin
Advisors: Prof. Carolyn Rose
Machine Learning Algorithms for Large Datasets
Java, Hadoop, AWS, Pig
Implemented Naive Bayes, Personalized Page Rank, Stochastic Gradient Descent in a memory constrained setting.
Improving Robustness of Classifiers using Ensemble of Ensembles
Java and Weka API
Built and evaluated a robust classifier learner that performed well with minimal tuning. We used 24 development datasets and 20 evaluation datasets for this project.
Team members: Saloni Potdar and Navneet Rao
Netflix Movie Recommender System
Java
Used collaborative filtering to predict the ratings of movies, and evaluated them on a subset of the Netflix Prize dataset.
Learning to Rank (LETOR)
MATLAB, SVM-Light
Implemented regularized logistic regression (RLR) with gradient descent, and ran experiments on adapted RLR and Support Vector Machine for LETOR.
Implementation of Information Retrieval Models using Lucene API
Java, Lucene API
Implemented IR models like the ranked and unranked boolean, BM25 and Indri. Also performed query expansion with pseudo relevance feedback.
