PROJECTS


  • Digital Stock Market

    [Carnegie Mellon University]

    Working on building a social web service that allows users to interact with web entities, through the 'investment' of digital funds. 'Investing' is being treated as an evolution of current social interactions such as 'Like-ing' or 'Share-ing', with distinct advantages.

  • Efficient Approximate Page Rank on a Very Large Graph

    [Carnegie Mellon University]

    Implemented a scalable Personalized PageRank algorithm on a very large graph using graph subsampling techniques. PageRank was approximated using "pushes" involving a residual score for each node, as well as a current approximation for the PageRank during each iteration. The size of the graph was large, of the order that even the complete vertex-weight vector would not fit in memory.

    Technologies used: Hadoop, Java

  • Phrase Finding with Hadoop

    [Carnegie Mellon University]

    Worked on being able to detect important phrases in a given corpus using statistical language models. Used pointwise KL-divergence between multiple language models to extract key phrases. Phrases were detected based on the "phraseness" and "informativeness" scores, as discussed by Tomokiyo and Hurst. Extracted phrases were ranked based on the unified phraseness and informativeness scores.

    Technologies used: Hadoop, Java

  • Lazy Learning for Opinion Classification on Product Reviews

    [Carnegie Mellon University]

    Used lazy learning techniques to perform sentiment analysis on product reviews to predict whether a product was sought after based on the reviews it received. Product reviews used for this project were reviews from Amazon. Lazy learning techniques were used since the nature of product reviews is such that there are always new reviews coming in, thus requiring new models to be built every time fresh data is to be considered. This project was documented as an academic research paper. Weka was used to run various lazy learning algorithms on the available dataset.

    Technologies used: Weka, Java

  • Implementation of Information Retrieval Models

    [Carnegie Mellon University]

    Used the Lucene API in Java to implement various information retrieval models such as BM25, Indri, Ranked and Unranked Boolean. Also performed query expansion and pseudo-relevance feedback to improve search results quality. The dataset used for this project was a 10% sample of Wikipedia.

    Technologies used: Lucene API, Java

  • Collaborative Filtering for Movie Recommendations

    [Carnegie Mellon University]

    Worked on using collaborative filtering to develop a movie recommendation system. Implemented memory-based and model-based methods to predict the ratings of movies. Used Pearson's Correlation Coefficient (PCC) to account for user and movie bias in the ratings. Evaluation was performed on a subset of the Netflix Prize dataset.

    Technologies used: Java

  • Learning To Rank (LETOR)

    [Carnegie Mellon University]

    Constructed SVM and Logistic Regression models for Learning To Rank (LETOR). Evaluated models using rank-based evaluation metrics - Normalized Discounted Cumulative Gain (NDCG), Mean Average Precision (MAP) and Precision@N (P@N). The experiments were run on a dataset from Microsoft Research Asia.

    Technologies used: Java

  • Text Tokenizer

    [Carnegie Mellon University]

    Built tokenizers to tokenize Twitter Data and News Commentary Data, using OpenFST. The tokenizers were capable of tokenizing words on white space, on punctuation and individual character wise.

    Technologies used: OpenFST

  • Chronic Kidney Disease Analysis

    [Carnegie Mellon University]

    Analysis of social media sites discussing kidney diseases, in order to extract information regarding topics being discussed, patients' reactions to treatments administered etc. Involves usage of text mining, sentiment analysis, machine learning and information extraction techniques.

    Technologies Used: C++, Indri API

  • Load Shedding in Smart Grids: A Game Theoretic Approach

    [Indian Institute of Science]

    This is a project I worked on for my senior year thesis. The project involved devising a fair load shedding algorithm in smart grids using solution concepts from Cooperative Game Theory, building on the Bankruptcy problem and Contested Garment Rule.

  • Paint Application using OpenGL

    [BMS College of Engineering]

    This project involved development of a simple paint application using OpenGL Library Functions. The application allows the user to select drawing color, pen type, etc Inbuilt drawing templates are available. External images can be imported into the drawing window.

  • Text Editor

    [BMS College of Engineering]

    The goal of this project was to develop a functional text editor with functionality to create, view, edit and save text documents.