Most parametric clustering algorithms in use today employ generative models that do not have a na... more Most parametric clustering algorithms in use today employ generative models that do not have a natural mechanism to give rise to overlapping clusters. The multiplicative mixture model has been recently proposed as a generative model that can naturally give rise to overlapping clusters. However, performing maximum likelihood parameter estimation for this model using a standard technique like expectation maximization is intractable. As a result, Monte Carlo algorithms have been developed to do parameter estimation. In contrast to these stochastic algorithms, we propose a complementary deterministic algorithm to perform approximate maximum likelihood parameter estimation in a tractable manner. We then derive an overlapping clustering algorithm that uses employs the multiplicative mixture model as a generative model.
Sentiment classification is one of the most challenging problems in Natural Language Processing. ... more Sentiment classification is one of the most challenging problems in Natural Language Processing. A sentiment classifier recognizes patterns of word usage between different classes and attempts to put unlabeled text into one of these categories in an unsupervised manner. Therefore, the attempt is to classify documents not by topic but by overall sentiment. We have used reviews of movies to train and test our classifier. Our system uses the Maximum Entropy method of unsupervised machine learning. We present our observations, assumptions, and results in this paper. We conclude by looking at the challenges faced and the road ahead.
Uploads
Papers by Priyank Patel