BigML at Predict Conference! 3rd October.

I am very excited to announce that this coming Tuesday at Predict Conference David Gerster Vice President of Data Science at BigML will speak. Here is some info on his talk. Tickets on sale.

Talk: Modeling with Latent Dirichlet Allocation

Overview: Many data science practitioners find themselves confronted with free text data. Often, the goal is simply to extract features from the text that are useful for predictive modeling, rather than performing a full semantic analysis. Latent Dirichlet Allocation is an unsupervised learning method that extracts groups of related words, or "topics", from a collection of documents. Surprisingly, the theme of each topic is usually obvious to a human reader, indicating that this method has uncovered real information about the documents.

Using a collection of 50,000 movie reviews, we extract topics from the review text, then use these topics to create a model that predicts the review sentiment with very high accuracy. We will also cover the conceptual underpinnings of LDA as originally introduced by Blei et. al.

Here is David speaking about BigML's machine learning approach.