GraphLab Sentiment Analysis?

User 512 | 10/15/2014, 9:59:35 PM

I know that GraphLab is able to extract text topics, but can it do sentiment analysis on the texts? For example, suppose I have a table below:

Customer Comment 1 I like this product 2 This product is not good

Then after the text analysis, the output table will be

Customer Comment Sentiment 1 I like this product Positive 2 This product is not good Negative

I cannot find much information on this. Any help would be appreciated.

Comments

User 794 | 10/15/2014, 11:44:54 PM

As far as I can tell, there are no sentiment analysis tools in GraphLab, but you should be able to build your own classifier using public data, or use a third party classifier.

A couple of hints for building your own classifier that can work "pretty-well":

If you have labeled data: - TF-IDF transform your text data using N-gram lengths between [1,3] and train a classifier using the TF-IDFed data and the sentiment labels

If you have some labeled and a lot of unlabelled data: - Train a topic model on your labeled + unlabelled data. - Transform your labeled data to the "topic space" - Train a sentiment classifier on the transformed data

To find some training data, just google search for sentiment analysis datasets. Some examples of free datasets: http://ai.stanford.edu/~amaas/data/sentiment/ http://www.cs.cornell.edu/people/pabo/movie-review-data/


User 512 | 11/7/2014, 5:58:43 PM

Thanks for the information! This is very helpful.