jalchr / dragon

C# implementation for Naive Bayes sentiment classification engine (evidence files included)

Dragon Sentiment Classifier

Dragon Sentiment Classifier is primarily used in the Product Review Search App - Solvy http://solvy.cloudapp.net/

Description

Dragon Sentiment API is a C# implementation of the Naive Bayes Sentiment Classifier to analyze the sentiment of a text corpus. Sentiment analysis calculates the attitude or opinion towards something, such as a product, location, organization or person. This API provides easy to use mechanism to identify the positive or negative sentiment of an input document. Please note that this API works best on a large corpus of words (e.g. product reviews or blogs with 1000+ words) and targeted towards electronic/gadget reviews.

Training the Dragon

Dragon API is a machine learning algorithm that first needs to be taught how to classify a random collection of words and this training is performed using a couple of included evidence files (Postive.Evidence.csv and Negative.Evidence.csv) that contain the frequency map for words that commonly occur in electronic gadget reviews.

Getting Started

In order to classify plain text contents:

//positiveReviews and negativeReviews are the training set used by our Dragon Classigfier
var positiveReviews = new Evidence("Positive", "Repository\\Positive.Evidence.csv");
var negativeReviews = new Evidence("Negative", "Repository\\Negative.Evidence.csv");

//Instantiate the classifier using the training data set
var classifier = new Classifier(positiveReviews, negativeReviews);

//testData – String variable that contains the readable plain text contents of the document that needs to be classified (Strictly no HTML)
//The second parameter is the list of words that are excluded to improve classification performance
var scores = classifier.Classify(testData, DragonHelper.DragonHelper.ExcludeList);
Console.WriteLine("Positive Score - " + scores["Positive"]);