Awesome

Twitter Sentiment Analysis using ConvNet

A tool to

• predict sentiment "positiveness" for tweets

How to use it?

>> from sentiment import sentiment_score
>> print sentiment_score(u"I love you")
0.9999

It returns a sentiment index ranging from 0 (negative sentiment) to 1 (positive sentiment).

Online Demo

• predict sentiment "positiveness" for single tweets
• gain an overview of the "positiveness" of hashtags

Click here

Algorithm

Please refer to A Convolutional Neural Network for Modelling Sentences for more information about the algorithm.

Technical choices

• Tornado as the web framework
• Theano as the neural network training implementation
• Scipy as the neural network classification(online version) implementation

Training techniques

1. Fan-in, fan-out initialization
2. Dropout
3. AdaDelta

Contributors

Han Xiao and Yao Lu