search

FlowBoat / FlowLabs-Malarkey

A project that aims to sniff and snuff out the bullshit on the web.
MIT License
0 stars 0 forks source link

Research on Sentiment Neurons #3

Open Atif-Mahmud opened 6 years ago

Atif-Mahmud commented 6 years ago

Discussion on Sentiment Neurons and the best way to classify the sentiment of a series of words.

Atif-Mahmud commented 6 years ago

As I mentioned before, there is a great article visible here, that deals with the exact thing we're looking into.

Atif-Mahmud commented 6 years ago

READ THIS PAPER: There is so much to learn

Atif-Mahmud commented 6 years ago

We first trained a multiplicative LSTM with 4,096 units on a corpus of 82 million Amazon reviews to predict the next character in a chunk of text. Training took one month across four NVIDIA Pascal GPUs, with our model processing 12,500 characters per second. These 4,096 units (which are just a vector of floats) can be regarded as a feature vector representing the string read by the model. After training the mLSTM, we turned the model into a sentiment classifier by taking a linear combination of these units, learning the weights of the combination via the available supervised data.

@88alex @zmct What does it mean by "linear combination"?

Atif-Mahmud commented 6 years ago

It's all open source, no one has done anything cool with this yet. I'm super excited!

Original REPO PyTorch Port