Open lihy-1224 opened 5 years ago
Tree_CNN_LSTM should be better, in my experiment. However the result variance (up and down), so I suggest you run each experiment multiple time (remember to clear cache data https://github.com/ttpro1995/TreeLSTMSentiment/issues/2)
https://github.com/ttpro1995/TreeLSTMSentiment was based on paper Improved Semantic Representations From Tree-Structured Long Short-Term Memory Networks.
This repo Tree_CNN_LSTM is based on my paper Combining Convolution and Recursive Neural Networks for Sentiment Analysis. If you cannot download paper from that link, download from my latex repo, the camera-ready version
For more very extensive detail, read my thesis
you should train Tree_CNN_LSTM longer, until it does not improve anymore, not just 10 epoch, but 100 epoch or so.
Don't just believe a stranger in everything. Please believe in yourself. Try to prove and confirm (or not) by yourself. Through pain, you will gain wisdom.
thanks for your kind advice!
I use Tree_CNN_LSTM model in Sentiment Classification task, but the accuracy seems to be lower than TreeLSTM model in your github repository. Is it right in your test?