A tensorflow implementation of Richard Socher's paper(2012): Semantic Compositionality through Recursive Matrix-Vector Spaces

1. [TODO] The parameters given in the files are not the best one. Tuning the parameters is still a work to do. Note that under the setting in the current file, we obtain a training accuracy of 0.958, and test accuracy of 0.575, Overfitting!!!
2. I write the code based on code from stanford cs224d (assignment 3). So I use the same dataset as that one.
3. [TODO] There are still some tricks in the paper that haven't applied in this implementation, such as word representation's initialization, training approach, etc.