Learning Structured Text Representations

Code for the paper:

Learning Structured Text Representations
Yang Liu and Mirella Lapata, Accepted by TACL

Dependencies

This code is implemented with Tensorflow and the data preprocessing is with Gensim

Document Classification

Data

The pre-processed YELP 2013 data can be downloaded at https://drive.google.com/open?id=0BxGUKratNjbaZjFIR1MtbkdzZVU

Preprocessing

To preprocess the data, run

python prepare_data.py path-to-train path-to-dev path-to-test

This will generate a pickle file, the format for the input data can be found in the sample folder

Training

python cli.py --data_file path_to_pkl --rnn_cell lstm --batch_size 16 --dim_str 50 --dim_sem 75 --dim_output 5 --keep_prob 0.7 --opt Adagrad
--lr 0.05 --norm 1e-4 --gpu -1 --sent_attention max --doc_attention max --log_period 5000

This will train the Tree-Matrix structured attention model in the paper on the training-set and present results on the devset/testset

License

MIT