Awesome

RWG_DV-Seq2Seq

Code for the Relevant Word Generator + Dual Vocabulary Sequence-to-Sequence generative framework.

As presented in paper: A Deep Generative Approach to Search Extrapolation and Recommendation

model_rwg.py: contains implementation for the RWG model.

model_dv_seq2seq.py: contains implementation for the DV_Seq2Seq model.

run_rwg.py: driver file for training/testing the RWG model.

run_dv_seq2seq.py: driver file for training/testing the DV_Seq2Seq model.

A pre-trained word vector file is needed. Put the file inside the libs/ folder and set the name using command line argument -word_vec_file

The word vector file must have a format that works with gensim.models.KeyedVectors.load_word2vec_format(...)

Due to privacy concerns, the original data cannot be released. Put your own data inside data/:

The expected data format for the RWG model (one line per instance):

Input words | target relevant words

The expected data format for the DV_Seq2Seq model (one line per instance):

Input words | relevant words | target output words

Words must be segmented by a single space

For each model there should be a train data file, a validation data file and a test data file, the filenames can be changed inside the driver files.

Requires pytorch 0.4.1+ and Python 3.