Context-to-Session Matching

This repo contains the code and data for paper Context-to-Session Matching: Utilizing Whole Session for Response Selection in Information-Seeking Dialogue Systems in KDD 2020.

Code

1. Preprocess: cd utils ; python compose_data.py
2. The configure is located in main.py (detailed introductions are in main.py)
3. How to train: python main.py train
4. How to test: python main.py test TestRandNegCand $chenkpoint_file

Data Structure

1. The data sets are located in data_ali directory.
   • cc.cc.train(.zip) represents the train set for CSM, please uncompress it first.
   • cc.cr.train(.zip) represents the train set for CRM, please uncompress it first.
   • cc.cc.{valid}/{test}/{test.human} represent the valid/TestRandNegCand test/TestRetrvCand test set for CSM and CRM
   • vectors.txt contains the pre-trained word embedding.
2. Format of each line in the files: query context|response context|response|label
3. Please note that: the labels in cc.cc.test.human are all 1 and the right human annotated labels of TestRetrvCand are in human/human.res.crowd

ACK

The code is developed referring DAM.