This is our code for Microsoft Beauty Of Programming 2017 Competition.
Our main contribution:
- Propose a sequence to sequence approach to judge the Matching Degree of question-answer pair.
- Propose an algorithm to answer questions based on a seqtoseq approach and information distilling techniques.
- Develop a bot service based on Microsoft Azure for answering questions of everything about SJTU.
- This project ranks 13/1198 in Microsoft Beauty of Programming 2017.
Folder data contains train and validation dataset for question-answer pair, as well as data we get from web.
Folder model contains model of the LSTM trained on the train dataset and a similar implementation for LSTM in keras.
Folder script contains the script for training, testing.
If you have questions or ideas, just create an issue. Ackonwledegment for my teammates Weichao Mao, Zhiyuan Tang and Xiang Song .