Structure:

.
├── materials: Contest descriptions
├── model: Baseline model
├── raw_data: ignored by .gitignore
├── data: ignored by .gitignore
├── utils: Data Preprocessing
└── README.md

Details can be found here

First clone the repo by

git clone ...

• Never commit to the master directly.
• Always do your work on a different branch from master.

• update your local repo with git pull origin master,
• create a working branch with git checkout -b MyNewBranch
• make your changes on your branch and stage them with git add,
• commit your changes locally with git commit -m "description of your commit", and
• upload the changes (including your new branch) to GitHub with git push origin MyNewBranch
• Go to the main repo on GitHub where you should now see your new branch
• click on your branch name
• click on “Pull Request” button (URC)
• click on “Send Pull Request”

Official description can be found here.

Given a question $q$, and a set of documents $D = {d_1, d_2, ..., d_n}$, we are expected to give an answer $a$ as close as possible to reference answers $Ar = {ar_1, ar_2, ..., ar_m}$, according to the evidences in document $$D$$.

It will have a set of evaluation metrics (Bleu-4, Rouge-L etc.) to measure the closeness of our answer $a$ to the reference answer $Ar$.

把答案加入encode：由于description类的问题的某些备选答案是携带语义信息的，故将备选答案也做encoding处理。