xiemeigongzi / graph-neural-networks.github.io Goto Github PK

EMNLP 2020 Virtual Conference

This virtual conference page is based on MiniConf by Alexander Rush and Hendrik Strobelt. It was extended by the amazing team of ACL 2020. The ACL version is the base for this repository.

The website is based on Flask and Frozen-Flask. It uses data files like .csv, .yaml or .json that contains the information about events, papers, ... to populate the HTML templates. From this it generates a static website which can then deployed easily via an HTTP server. We strongly recommend deploying it via Amazon CloudFront and using Amazon Cognito as the authentication provider. See here for a guide to do this.

Quick Start

pip install -r requirements.txt
pip install -r requirements-dev.txt
make run

When you are ready to deploy run make freeze to get a static version of the site in the build folder.

Project Structure

The repository consists of the following main components:

1. Datastore sitedata

Collection of data files representing the papers, speakers, workshops, and other important information for the conference.

2. Routing main.py

This file contains defines the Flask app and the routes

3. Templates templates

Contains all the pages for the site. See base.html for the master page and components.html for core components.

4. Frontend static

Contains frontend components like the default css, images, and JavaCcript libs.

5. Scripts scripts

Contains additional preprocessing to add visualizations, recommendations, schedules to the conference.

Pages

This section describes all pages that are in this version of MiniConf and how to customize them.

Index

This page is mainly configured via sitedata/config.yml. One mainly needs to change the conference name, date, number of workshops/tutorials and help documents here. Before the conference starts, also update the acknowledgements. In static/js/time-extend.js, you also need to change start and end times of the conference to make sure that daylight saving is handled correctly.

Schedule

The schedule is based on FullCalendar. The entries are automatically generated from the data of the paper sessions/workshops/plenary. For the weekly view, we compute blocks of events and show only a generic name, on the day view, they are shown as is. Additional events can be added via sitedata/overall_calendar.yml. If you add new event types, make sure to assign them a color in load_site_data::build_schedule.

Plenary

Plenary make up the official program of the conference. For EMNLP, we had live events and prerecorded keynote talks. Keynotes were prerecorded but livestreamed via SlidesLive. Panels were fully live and used CART real-time captioning. We did not add the prerecorded talks to the plenary page before the talk was done. Each plenary details page can either have one video or a list of videos. It can also have a RocketChat channel. The default is to just show one SlidesLive video. Refer to the ACL2020 repo for how multiple videos can be shown on a plenary details page.

Livestream

Plenary events were livestreamed via SlidesLive. Make sure that your contract with them covers this. Panels were fully live and used CART real-time captioning.

For EMNLP, we had prerecorded presentations and keynote speakers and live panels. Recordings were streamed live. For keynotes, after the prerecorded talk was shown, the keynote speaker, a volunteer and a SlidesLive person were in a Zoom call, this Zoom call was streamed after seamlessly after the keynote recording. Then the volunteer took questions from the #live RocketChat channel and asked the keynote speaker.

Before the livestream starts, change the livestream ID and CART URL on livestream.html. The Livestream SlidesLive IDs are different from the actual ID for the recording, e.g. if a keynote is livestreamed, then the livestream ID is different from the prerecorded ID. Make sure to update the ID before the stream start and ask SlidesLive to show a Livestream will start soon message. After the livestream is done, add the ID to the plenary event itself. You can use the livestream ID as a slideslive ID for the player, it will then show a recording of the livestream.

Papers

EMNLP hat 4 types of papers, main conference papers (long/short), CL, TACL, demo, workshop and findings papers. See below for the kind of things each paper had.

Demo papers can have arbitrary markdown under material to add info like code repo or screencast, see our demo_papers.csv for it.

For *ACL style conferences, we recommend to ask publication chairs early to give out a list of accepted papers with SoftConf ids, title, authors, paper type (long/short) and anthology IDs. Also ask them to give out the proceedings as soon as they are done, do not wait for them to be published in the ACL anthology page. We do not store PDFs directly, but link to the ACL anthology.

Visualization

We use SPECTER to generate document embeddings from abstracts and DyGIE++ to generate keywords. For the embeddings, we then use umap to project them to 2D. Recommentations are generated by using n-nearest neighbours.

We provide scripts/dataentry/projections.py to generate the projections, please refer to the respective repositories to find out how to install them. Install them in their seperate virtualenv, they all have conflicting dependencies.

Images

We extract images from the PDFs and upload them to Amazon S3. You can use our bad script under scripts/dataentry/extract_images.py to extract them. In order to allow authors to change their images, we set up an additional Github repository that contains the images. Pull requests there automatically deploy the update images to S3. Look at the Github workflow there and set the respective secrets for the autodeploy.

ConnectedPapers

Connected Papers is a visual tool to help researchers and applied scientists find academic papers relevant to their field of work. In addition to linking to the connected papers for each main paper, they built a custom page for EMNLP which is shown below every main paper presentation. If you want that also, then conference papers need to be indexed by SemanticScholar before the conference. Contact them via their website and ask nicely, then they will also help you.

Tutorials

Tutorials simply contain a schedule of the events, website, Zoom links, RocketChat and optional a prerecorded SlivesLive video. We asked tutorial organizers to fill in the information into Google Sheets and then use a script to download that sheet and parse it. You can refer to scripts/dataentry/tutorials.py and see how we loaded tutorials. We used this template for tutorial organizers to fill in. Tutorial blocks on the tutorial overview page and in the main schedule are computed automatically.

Workshops

Workshops contain a schedule of the events (can use markdown in there, website, Zoom links, RocketChat, a list of prerecorded SlivesLive videos for invited talks and a link to workshop papers. We asked workshop organizers to fill in the information into Google Sheets and then use a script to download that sheet and parse it. You can refer to scripts/dataentry/workshop.py and see how we loaded workshops. We used this template for workshop organizers to fill in. Workshop blocks on the workshop overview page and in the main schedule are computed automatically. We gave each workshop up to 5 Zoom links, but did not schedule events for them. Instead, we linked the personal meeting ID on our website.

Socials

We asked social organizers to fill in the information into Google Sheets and then use a script to download that sheet and parse it. You can refer to scripts/dataentry/socials.py and see how we loaded socials. We used this template for social organizers to fill in. Social blocks in the main schedule are computed automatically. We gave each social event one Zoom link or they could get a Gather room, but did not schedule events for them. Instead, we linked the personal meeting ID on our website.

Sponsors

Each sponsor has a booth that is handmade just for him. We assigned one volunteer to each sponsor to write their YAML. We collected them via Dropbox file request. Then we merged them into one file via scripts/dataentry/sponsors.py. The file format is the following:

Most of these are optional. If the sponsor should not get a booth, then you can link to their page via landingpage:

name: ISI
level: Bronze
logo: isi.png
website: https://www.isi.edu
channel: example_sponsor
landingpage: https://www.isi.edu/

Sponsor Accounts

Sponsors a number of visitor accounts based on their sponsor tier. They also can register 3 people that only exhibit. We recommend also creating one account per sponsor early enough so that they can see their booth live. We recommend naming it after the sponsor and not binding it to a specific person, e.g. better call it sponsor.deepmind. This account will then not be removed before the conference. Additional exhibitors need to register and can usually register for free.

Relevant documents

• Invitation to Dry Run
• EMNLP Instructions to sponsors

Chat

We use RocketChat throughout this page to have channels for keynotes, plenaries, live, papers, workshops, tutorials and many more. You can host RocketChat yourself, e.g. on AWS (refer to their guide) or pay them to host it. We normally first create a demo workspace and then upgrade it to full for one month with the number of anticipated users during the conference.

We integrate RocketChat via SSO into our Amazon Cognito user repository so that only one set of username and password is needed. For that, you can refer to this guide. You do not need to change the Lambda functions if you set up the project correctly when creating the AWS app.

You can refer to our checklist for RocketChat to have an idea what needs to be done.

We have scripts for RocketChat setup in scripts\rocketchat and here.

In order to use the Active Chat feature, you need to use RocketChat and host the statistics server. Please refer to scripts/channels-stats to see how. If you do not want it, remove it from base.html.

Help

Update the code of conduct to your own. The FAQ is generated by faq.yml.

Misc

Gather.Town

We use VirtualChair to manage Gather.Town for us. We strongly recommend to also book them. Ask them early enough. We use SSO with Gather.Town, you need to specifically ask for it and make sure that it lands in the contract.

Zoom

We buy one Master account and then create accounts for the specific events. For things like tutorials, socials and workshops we do not schedule events but use the personal meeting link. You can use these scripts to help you creating Zoom things.

Relevant documents

• Volunteers’ Documentation on Zoom
• Sponsors’ Documentation on Zoom
• Mentors’ Documentation on Zoom

User creation

For the first batch of accounts, we are given an Excel file and bulk create accounts. This is best to be done the week before the conference starts. Use these scripts. You might need to change custom:name to name and add support for affiliation if you choose to store it.

Late Registration

We use Zappier for that: After the first bulk creation, the registration company forwards a copy of all registration mails to us. Then we use Zappier to parse email, name, affiliation and host. This is sent to a AWS Lambda function that then creates the account for us. See also this issue.

Help Desk

We recommend setting up a helpdesk that is staffed by volunteers. For that, create a Google Group and add helpdesk members and also create a helpdesk channel. It is best to set the Google Group up early enough and also use it as the reply mail for the conference invitation account credentials mail and also to mention it there.

Favicons

We use this website to generate favicons from an image.

Acknowledgements

MiniConf was built by Hendrik Strobelt and Sasha Rush.

Thanks to Darren Nelson for the original design sketches. Shakir Mohamed, Martha White, Kyunghyun Cho, Lee Campbell, and Adam White for planning and feedback. Hao Fang, Junaid Rahim, Jake Tae, Yasser Souri, Soumya Chatterjee, and Ankshita Gupta for contributions.

It was extended by the ACL 2020 virtual infrastructure team, especially by Hao Fang and Sudha Rao with the help of many volunteers.

It was extended for EMNLP 2020 by Jan-Christoph Klie with the help of the EMNLP 2020 virtual infrastructure team and with the help of many volunteers.