https://github.com/mattdeitke/cvpr-buzz
🐝 Explore Trending Papers at CVPR
https://github.com/mattdeitke/cvpr-buzz
computer-vision cvpr cvpr-buzz cvpr2021 d3 emotionjs gatsby graphql react tsx typescript
Last synced: 2 months ago
JSON representation
🐝 Explore Trending Papers at CVPR
- Host: GitHub
- URL: https://github.com/mattdeitke/cvpr-buzz
- Owner: mattdeitke
- License: mit
- Created: 2021-06-12T08:51:01.000Z (about 4 years ago)
- Default Branch: main
- Last Pushed: 2021-06-22T05:39:43.000Z (about 4 years ago)
- Last Synced: 2025-04-10T19:17:47.146Z (3 months ago)
- Topics: computer-vision, cvpr, cvpr-buzz, cvpr2021, d3, emotionjs, gatsby, graphql, react, tsx, typescript
- Language: TypeScript
- Homepage: https://mattdeitke.com/cvpr-buzz
- Size: 13.1 MB
- Stars: 52
- Watchers: 1
- Forks: 3
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
🐝 CVPR Buzz 🍯Explore Trending Papers at CVPR 2021
## 📊 Dataset
The scraping code is in `tasks.py`. Data is cached on the website, which makes it extremely fast to use with GraphQL and allows for fewer dependencies to be relied on.
#### 👁️ CVPR Accepted Papers
The accepted papers and their abstracts are extracted from [CVPR](https://openaccess.thecvf.com/CVPR2021?day=all).
#### 📖 Citation Data
Citation data comes from [Semantic Scholar](https://semanticscholar.org). There is no easy way to go from the paper title to Semantic Scholar's paper ID (i.e., it's not possible from the API). So we just search for it with Selenium and apply a few checks to see if it's the same paper.
This may take an hour or so to get through all the papers. There are also occasional hit limits where you may have to pick up from a point. Thus, one may need to monitor the window that opens. (It's possible to automate this by checking for the existence of certain elements on the screen, though I haven't been bothered enough by it to implement this.)
With the paper ID, we can then use [Semantic Scholar's API](https://api.semanticscholar.org/) to easily fetch the number of citations for the paper.
To fetch the engagement for each paper on Twitter, [Twint](https://github.com/twintproject/twint) is used. Currently there are 2 queries for each paper:
1. **Paper title.** Searches for the title of the paper in quotes (e.g., ["VirTex: Learning Visual Representations From Textual Annotations"](https://twitter.com/search?q=%22VirTex%3A%20Learning%20Visual%20Representations%20From%20Textual%20Annotations%22&src=typed_query&f=live)). Paper titles are unique enough that I've found it extraordinarily rare for there to be a tweet with the title of a paper, and it not actually being about the paper. For papers where the title is a common phrase, I have attempted to remove its results.
2. **ArXiv URL.** Search for where the arXiv URL has been shared in quotes (e.g., ["arxiv.org/abs/2006.06666"](https://twitter.com/search?q=arxiv.org%2Fabs%2F2006.06666&src=typed_query&f=live)).Each unique tweet is only counted once.
One can now also add tweets manually (keep reading), which aren't caught in the above criteria.
## 👊 Adding Data
_Please open a PR to add new data!_
#### 🐤 Tweets
If you want to index Tweets, open a PR and add them to [manual-data.json](https://github.com/mattdeitke/cvpr-buzz/blob/main/manual-data.json). All that is needed is the account username and the `tweetId`.
For instance, if `https://twitter.com/quocleix/status/1349443438698143744` is the Tweet, one would format it as `["quocleix", "1349443438698143744"]`. The paper IDs (a.k.a. the keys) can be found in `paper-data`.
#### 🏙 Citation Data
Citation data comes from Semantic Scholar. If your paper is on Semantic Scholar, but it is not showing up on the site, please edit the `s2id` field in `paper-data/{paperId}.json`. The `s2id` is found at the tail end of the Semantic Scholar URL.
For instance, if the Semantic Scholar URL is `https://www.semanticscholar.org/paper/Meta-Pseudo-Labels-Pham-Xie/43497fe8aa7c730e075b08facc2aa560a6d4dd85`, the `s2id` would be `43497fe8aa7c730e075b08facc2aa560a6d4dd85`.
## 📄 License
MIT License.