用 PageRank 跑 arXiv 上面 CS paper 的排名

在「Ask HN: AI/ML papers to catch up with current state of AI?」這邊看到的,本來只是在討論有哪些 AI/ML paper 可以看,結果在 id=38654200 這邊看到這個網站,上面的資料是每天更新一次:

https://trendingpapers.com/

This tool can help you find what's new & relevant to read. It's updated every day (based on ArXiv).

You can filter by category (Computer Vision, Machine Learning, NLP, etc), by release date, but most importantly, you can rank by PageRank (proxy of influence/readership), PageRank growth (to see the fastest growing papers in terms of influence), total # of citations, etc...

依照「Frequently Asked Questions」的說明,是用 PageRankarXiv 上面的 paper,主要是 CS 為主。

難得看到 PageRank 出現而且是用在 paper citation 上面...

引用自己論文的問題...

Nature 上點出來期刊論文裡自我引用的問題 (這邊的自我引用包括了合作過的人):「Hundreds of extreme self-citing scientists revealed in new database」。

開頭舉了一個極端的例子,Vaidyanathan 的自我引用比率高達 94%,而學界的中位數是 12.7%,感覺是有某種制度造成的行為?

Vaidyanathan, a computer scientist at the Vel Tech R&D Institute of Technology, a privately run institute, is an extreme example: he has received 94% of his citations from himself or his co-authors up to 2017, according to a study in PLoS Biology this month. He is not alone. The data set, which lists around 100,000 researchers, shows that at least 250 scientists have amassed more than 50% of their citations from themselves or their co-authors, while the median self-citation rate is 12.7%.

會想要提是因為想到當年 Google 的經典演算法 PageRank,就是在處理這個問題... 把 paper 換成 webpage 而已。

把 npm 的 dependency 當作 PageRank 的資料來源,分析 npm 目前的生態...

在「An Analysis of the JavaScript Package Ecosystem npm」這篇看到作者把 npm 的 dependency 當作資料來源,計算出 npm 的 PageRank:

可以看到 Underscore.js 的 PageRank 一直都維持在第一位... 這個方法頗有趣的,不知道有沒有其他語言的 :o

Google PageRank 資料將不再公開

Google 將不再對外公開 PageRank 資訊:「Google has confirmed it is removing Toolbar PageRank」與「RIP Google PageRank score: A retrospective on how it ruined the web」。

PageRank 資訊是透過 Google Toolbar 再反向被挖出來的,而 Toolbar 上的資訊將會拿掉,也預期對應的 API 應該也會關閉:

Google has confirmed with Search Engine Land that it is removing Toolbar PageRank. That means that if you are using a tool or a browser that shows you PageRank data from Google, within the next couple weeks it will begin not to show any data at all.

Google 內部還是會用,只是不會公開了...

Google 發表計算網頁真實性的演算法 (Knowledge-Based Trust)

Slashdot 上看到 Google 發表了計算網頁真實性的演算法,Knowledge-Based Trust (KBT):「Google Wants To Rank Websites Based On Facts Not Links」,原始的論文 PDF 檔案可以在「Knowledge-Based Trust: Estimating the Trustworthiness of Web Sources」這邊取得。

論文本身的原理不難懂 (其實方法相當有趣),主要是給出了三個貢獻。

首先是能夠區分是取出資訊的方法有問題 (extract 的演算法不夠好),或是網站本身就給出錯誤的資訊:

Our main contribution is a more sophisticated probabilistic model, which can distinguish between two main sources of error: incorrect facts on a page, and incorrect extractions made by an extraction system.

第二個則是在效能上的改善:

Our second contribution is a new method to adaptively decide the granularity of sources to work with: if a specific webpage yields too few triples, we may aggregate it with other webpages from the same website. Conversely, if a website has too many triples, we may split it into smaller ones, to avoid computational bottlenecks (Section 4).

第三個則是提出好的分散式演算法,可以螞蟻雄兵計算出來:

The third contribution of this paper is a detailed, large-scale evaluation of the performance of our model.

KBT 並不是要取代 PageRank,而是跟 PageRank 互相配合,可以有效打擊內容農場 (Content farm) 這類網站,畢竟 PageRank 的假設在一般的狀況下是有邏輯的。

在「High PageRank but low KBT (top-left corner)」這段講到了這件事情:

We consider the 15 gossip websites listed in [16]. Among them, 14 have a PageRank among top 15% of the websites, since such websites are often popular. However, for all of them the KBT are in the bottom 50%; in other words, they are considered less trustworthy than half of the websites. Another kind of websites that often get low KBT are forum websites.

再找時間細讀其他類似的演算法...