This started as a README for a potential open source project. I decided to make a a blog post to maybe get some feedback or suggestions.
Just an idea at this point.
There are "readability" libraries such as:
These take a webpage containing an article and strip out all the navigation and ads. Most of these work great for pages that are articles and some, like Mozilla's do a pretty good job of identifying when a page is NOT an article. However, there isn't much these libraries can currently do with pages that are lists of articles like:
Here, I'm considering making a library that strips out extraneous side navigation/ads/other junk from web pages like these, and either returning a list of article URLs or a very simple page with article links in a bulleted list.
By combining this with a standard readability algorithm/library, one could create a simple and/or text-only view of a website than what is typically rendered by text browsers like w3m and lynx.
A browser built using this could sit in a missing place in the continuum of browser complexity:
Offpunk < THIS THING < w3m/lynx/elinks < visurf/netsurf/surf < qutebrowser/chrome/firefox
- Start with a given HTTP/S URL
- retrieve that page
- Parse the page and get a list of all the links on that page
- Remove any links that are to domains other than the one from the original link
- Retrieve a few, say 10,maybe random links from the remaining of the list
- Get a similar list of links for each of those pages
- Get a list of the links that are common across all of the retrieved pages. It is a reasonable assumption that these would we be navigational links.
- Finally, we'd return to the list of links from the original page, and remove the links we've determined are navigational
- At this point, we're left with links only to unique content pages!
- If a website makes heavy use of cross linking in articles, those articles may be unfairly excluded from the final list of articles
- For some sites, this would be redundant with RSS/Atom feeds a be of lower quality