Newsfeed_Sanity

A simple and sane newsfeed for m.phys.org - or how to quickly recommend the best topics

Requirements

• pandas
• numpy
• bs4
• cython
• gensim
• scipy
• webbrowser
• jupyterlab
• https://code.google.com/archive/p/word2vec/ (download: GoogleNews-vectors-negative300.bin)

Goal

During the last few years, I was spending much time to read once per week the newsfeed from http://m.phys.org. That's a good exercise for keeping touch with the latest advancements in Science. But with the drawback that it's taking time, a lot of time, and mainly because I had to go through a long list of different topics that I didn't need. Also, I needed a newsfeeder (recommander) which respects my tastes without having to go through undesired proposals.

Solution

Compress my reading habits using word2vec from gensim in order to automatically extract the relevant science subjects to me. Also, as I was manually scrapping the news through the http://m.phys.org website, I was of course tempted to open this or this article according to its title. This is fairly simple I know, but this is generally what we do, right? So let's start.

The notebook

The notebook newsfeed_sanity.ipynb above performs the following tasks:

• Check your browsing history from Google Chrome for the last three months
• Extract the browsed webpages from http://m.phys.org
• Use gensim to vectorize the articles' topic with word2vec
• Compute the score = (1-distance) (higher is better here) between past and new topics
• Rank the new topics by score
• Display the new topics as hyperlinks in a html page.

Feel free to fork it, copy it, change it and improve it for your tastes and targeted websites.

This was an individual excercice, not a tutorial.