Digital daily Scroll.in has launched a platform called SocialWire, that curates the top influential news stories of the day. The publication claims that it’s using a patent-pending collection of algorithms called the Scroll Machine to identify the most influential news stories from several social media platforms.
The software will create a “list of influencers” curated by its editors, and identify the most influential stories by going through ‘hundreds of thousands of users, stories and shares’. Scroll claims that currently, the software analyses over 100,000 posts every day. The top 10 influential stories on SocialWire are updated every few hours, combing through news sources and Twitter.
It will use an “influence score” on the scale of 0-100 to calculate the size of the influencer’s audience, news relevance, and the degree to which other influencers re-share their posts. It claims this will help solve the popular-only (most shared) posts problem. The score is normalised by mapping the share and follower count using a sigmoid function.
At first glance, the site seems to populate international and Indian news through Twitter influencers and displays a list of influencers on the right hand side, which is mostly politicians and journalists. Clicking a story takes the user to the Twitter influencer’s update or, if the tweet has been retweeted, the original tweet.
Machine learning algorithms to generate lists
The Scroll Machine, Scroll claims, is a collection of machine learning algorithms which analyse social media posts to identify interesting and important stories. The idea is to automate the lists or news it picks real time. Scroll claims that the Machine processes more than 100,000 social media posts every day. It plans to launch other lists, namely the Machine’s feeds, which its editors have been reviewing, soon. These will include real time debates happening on Twitter.
An example of the Scroll Machine’s list is a “sickular vs bhakts” list. The algorithm is fed a list from the Scroll editorial team, based on which it discovers the most influential people on the ‘indian politics’ topic. These influencers are used for better information discovery, Scroll says. It then processes politics data over a period of time (unknown) from these people to discover the most ‘actively biased’ Twitter users from it, dividing them into two groups based on analysis, the content they share and their network. It is not clear who, the machine or Scroll’s editorial team, filters spam and non relevant data.
Scroll says that it uses interactions between the clustered groups and filtered content from influencers to put content into the ‘sickular’ and ‘bhakt’ buckets. Since data is classified on a per tweet basis, Scroll says that it is possible to have a tweet go into both buckets but will add stability over time. The tweets are then ranked to generate a feed using certain parameters which include account authority, content score, polarity score and time sensitivity to display the top results on the website.