Archillect has an algorithm that is fed a list of keywords. Instead of posting the search results directly, she wiki-walks between pages and posts, collecting data on various items: image, poster, recent interactions and the visible audience of the post. She maps the social structure of these items by mining as much data as possible from each one of them.
The balance and threshold of keywords and picks are dynamically adjusted as Archillect’s posts on social media gain attention. This not only makes the decision making process nearly human but also gives her a primitive trend perception.
Archillect's curation process is completely automated. She aims to make her posts reach as far as possible considering the potential followers that may share the content with their own followers. As a result she likes attention from accounts with high potential of making her posts survive in the social web and she increases her trust in the accounts that helps her make the correct choices that made earlier posts reach further.
As a result she learns, evolves, communicates and becomes happy in her own ways.
Archillect is limited to the data that's available at the discovered image source. This makes creator/work identification unreliable as the source can be any website, mainly, social media.
Content source is one of the valuable topics for Archillect. There is a great amount of effort going into researching different methods for locating real source of the content that can be used for fully automated attribution, such as the experimental Archillinks, a dedicated account/bot that finds possible sources through web searches. Feel free to have a look at the official blog for more information about our technical approaches and public discussions on automated attribution.
Additionally, on this archive