MICO’s recommendation is about providing a framework that uses both collaborative filtering and content-based approaches for recommendations within the platform.

The relevant use cases can be separated into two domains:

  1. Recommendation use cases which require collaborative filtering only, e.g., evaluating user likes
    or site accesses.
  2. Cross-media recommendation use cases, for which the problem may be defined as follows: The
    task of finding a suitable selection of media items, where there are at least two different types of
    input media sharing the same context.

For the collaborative filtering cases, MICO uses prediction.io as a highly customizable recommendation framework. Within the MICO platform,Inside MICO it is coupled within an optional Docker module. Example configurations are available for Site statistics and image likes. Inside the MICO platform, a REST api handles the communication with prediction.io for getting recommendations. See the section on “Getting recommendations” on how to install the WP5 modules.

For the cross-media recommendation, the project focuses on providing infrastructure for implementing a cross media workflow as described by Köllmer et.al in A Workflow for Cross Media Recommendations based on Linked Data Analysis. The main application being the “Editor Support Showcase”. The second cross media recommendation application is filtering of Zooniverse subjects in debated items, to recommend them for expert review. 

Integration of recommendations within the MICO platform.

The overall infrastructure is depicted in the figure. Note, that the recommendation gets content metadata out of the Marmotta triple store via Anno4j, i.e. uses metadata provided by MICO extractors. For further information, see the recommendation repository and the description in “Getting recommendations“.