POC: basic recommendation capabilities of the EOSC Portal recommendation system

[roksanaer]

Taken from the confluence problem description (https://docs.cyfronet.pl/pages/viewpage.action?spaceKey=FID&title=First+prototype+of+the+recommendation+system): Please update along with the algorithm development so info gathered here reflects the solution most up to date.

POC system should meet the requirements received by the project partners:

1. Get recommendations for new EOSC Resources based on similar users. The authenticated User can receive recommendations for new EOSC Resources based on profiles and common interests of similar users (scientific area, etc) on the EOSC Portal.
2. Get recommendations for new EOSC Resources based on EOSC Resources a user already searched or ordered. The authenticated User can receive recommendations for new EOSC Resources based on previous searches or ordered EOSC Resources on the EOSC Portal. Please note that the main focus of the recommendations come from requirements for authenticated users.

Recommendation algorithm first specification

Recommendation factors:

1. The current user's engagement in the Portal a) User profile (for authenticated users) - Promote search results that are matching information in the user profile
   b) User activity - Promote search results that are similar to the ones the user have already engaged with Activity type (user interest type)
   • The current user clicks
   • The current user orders (in general project_items)

Resource similarity factors

• Engagement by similar users (this may be overlapping with or replacing "Similar user engagement" factor)
• Common "Target users"
• Common "Providers"
• Common "Geographical Availabilities"
• Common "Language Availabilities"
• Same category
• Similar tags
• Resources frequently compared by users
• other...  (Access types, Access modes)
• Common use in  "Marketplace Projects" (among all users)
• Providers indication of "Related resources"
• other... (Access types, Access modes)
  2. General public reception / reputation (see Projekt implementacji algorytmów rekomendacji) a) Users' interest for the resource - Promote search results that are 'valued' by the Portal users (similar users may be rated higher)
• hover-rate,
• click-rate
• order factor
• user engagement (number of ranking, opinions, without measuring quality, questions for providers), b) Users' satisfaction with the resource - Promote search results that are (questionable factor - low credibility)
• resource rating
• opinion content
• service reuse (??) c) Resource maturity
• TRL
• Life Cycle status
• Certifications
• Number/Quality of metadata provided?
  3. Similar users engagement (may be replaced with 1.b.I) a) Promote search results that would be promoted for similar users (using the same set of promotion rules)

Each factor should be applied with an adequate weight (or each component of a factor if we treat the top-level hierarchy as "factors").

[roksanaer]

duplicates #1756 Closing issue