TL;DR:
- The design of both search and recommendations is to find and filter information
- Search is a "recommendation with a null query"
- Search is "I want this", recommendations is "you might like this"
- In recommendations, man "is this search query", from Computing Taste by Nick Seaver
- Search and recommendations are at opposite ends of an extreme spectrum
- Search is more about retrieving information versus filtering for preferences, and in search the user has more agency versus recommendations where the recommender system has more agency
Marcia Bates on Search Systems:
- People like to see where topics are embedded in the contextual landscape because effective searching is not intuitive and there is no explicit index for search strategies
Michael Ekstrand on search vs recsys
- Search is directed information-seeking and we care about
- Item properties
- User properties (preferences and interaction history)
- Query
- Context
In general, these things make up the following formulation:
- Item + User context: Traditional recommendations:
- Item + Query context: Non-personalized search
- Context + user: context-aware recommender
- item, query, query, context: Context-aware personalized search
- Computing Taste by Nick Seaver, on music recommendation
- The Revolt of the Public, about the collapse of top-down information sources
- A Survey of Diversification Techniques in Search and Recommendation
- Recommender Systems Notation
- Wide and Deep Learning for Recommender Systems
- The design of browsing and berrypicking techniques for the online search interface
- James Kirk on lessons from industrial recommender systems
- Marcia Bates interview on search systems
- Michael Ekstrand on the difference between search and recommendations and on recsys notation
- Eugene Yan on system design for search and recommendations
- New Search system based on Stable diffusion
- Expanding AI Dark Forest by Maggie Appleton
