9 Personalized Search
This chapter covers
- The personalization spectrum between search and recommendations
- Leveraging a user’s signals to generate collaborative personalization from latent features
- Leveraging embedding vectors to generate personalization profiles
- Mixing user signals and content-based attributes to generate multi-modal personalization
- Clustering products by embeddings to create personalization guardrails
- Avoiding the pitfalls of personalized search
The better your search engine understands your users, the more likely it will be able to successfully interpret their queries. In chapter 1, we introduced the three key contexts needed to properly interpret query intent: content understanding, domain understanding, and user understanding. In this chapter, we dive into the "user understanding" context.
While we’ve focused on learning domain-specific context from documents (chapter 5: semantic knowledge graphs) and on the most popular results according to many different users (chapter 8: signals boosting), it’s not always reasonable to assume the idea of the "best" result is shared across all users. Whereas signals boosting models find the most popular answers across all users, personalized search instead attempts to learn about each specific user’s interests and to return search results that cater to those interests.