LinkedIn's LLM fix for keyword search: content value measured by relevance and dwell time

_jaydeepkarale · x · 2026-09-16

This thread discusses LinkedIn's reflection on content retrieval: traditional keyword matching either returns zero results (no post contains every query keyword) or returns posts that include all keywords yet fail to answer the question, due to lack of conceptual understanding of the query.

For content quality, LinkedIn relies on two metrics only: on-topic rate (how relevant content is to the user's need) and long-dwell (how long people spend on a post). Likes and comments are excluded since they can easily be farmed via engagement farming.

Original post →

More from Research

Research channel →