Kopp: Patent Search System Having Task-Based Machine-Learned Models
- vor 7 Tagen
- 1 Min. Lesezeit
Key Takeaways:
Kopp analyzed search patents related to search and AI search, adding a new patent on Search System Having Task-Based Machine-Learned Models
The patent is a about a search system that
breaks broad user queries into
ranked subtasks using LLMs + specialized multi-task ML models — then
surfaces, ranks and dynamically re-ranks content
based on aggregated user interaction scores across the full task session.
He outlined the process:
Query Classification — is this a goal-oriented task or a simple query?
Task Decomposition — LLM + multi-task model splits the goal into ordered subtasks
Per-Subtask Search — independent content retrieval for each subtask
Interaction Scoring — content ranked by aggregated user engagement signals
Dynamic Re-ranking — user behavior updates scores in real time
Session Context Persistence — task intent carried across multiple queries
Advertiser Targeting — bid on tasks/subtasks, not just keywords
He proposes the following measures to increase visibility in AI Search:
Keywords are dead as the primary unit — tasks and subtasks are the new retrieval target
Build content clusters mapped to full task journeys, not single queries
Engagement signals compound — early clicks on subtask content drive long-term ranking advantage
Ambiguous pages risk wrong subtask mapping — specificity wins
Optimize for task sessions, not isolated queries
Further tactics to increase visibility in AI Search right here
Sources:
