Personalizes query suggestions by computing similarity between candidate suggestions and the user's past queries through matching entities, surfacing suggestions tuned to the individual's recurring topical interests.
Patent Overview
- Inventor
- Nitin Gupta
- Assignee
- Google LLC
- Filed
- 2017-06-30
- Granted
- 2019-12-03
- Application Number
- US 15/640,316
The Challenge
Suggestions Default To Population Averages, Not Personal Patterns
Standard query suggestions are tuned to the population. A user with a distinct topical profile (specialist, hobbyist, professional in a niche) sees suggestions skewed toward what most users want, not what they want. Personalizing suggestions requires reading the user's history and tuning the surface to their actual interests, not the median user's.
- Average Users Don't Exist — Real users have specific interests. Population-average suggestions serve nobody perfectly.
- Past Queries Carry The Profile — What a user has searched for in the past reveals what they care about. The personalization should read this history.
- Entity-Level Match Is The Right Signal — Surface-form match between past queries and candidate suggestions is too literal. Entity-level match captures the underlying topical alignment.
- Similarity Measure Must Reward Topical Fit — Candidates whose entities overlap with the user's past-query entities should rise. The similarity computation drives the personalization.
Innovation
Entity-Match Similarity For Personalized Suggestions
When candidate query suggestions are produced for a current query, each candidate is scored by a similarity measure against the user's past queries. The measure is based on matching entities related to the candidate and entities related to the past queries. Candidates with high similarity scores surface higher than candidates that don't match the user's topical profile.
- Identify Candidate Suggestions — Standard suggestion pipeline produces candidates responsive to the current query.
- Extract Entities From Candidates — For each candidate, extract the entities it references.
- Retrieve User's Past Queries — Pull the user's recent past queries from session and account history.
- Extract Entities From Past Queries — For each past query, extract its entities. The aggregate forms the user's entity profile.
- Compute Similarity Per Candidate — Score each candidate's entity overlap with the user's entity profile. High overlap indicates topical fit.
- Rerank Candidates By Similarity — Promote high-similarity candidates in the final suggestion list. Low-similarity candidates drop in rank.
- Surface Personalized Suggestions — Display the reranked list. The user sees suggestions tuned to their past interests rather than population-average defaults.
Entity Match Drives Personal Fit
The patent uses entity-level matching as the personalization signal rather than literal query-string matching. Entities persist across phrasing variations and capture the user's underlying topical profile cleanly.
Past Entities Predict Current Fit
When candidate suggestions share entities with the user's past queries, the candidates are likely to fit the user's current intent too.
- User's Past Query Entities — Aggregated entities from the user's history. Represents the topical profile.
- Candidate Entity Match — Per-candidate similarity to the user's profile via entity overlap.
What This Means for SEO
What This Means for SEO
Personalized suggestions shape what every user actually types. Knowing the entity-match mechanism informs how to position content for users with specific topical profiles.
- Entity-Strong Content Surfaces To Aligned Audiences — Pages with strong, recognized entities surface to users whose past queries include those entities. Entity alignment is the personalization wedge.
- Niche Audiences See Niche Suggestions — Users who consistently search niche topics see niche suggestions. If your content serves a niche, the personalization pulls you toward that niche's audience directly.
- Audience-Defined SEO Beats Generic Keyword Targeting — Optimizing for entity-defined topics aligned with your target audience captures personalized-suggestion traffic that pure keyword targeting misses.