Reranks autocomplete query completions by checking which completions are likely to co-occur with reference queries in user activity sessions, surfacing completions tuned to session context rather than raw frequency.
Patent Overview
- Inventor
- Nitin Gupta
- Assignee
- Google LLC
- Filed
- 2014-06-12
- Granted
- 2016-03-29
- Application Number
- US 14/303,162
The Challenge
Default Autocomplete Surfaces Population-Average Completions
Autocomplete typically ranks completions by global frequency or recency. The user typing a prefix sees the population's most-common completions, even when the user's current session context suggests a different intent. The system needs to rerank completions based on how likely each is to fit the current activity session, surfacing context-appropriate options rather than population averages.
- Global Frequency Ignores Session Context — A prefix has many possible completions; population frequency picks the same top set for everyone. Session-aware reranking adapts to what the current user is actually doing.
- Session Activity Reveals Intent — Pages visited, queries issued, and content engaged with in the current session signal the user's topic. Completions matching that topic should rise; off-topic completions should fall.
- Co-Occurrence Is The Signal — Completions that co-occur with the user's reference queries in past activity sessions are likely to fit. The system mines this co-occurrence pattern across the global session corpus.
- Reranking Without Replacing — The reranking adjusts the existing completion set rather than replacing it. Users still see the candidate completions; the order reflects session fit.
- Latency Constraint — Autocomplete runs at every keystroke. The reranking must be cheap enough to fit within sub-100ms response budgets.
Innovation
Co-Occurrence Reranking By Session
When the user types a query prefix, the system fetches candidate completions as usual. It also identifies one or more likely queries that are likely to co-occur with a reference query in past user activity sessions. If one of those likely co-occurring queries matches a candidate completion, that candidate is promoted in the rerank. Completions matching the user's session context rise to the top.
- Receive Query Prefix — User types a partial query in the search box. The prefix triggers autocomplete.
- Fetch Candidate Completions — Standard autocomplete generates a set of candidate completions for the prefix.
- Identify Reference Queries In Session — Determine the reference query or queries for the current session. The reference can be recent queries issued in this session or the active session's topical anchor.
- Lookup Likely Co-Occurring Queries — From historical session data, identify queries that frequently co-occur with the reference queries in user activity sessions.
- Match Co-Occurring Queries To Candidates — Check whether any of the likely co-occurring queries match candidate completions. Matches indicate the candidate fits the session context.
- Promote Matching Candidates — Move matched candidates up in the completion ranking. Non-matching candidates retain their default rank or move down to make room.
- Surface Reranked Completions — Display the reranked completion list. The user sees session-appropriate completions at the top.
Session Co-Occurrence As Rerank Signal
The patent uses cross-user session patterns as the reranking signal. When many users follow query X with query Y in the same session, that pattern indicates Y is a likely next-step from X. Applying it to the current user's session produces context-aware completions.
Past Session Patterns Predict Current Intent
User behavior in sessions is patterned. Queries that follow query X cluster around a small set of common successors. The current user is likely to follow the same pattern.
- Reference Query — A recent query in the current session that anchors the rerank. May be the immediately preceding query or the session's dominant topic.
- Likely Co-Occurring Set — Queries that frequently co-occur with the reference query across sessions. Derived from historical session co-occurrence statistics.
- Match-Driven Promotion — When a candidate completion matches a likely co-occurring query, it gets promoted. Strong matches override default frequency ranking.
Autocomplete becomes session-aware by mining what queries actually follow what queries.
<\/section>Technical Foundation
Session Co-Occurrence Data
The reranking depends on a precomputed table of query co-occurrence patterns across sessions.
- Session Definition — A user activity session bounded by inactivity timeouts or explicit session boundaries. Within a session, queries are temporally and topically linked.
- Co-Occurrence Table — Pairwise frequency of queries appearing in the same session across the user population. The table is what the rerank consults.
- Reference Query Identification — The query (or queries) from the current session that anchors the rerank. Often the most recent or most central query in the session.
- Match Function — How the rerank decides whether a candidate completion matches a likely co-occurring query. Exact match, normalized match, and semantic similarity all work.
Key Insight: Population-level session co-occurrence is a cheap, scalable source of personalization. Without needing per-user models, the reranker adapts completions to whatever the current user has been doing in the session. The same mechanism powers 'people who searched for X also searched for Y' surfaces broadly.
<\/section>What This Means for SEO
What This Means for SEO
Session-aware autocomplete reranking shapes which queries users actually issue. Knowing the co-occurrence mechanism informs how to think about session-level content strategy.
- Session Patterns Matter For Discoverability — If your target query frequently co-occurs with another query in user sessions, both queries cross-promote each other through autocomplete. Strong session-level patterns multiply your suggestion exposure.
- Reference Queries In A Session Drive Next Suggestions — What users searched five seconds ago shapes what suggestions they see for their next keystroke. If your content is the next-natural-step from a popular session-anchor query, autocomplete pulls users toward your content.
- Content That Spans Common Session Sequences — When a single page can satisfy multiple steps in a typical session (e.g., topic intro + how-to + comparison + buying guide), it benefits from being relevant across the session's evolving queries.
- Topical Hubs Capture Session Co-Occurrence — Hub pages that link to and from related sub-queries embody the session-co-occurrence pattern at the content level. Strong internal linking around session sequences strengthens this signal.