Combines parameters across queries in the same session that share a line of inquiry, so the system can answer the composite intent rather than treating each query as a fresh start, especially useful for multi-step research and exploration.
Patent Overview
- Filed
- 2020-03-12
- Granted
- 2023-09-19
- Application Number
- US 16/816,938
The Challenge
The Challenge
Users often ask several related queries in a row, each refining an underlying line of inquiry. Treating each query independently throws away the accumulated context. The system needs to read the session as one investigation and serve the composite intent.
- Independent Treatment Loses Context — When the user issues three related queries in a session, the third may need parameters from the first two to make sense. Independent treatment forces the user to re-include every constraint.
- Lines Of Inquiry Are Common — Research, comparison, planning, and learning sessions involve multi-query investigations. The pattern is too common to ignore.
- Parameter Identification Is Subtle — Each query carries parameters (location, time, entity, comparison target). Identifying which parameters should persist across queries requires careful parsing.
- User May Switch Topics — Not every consecutive query continues the prior line. The system must detect topic shifts and reset parameters when the inquiry changes.
- Combination Must Stay Predictable — Users should be able to predict which parameters carry over. Surprising combinations feel like the system has gone rogue.
Innovation
How The System Works
The system identifies session-shared queries via topical-coherence detection, extracts parameters from each query, accumulates parameters across the line of inquiry, applies the accumulated parameter set to each new query, and resets when topic-shift is detected.
- Track Session Query History — Sessions record each query with timestamps and parsed parameters. The history forms the input to line-of-inquiry detection.
- Detect Line Of Inquiry — A topical-coherence model identifies queries that share a line of inquiry. Coherence uses entity overlap, topic similarity, and follow-up patterns.
- Extract Parameters Per Query — Each query is parsed for parameters: entities, locations, times, comparisons, filters. The parameter set becomes the basis for combination.
- Accumulate Across Line — Parameters from coherent queries accumulate into a session parameter set. Later queries inherit earlier parameters unless explicitly overridden.
- Apply Composite Parameters — The current query plus accumulated parameters compose a richer effective query. Retrieval and ranking work on the composite.
- Detect And Handle Topic Shifts — When a query no longer coheres with prior ones, the accumulated parameters reset. The system starts fresh for the new inquiry.
- Surface Composite Intent To User — The SERP can subtly indicate which parameters are being applied (filter chips, query-summary block) so users can verify or edit.
Session-Aware Parameter Accumulation
The patent's load-bearing idea is that a session of related queries is one composite intent. Tracking and accumulating parameters across the line of inquiry lets retrieval serve the composite directly.
The Session Is The Unit Of Intent
Single-query treatment misses session-level intent. Aggregating across a coherent line of inquiry makes retrieval responsive to the user's evolving question rather than each query in isolation.
- Coherence Detection — Topical-coherence model identifies which queries share a line of inquiry. Coherence triggers parameter accumulation.
- Parameter Extraction — Each query's parameters are extracted into a structured set. Accumulation operates on the parameter structure, not raw text.
- Reset On Topic Shift — When coherence breaks, accumulated parameters reset. Topic shift detection prevents stale parameters from polluting unrelated queries.
Technical Foundation
Technical Foundation
The patent specifies the session history, the coherence detector, the parameter extractor, the accumulator, and the integration with retrieval and ranking.
- Session History Store — Per-user session history records query text, timestamps, and parsed parameters. The store is pseudonymized and retention-bounded.
- Coherence Detector — Topical similarity model decides whether a new query continues a prior line of inquiry. Inputs include entity overlap, embedding distance, and time-since-prior-query.
- Parameter Extractor — Parses queries into typed parameters: entities, locations, times, filters, comparisons. Output is a structured parameter set.
- Parameter Accumulator — Maintains the running parameter set for the current line of inquiry. Updates on each coherent query, resets on topic shift.
- Composite Query Builder — Combines the current query with accumulated parameters to produce a richer effective query for retrieval and ranking.
- User-Visible Indicator — Filter chips or query-summary blocks show which parameters are active. Users can edit or remove parameters they did not intend to accumulate.
The Process
The Process
The pipeline runs in the query path. Latency is minimal because session state is already cached per user; the additions are coherence detection and parameter composition.
- Receive Query — Query arrives with session ID. The session history is fetched from cache.
- Parse Parameters — Extract parameters from the incoming query into a structured set.
- Detect Coherence — The coherence detector compares the new query to recent session queries. Output is coherent or not.
- Accumulate Or Reset — If coherent, accumulate the new parameters with prior. If not, reset the accumulator.
- Build Composite Query — The composite query is the current query plus accumulated parameters. This goes to retrieval.
- Render Results With Indicators — Results return with active-parameter indicators in the UI. Users can verify or edit.
- Update Session State — The session history updates with the current query and updated parameter accumulator state.
Quality Control
Quality Control
Wrong accumulation surprises users. The patent specifies safeguards.
- Coherence Threshold — Below-threshold coherence triggers reset. Conservative threshold prevents accidental accumulation across topics.
- User Override — Users can explicitly clear accumulated parameters. The override is a first-class UI element, not a hidden setting.
- Visible Accumulation Indicators — The user sees which parameters are active. Transparency lets users verify and trust the system.
- Time-Window Bounds — Old session queries do not accumulate forever. A time window bounds how far back coherence detection looks.
- Privacy Boundaries — Sensitive query types are excluded from accumulation. Health, finance, or other sensitive categories reset on each query.
Real-World Application
Session parameter accumulation underpins Google's follow-up-query handling, Assistant multi-turn dialogues, and the conversational layer of Search Generative Experience.
- Session-aware Retrieval Scope — Retrieval considers the line of inquiry, not just the latest query. The composite intent is the unit of search.
- Topic-reset State Management — Accumulated parameters reset on topic shifts. The user is not trapped in stale state.
- Visible User Interface — Active parameters surface in the UI so users can verify, edit, or clear them.
Why Follow-Up Queries Reward Pillar Pages
Pages covering a topic broadly handle follow-up queries gracefully because their content already addresses many sub-aspects. Narrow pages that satisfy only one query in the line lose subsequent slots to broader pages that satisfy several.
Why Conversational Search Inherits This Logic
Each turn in a conversation is a query in a line of inquiry. The accumulation primitives in this patent are the substrate for multi-turn conversational behavior in modern Google products.
<\/section>What This Means for SEO
What This Means for SEO
When the engine fuses signals from a user's recent queries, your page is judged against the implied composite intent rather than the literal current query.
- Session Context Reshapes Intent — A query that looks ambiguous in isolation may be unambiguous given the previous two queries in the session. Pages that satisfy the broader session arc rank for more refined follow-on queries.
- Topical Authority Beats Single-Query Match — When the system reads a multi-query session, breadth of coverage on the topic matters more than precision on the literal current query. Topic-cluster content architectures align with how the system actually reads intent.
- Query Refinement Patterns Reveal Gaps — The follow-up queries users issue after landing on your page tell you what your page failed to answer. Mine refinement queries from your analytics, every one is a new section to add.