Transformer-Based Search Query Recommendations from User Interactions

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional online systems fail to account for various interactions by customers when recommending search queries, leading to increased user interaction and time spent on search interfaces due to the limitations of relying solely on previous search queries.

Innovation Solution

An online system generates a user sequence based on prior search queries and interactions, applies a query relevance model using a transformer-based architecture to predict relevant search queries, and displays these queries proactively to reduce user input.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional online systems display recommended search queries based solely on previously received search queries, then the system can provide some search recommendations, but the recommendations do not account for other kinds of interactions by the customer with the online system or relationships between different previously received search queries, limiting the quality and relevance of recommendations

Engineering Contradiction:

Improvescope of interactions consideredVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent transitions from analyzing only search query text to incorporating multiple dimensions of user interaction data including clicks, views, purchases, and item details. This multi-dimensional approach allows the system to capture relationships between different types of interactions and generate more comprehensive search recommendations that reflect actual user behavior patterns across various interaction modalities

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent introduces a transformer-based language model as an intermediary component that processes and integrates diverse interaction data types. This intermediary model encodes user interactions, search queries, and item information into unified representations, enabling the system to capture complex relationships between different interaction kinds without requiring complex rule-based processing logic

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If the system displays more comprehensive recommended search queries based on multiple interaction types, then the relevance and quality of recommendations improves, but the computational resources and processing time required increases

Engineering Contradiction:

Improvequery recommendation accuracyVSAvoidcomputational energy consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent pre-trains the transformer-based language model on large corpora of user interaction data before deployment. This preliminary training allows the model to learn general patterns of user behavior and item relationships in advance, so that during actual operation, the system can generate recommendations more efficiently by applying the pre-learned knowledge to specific user interaction sequences rather than computing from scratch

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent employs attention mechanisms in the transformer model that dynamically adjust parameter weights based on the input interaction sequence. The attention mechanism selectively focuses computational resources on the most relevant interactions and items for each recommendation task, reducing unnecessary computations and energy consumption while maintaining high recommendation accuracy by concentrating processing power where it matters most

Inventive Principle:

Principle #35Parameter changes

3Ease of operation

If the system requires users to manually enter search terms, then users can perform precise searches, but the amount of interaction with the search interface increases and time spent on the search interface increases

Engineering Contradiction:

Improvesearch interface ease of useVSAvoidtime spent on search interface

Core Design Contradiction:

Ease of operationVSLoss of time

Solution Approach 1:

The system proactively generates and displays recommended search queries based on user interaction history before the user needs to perform a search. This preliminary provision of relevant search options allows users to quickly select from pre-computed recommendations rather than manually constructing search queries, reducing both interaction steps and time spent on the search interface while maintaining search precision through context-aware recommendations

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250321971A1Search Engine for Recommending Search Queries Based on User Interactions Using a Transformer-Based Language Model

Publication Date: 2025.10.16 MAPLEBEAR INC

AI summary

An online system provides a search interface for a user to identify items. The search interface may present suggested search queries to the user, allowing the user to select a suggested search query rather than manually entering search terms to form a search query. To identify search queries most likely to be selected by the user, the online system gets a set of candidate search queries and generates a relevance score for each candidate search query by applying a trained query relevance model to each candidate search query. The scored candidate search queries are selected and ranked using the relevance scores, and the selected candidate search queries are displayed using the ranking in the search interface. The query relevance model is a transformer-based small language model receiving a user sequence of prior search queries and items with which the user interacted and the candidate search terms as input.