Category-Specific Instruction Suggestions Using Token Frequency Normalization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional natural language processing techniques fail to generate effective suggested instructions for online concierge systems, as they often provide generic instructions across different item categories, lacking specificity for individual item types.

Innovation Solution

An online concierge system scores and normalizes candidate instruction representations based on intra-category and inter-category frequency of use of instruction tokens to select and generate item-specific suggested instructions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional NLP techniques are used to generate suggested instructions, then the system can provide instructions for multiple item categories, but the instructions become generic and lack specificity for individual item types

Engineering Contradiction:

Improveinstruction applicability across item categoriesVSAvoidinstruction specificity for item type

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent segments the instruction generation process by creating separate frequency analysis for intra-category and inter-category instruction tokens. This segmentation allows the system to evaluate and select instructions that are specific to each item category rather than generating generic instructions applicable to all categories, thereby resolving the contradiction between versatility and specificity.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by making the instruction generation adaptive to local characteristics of each item category. By calculating intra-category frequency (local to the category) and comparing it with inter-category frequency (global across categories), the system tailors suggested instructions to the specific needs of each item type while maintaining overall system versatility.

Inventive Principle:

Principle #3Local quality

2Measurement precision

If human experts generate suggested instructions for each item, then the instructions are highly specific and accurate, but the process becomes incredibly expensive and time consuming

Engineering Contradiction:

Improveinstruction accuracy and specificityVSAvoidinstruction generation efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent implements self-service by enabling the system to automatically generate suggested instructions using NLP techniques combined with frequency analysis of user-provided instructions. Instead of relying on human experts for each item, the system learns from existing instruction data and autonomously generates category-specific instructions, achieving both high accuracy and scalability.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent uses feedback from actual user instructions to improve instruction generation. By analyzing the frequency and effectiveness of instructions provided by users across different categories, the system continuously refines its suggested instructions, achieving expert-level accuracy through automated learning rather than manual curation.

Inventive Principle:

Principle #23Feedback

3Measurement precision

If the system prioritizes intra-category frequency of instruction tokens, then the suggested instructions become more specific to the item category, but the system may overlook instructions that are universally applicable across categories

Engineering Contradiction:

Improveinstruction category specificityVSAvoidinstruction general applicability

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent changes the evaluation parameters by introducing a dual-frequency analysis approach. Instead of relying on a single frequency metric, the system simultaneously considers intra-category frequency (for specificity) and inter-category frequency (for general applicability), allowing it to balance category-specific instructions with universally applicable ones through parameter optimization.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20260004074A1Generating suggested instructions through natural language processing of instruction examples

Publication Date: 2026.01.01 MAPLEBEAR INC

AI summary

An online concierge system generates suggested instructions for presentation to a user. The online concierge system access instruction examples corresponding to a target item category and generates candidate instruction representations based on instruction messages within each instruction example. The online concierge system generates preliminary scores for the candidate instruction representations that are directly related to an intra-category frequency of use of the instruction tokens of the candidate instruction representation within the target item category. The online system normalizes these preliminary scores for the candidate instruction representations based on the inter-category frequency of use of the instruction tokens in all item categories to generate final scores for the candidate instruction representations. The online concierge system selects a set of instruction representations based on these final scores and generates suggested instructions based on the set of instruction representations.