LLM Intention Vectors for Example-Based Product Recommendation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional product recommendation systems struggle to accurately reflect a user's current intention due to their reliance on past preferences, making it difficult to provide relevant recommendations.

Innovation Solution

A method and system that utilizes a Large Language Model (LLM) to derive a user's intention vector from example products and text input, incorporating probability distributions and user history to infer current intentions, enabling the recommendation of both tangible and intangible products.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional recommendation algorithms (collaborative filtering, sequential recommendation, contents based recommendation) are used to recommend products based on user's previous tastes, then the system can provide recommendations, but the recommendation cannot reflect the user's current intention

Engineering Contradiction:

Improveaccuracy of reflecting user's current intentionVSAvoidability to adapt to user's changing preferences

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The system dynamically updates user profiles by incorporating real-time text inputs and example products, allowing the recommendation model to adapt to changing user preferences. The user profile is not static but evolves with each new interaction, enabling the system to capture current intentions rather than relying solely on historical data.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system uses text inputs from users and their interactions with example products as feedback to continuously refine the user profile. This feedback loop allows the system to learn from user behavior patterns and adjust recommendations accordingly, improving accuracy in reflecting current intentions over time.

Inventive Principle:

Principle #23Feedback

2Device complexity

If the system uses only past purchase and click information for recommendations, then the recommendation process is simple, but it cannot capture the user's current needs and intentions

Engineering Contradiction:

Improvecomplexity of recommendation processVSAvoidaccuracy of user intention identification

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The system segments the recommendation process into distinct components: text input processing, example product analysis, user profile updating, and recommendation generation. Each component handles a specific aspect of the task, making the overall complex process more manageable and efficient. The segmentation allows for specialized processing of different data types (text, product information, user history).

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The user profile acts as an intermediary that bridges the gap between raw data (text inputs, purchase history) and final recommendations. It consolidates and interprets various data sources into a unified representation of user preferences, simplifying the recommendation process while improving accuracy through comprehensive data integration.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Measurement precision

If the system incorporates text input and example products to infer current intention, then the recommendation accuracy improves, but the system complexity increases

Engineering Contradiction:

Improveaccuracy of current intention identificationVSAvoidcomplexity of intention derivation process

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The user profile serves multiple functions simultaneously: it stores historical data, processes text inputs, analyzes example products, and generates recommendations. This multi-functionality reduces the need for separate dedicated components for each task, thereby managing system complexity while maintaining high recommendation accuracy through unified data processing.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20250315878A1Method and system for product recommendation based on example products and text input

Publication Date: 2025.10.09 LG MANAGEMENT DEV INST CO LTD

AI summary

Provided are an example product and text input-based product recommendation method and system, which are configured to provide a recommendation product that matches a current intention of a user based on an example product and needs-related text entered by the user.