Beauty Recommendation LLM Classification for Hallucination Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing large language models (LLMs) struggle with hallucinations and inaccuracies in providing personalized beauty recommendations, lacking the ability to effectively classify user inputs and provide context-specific responses.
Innovation Solution
A system that utilizes a large language model (LLM) to classify user inputs, append contextual information, and reduce hallucinations by using prompt engineering and multiple classifiers to ensure accurate and relevant responses, integrating skin analysis and virtual try-on capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing large language models are used to provide beauty recommendations, then the system can generate responses quickly, but the model produces hallucinations and inaccuracies
Solution Approach 1:
The patent introduces an intermediary classification system between the user input and the LLM. The classifier categorizes user inputs into different beauty domains (skincare, makeup, haircare) and passes these structured classifications to the LLM as contextual guidance. This intermediary layer reduces hallucinations by providing the LLM with predefined categories and constraints, while maintaining fast response generation through automated classification.
Solution Approach 2:
The system segments the beauty recommendation task into distinct classification categories (skincare, makeup, haircare, etc.). Each category is handled with domain-specific prompts and contextual information tailored to that segment. This segmentation allows the LLM to focus on specific beauty domains with appropriate expertise, reducing cross-domain hallucinations while maintaining overall system productivity.
2Measurement precision
If the LLM is prompted to provide classifications for user input, then the system can improve response relevance, but the number of requests to the LLM increases
Solution Approach 1:
The system applies partial classification by only categorizing user inputs that require beauty domain expertise. Simple queries or those clearly outside beauty topics are handled without full classification workflows. This selective approach maintains high classification accuracy for relevant inputs while reducing the total number of LLM requests by avoiding unnecessary classification steps for irrelevant queries.
3Reliability
If contextual information is appended to user inputs, then the LLM can provide more accurate responses, but the processing time and complexity increase
Solution Approach 1:
The system performs preliminary actions by pre-processing user inputs through classification and appending relevant contextual information (such as user profile data, product database entries, or category-specific guidelines) before the LLM generates responses. This preliminary structuring of context reduces the LLM's cognitive load during response generation, improving accuracy while managing processing time through efficient pre-computation of contextual data.
Data Source
AI summary
A computer system transmits user input and contextual information to a large language model (LLM) and requests the LLM to confirm the user input relates to one or more beauty topics. Based on the confirmation, the system requests the LLM to provide a response to be presented to a user via a user interface (UI), which relates to the beauty topic(s) and is based on the user input and contextual information. The confirmation may include requesting the LLM to provide one or more classifications of the user input. The UI may include elements such as a skin analysis request element, a product information element, or a content selection element. For example, the skin analysis request element may be activated to obtain a digital model of a face of the user, and a product or care routine recommendation can be generated based on the digital model.