Domain-Specific Recommender Models for Multi-Domain Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
General large language models (LLMs) for recommendation systems are computationally expensive due to their large number of parameters, and training them on diverse product domains is hindered by out-of-domain training samples, leading to inefficiencies in training and inference.
Innovation Solution
Implement a domain-specific recommender model framework that uses a base recommender model and smaller, domain-specific models trained on relevant items within a specific domain, utilizing learnable coefficients and thresholds to determine domain relevance, thereby reducing model size and computational requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a general large language model is used for recommendation across multiple domains, then the model can handle diverse product types, but the model size and computational cost increase significantly
Solution Approach 1:
The patent divides the general recommendation model into multiple domain-specific models, each trained on a specific domain's data. This segmentation allows the system to maintain versatility across domains while keeping each individual model smaller and more computationally efficient. The framework includes a domain classifier that routes queries to appropriate domain-specific models,实现ing multi-domain coverage without requiring a single large model.
2Adaptability or versatility
If training data from multiple domains is used to train a general model, then the model can make recommendations across diverse domains, but out-of-domain training samples reduce recommendation accuracy
Solution Approach 1:
The patent extracts and separates training data by domain, creating dedicated training sets for each domain. The domain classifier identifies the relevant domain for each query and routes it to the corresponding domain-specific model, ensuring that only in-domain training samples are used for each recommendation task. This extraction of domain-specific training data eliminates the negative impact of out-of-domain samples on recommendation accuracy.
Solution Approach 2:
The patent applies local quality by training each domain-specific model with high-quality, domain-relevant training data. Instead of using a uniform training approach across all domains, the system tailors the training data and model parameters to each specific domain's characteristics, thereby maximizing recommendation accuracy within each domain while maintaining overall multi-domain capability.
3Measurement precision
If a domain-specific model is trained only on relevant domain data, then recommendation accuracy improves, but the model cannot handle queries outside its domain
Solution Approach 1:
The patent creates a universal recommendation framework that combines multiple domain-specific models with a domain classifier. The classifier acts as a routing mechanism that directs queries to the appropriate domain-specific model, enabling the system to maintain high accuracy within each domain while achieving versatility across multiple domains. This multi-functional architecture allows the system to handle both domain-specific accuracy requirements and multi-domain flexibility.
Data Source
AI summary
Embodiments described herein provide A method of item recommendation at an electronic recommendation system, the method comprising: receiving, via a data interface, a query; generating, via a base neural network based recommender model, a first recommendation of at least a first item based on the query, wherein the first item is associated with a vector of tags; computing a scalar value indicative of a relevance level between the first item and a specific domain based on a multiplication of the vector of tags with a learnable coefficient vector associated with the specific domain; and displaying the first item via a user interface in response to the scalar value surpassing a threshold.