Multi-Layer Neural Network for Brand Matching Precision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for brand matching face challenges in resolving duplication and classification issues due to heterogeneous data sources, multiple languages, and ambiguous brand meanings, leading to sparse product coverage and limited representativeness.
Innovation Solution
A multi-layer machine learning model utilizing SBERT layers for sentence analysis, deep averaging networks for aggregating features, and a combination of twin and difference neural networks to analyze similarities and differences, along with a Cross-Lingual Inference Model for Brands, which encodes metadata and textual/image information to generate universal entity vectors for brand matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single-layer machine learning model is used for brand matching, then the model complexity is low, but the matching precision and recall are insufficient due to duplication and classification issues
Solution Approach 1:
The patent divides the brand matching model into multiple layers: an embedding layer for feature extraction, a twin neural network layer for similarity computation, a difference neural network layer for distinction analysis, and a concatenation layer for integrated decision-making. This segmentation allows each layer to specialize in specific aspects of brand matching, improving precision while maintaining manageable complexity through modular design.
Solution Approach 2:
The patent transitions from traditional single-layer matching to a multi-layer neural network architecture that operates in multiple dimensional spaces. The twin and difference networks analyze brand similarities and differences in separate dimensional spaces before combining results, enabling more nuanced matching that resolves duplication and classification issues.
2Adaptability or versatility
If traditional machine learning models are used, then the computational resources required are low, but the ability to handle heterogeneous data sources and multiple languages is limited
Solution Approach 1:
The patent implements a universal embedding layer that processes multiple data types (text, images, metadata) and multiple languages through a single unified neural network architecture. This multi-functional design enables the model to handle heterogeneous data sources and cross-language matching without requiring separate specialized models, improving adaptability while optimizing computational efficiency through shared processing pathways.
3Measurement precision
If simple brand matching algorithms are used, then the processing speed is fast, but the recall rate is low due to sparse product coverage
Solution Approach 1:
The patent employs an embedding layer that pre-processes and encodes brand features from heterogeneous data sources before the main matching operation. This preliminary action creates comprehensive brand representations that improve recall by capturing subtle similarities, while the pre-computed embeddings enable efficient subsequent matching operations that maintain processing speed.
Data Source
AI summary
Devices, systems, and methods are provided for brand matching using multi-layer machine learning. A method may include generating, based on a first embedding vector and a second embedding vector as inputs to a twin neural network, a third embedding vector and a fourth embedding vector; generating, based on the first embedding vector and the second embedding vector as inputs to a difference neural network, a difference vector indicative of a difference between the first embedding vector and the second embedding vector; generating a concatenated vector by concatenating the third embedding vector with the fourth embedding vector and the difference vector; generating, based on the concatenated vector as an input to a feedforward neural network (FFN), a score between zero and one, the score indicative of a relationship between a first entity and a second entity.