Three-Level Neural Network Model for Discrete Feature Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning models struggle with applying their versatility to large-scale discrete feature data common in industries, leading to inefficiencies in model training and prediction.
Innovation Solution
A three-level neural network model training method that learns feature information representations, interaction representations, and prediction results through under-layer, intermediate, and top-layer neural networks, respectively, allowing for effective extraction and prediction from feature data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional deep learning models are used, then prediction results can be obtained, but the models lack versatility when applied to large-scale discrete feature data common in industry
Solution Approach 1:
The neural network model is divided into three distinct levels: a first-level model for learning feature information representations, a second-level model for learning interaction representations, and a third-level model for learning prediction results. This segmentation allows each level to specialize in specific tasks, improving both versatility across different data types and prediction accuracy through dedicated processing stages.
Solution Approach 2:
The patent introduces an additional dimension of processing by adding intermediate interaction representations between features. Instead of directly mapping features to predictions, the model creates a intermediate layer that captures feature interactions, enabling better handling of complex discrete feature data while maintaining versatility across applications.
2Adaptability or versatility
If existing neural network models are trained on industry data, then predictions can be made, but the model structure must be customized for each business type reducing reusability
Solution Approach 1:
The three-level neural network model is designed with universal components that can handle various types of discrete feature data across different industries. The first-level feature representation, second-level interaction representation, and third-level prediction architecture serve multiple functions across different applications, reducing the need for custom model structures for each business type while maintaining high reusability.
3Measurement precision
If feature information and interactions are extracted separately, then prediction accuracy improves, but the computational process becomes more complex
Solution Approach 1:
The computational process is segmented into three sequential stages: extracting feature information representations in the first level, extracting interaction representations in the second level, and generating prediction results in the third level. This segmentation organizes the computational complexity into manageable stages, improving prediction accuracy through systematic extraction while keeping the computational process structured and manageable.
Data Source
AI summary
A training method and system of a neural network model including a three-level model, and a prediction method and system are provided. The training method includes: acquiring a training data record; generating features of a training sample based on attribute information of the training data record, and using a label of the training data record as a label of the training sample; training the neural network model using a set of the training samples, learning an interaction representation between corresponding input items respectively by a plurality of intermediate models comprised in a second-level model of the neural network model, learning a prediction result at least based on the interaction representations output by the second-level model by a third-level model of the neural network model, and adjusting the neural network model at least based on a difference between the prediction result and the label.