Classification System Using LSTM and CNN Semantic Signatures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Item classification in large categories is challenging due to data size, category skewness, and noisy metadata, particularly in e-commerce applications where precise classification is essential but hindered by the complexity of handling structured and unstructured data.
Innovation Solution
A system utilizing Deep Learning methodologies, specifically combining Long Short Term Memory (LSTM) and Convolutional Neural Network (CNN) layers to analyze unstructured and structured data separately, generating semantic signatures that indicate the likelihood of an item belonging to specific categories, and then using bidirectional LSTM and CNN layers to integrate these signatures for accurate classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional classification methods are used to handle large categories, then the system is simpler to implement, but classification accuracy deteriorates due to data size, category skewness, and noisy metadata
Solution Approach 1:
The patent segments the classification task into multiple stages: initial category prediction using a lightweight model, followed by refined classification using a more complex model only for relevant categories. This segmentation reduces the overall computational burden while maintaining high accuracy for the most important categories, directly addressing the contradiction between accuracy and complexity.
Solution Approach 2:
The patent performs preliminary category prediction using a simple, fast model before applying the more complex classification model. This preliminary action filters out irrelevant categories early, preventing the complex model from processing unnecessary data, thus improving accuracy where needed while controlling overall system complexity.
2Measurement precision
If complex Deep Learning models (LSTM and CNN layers) are used to analyze structured and unstructured data, then classification accuracy improves by 22% in F1 Score, but computational resources and processing time increase
Solution Approach 1:
The patent segments the data processing into two streams: unstructured text data processed by LSTM layers and structured attribute data processed by CNN layers. These segmented processing streams operate independently and can be optimized separately, allowing the system to achieve high accuracy through specialized processing while managing computational resources more efficiently than a monolithic model.
Solution Approach 2:
The patent applies complex Deep Learning processing selectively rather than uniformly to all data. By using ensemble methods and focal loss, the model focuses computational effort on difficult-to-classify samples and critical features, achieving high accuracy without wasting resources on easily classifiable instances, thus addressing the resource consumption issue.
3Measurement precision
If focal loss with alpha modulation is applied to handle category skewness, then classification balance across imbalanced categories improves, but the complexity of the loss function and training process increases
Solution Approach 1:
The patent dynamically adjusts the alpha parameter in the focal loss function based on category difficulty and data distribution. This parameter modulation allows the loss function to adapt to different levels of category imbalance without requiring fundamentally different training approaches, improving classification balance while keeping the training process manageable through a single adjustable parameter rather than complex multi-parameter optimization.
Data Source
AI summary
Systems and methods including one or more processing modules and one or more non-transitory storage modules storing computing instructions configured to run on the one or more processing modules and perform acts of receiving attribute data comprising a set of unstructured attribute data and a set of structured attribute data, analyzing the set of unstructured attribute data by processing through a first set of one or more Long Short Term Memory (LSTM) layers, to obtain an unstructured semantic signature, analyzing the set of the structured attribute data by processing through a first set of one or more Convolutional Neural Network (CNN) layers, to obtain a structured semantic signature, analyzing the unstructured semantic signature and the structured semantic signature, and classifying the item in one or more item categories. Other embodiments are disclosed herein.