Scalable Deep Learning Architecture for Sparse User Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning models fail to accurately generate user representations due to sparse user-item interaction data, struggling with non-linear relationships and requiring extensive resources, which limits their scalability and frequency of updates in e-commerce item recommendations.
Innovation Solution
A scalable deep learning architecture (SDLA) is employed to process sparse user-item interaction data by splitting it into training and inference batches, training a deep learning model, and generating inferred user embeddings in a latent space, allowing for efficient and frequent updates of user representations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional machine learning models are used to generate user representations, then the system can process user-item interaction data, but the models fail to capture non-linear relationships and produce poor performance on sparse datasets
Solution Approach 1:
The patent replaces conventional machine learning models with a deep learning model that can automatically capture non-linear relationships in user-item interaction data. The deep learning architecture substitutes traditional feature extraction and manual model design with neural networks that learn complex patterns directly from sparse data, improving both accuracy and reliability of user representations.
2Ease of manufacture
If traditional machine learning techniques are employed, then features can be extracted manually, but the process requires extensive manual feature engineering and fails to scale efficiently
Solution Approach 1:
The deep learning model performs automatic feature extraction and representation learning without requiring manual feature engineering. The system serves itself by learning relevant features directly from the input data through the neural network's training process, eliminating the need for extensive manual feature preparation and improving scalability.
3Measurement precision
If deep learning models are trained on massive datasets, then accurate user representations can be generated, but the resource consumption and training time increase significantly
Solution Approach 1:
The patent segments the large-scale user-item interaction dataset into smaller training batches that can be processed efficiently. By dividing the data into manageable chunks and using parallel processing across multiple devices, the system maintains high user representation accuracy while reducing the computational burden and energy consumption associated with training on the entire dataset at once.
4Adaptability or versatility
If user representations are updated frequently to reflect latest user preferences, then recommendation quality improves, but the computational cost and processing time increase
Solution Approach 1:
The patent pre-trains the deep learning model on historical user-item interaction data to learn robust user representation patterns. This preliminary training allows the system to quickly adapt to new user preferences with minimal additional processing when new interaction data arrives, enabling frequent updates without proportionally increasing computational costs.
5Quantity of substance
If sparse user-item interaction data is used directly for training, then the model can be trained with available data, but the sparsity leads to poor generalization and inaccurate predictions
Solution Approach 1:
The deep learning model acts as an intermediary that transforms sparse user-item interaction data into dense user representations in a latent space. The neural network learns to map sparse input data to meaningful dense vectors that capture user preferences, effectively bridging the gap between limited input data and accurate prediction requirements.
Data Source
AI summary
Systems and methods for recommending items based on enhanced user representations are disclosed. A sparse part and a dense part of user-item interaction data are generated. While the dense part is split into a plurality of training data batches, the sparse part is split into a plurality of inference data batches. A deep learning model is trained based on the plurality of training data batches. Inferred user embeddings are generated by applying the trained deep learning model to the plurality of inference data batches in parallel. The inferred user embeddings are non-zero user representations in a same latent space. Based on user session data of a query user and the inferred user embeddings, recommended items are generated and transmitted to a user device for display to the query user.