Lightweight Recommendation Model Using Parallel CPU Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing item recommendation technologies face challenges in efficiently processing large datasets and providing real-time recommendations, especially due to the limitations of conventional collaborative filtering methods which struggle with scalability and computational efficiency.
Innovation Solution
The technology employs a model that converts a binary matrix of user interactions into a design matrix, which is then processed in parallel by central processing units (CPUs) using a greedy coordinate descent algorithm, allowing for efficient training and inference with reduced computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional collaborative filtering methods are used to process large datasets, then recommendation accuracy can be maintained, but computational efficiency and scalability deteriorate
Solution Approach 1:
The patent segments the recommendation system into two distinct phases: an offline training phase where a lightweight model is trained on aggregated user interaction data, and an online inference phase where the trained model quickly generates recommendations for individual users. This segmentation allows the system to process large datasets efficiently during training while maintaining fast response times during service.
Solution Approach 2:
The system performs preliminary training of a lightweight recommendation model during an offline phase before actual recommendations are needed. By pre-training the model on aggregated data from many users, the system prepares computational resources in advance, allowing for fast real-time recommendations without requiring heavy computational resources during the actual recommendation generation.
2Measurement precision
If more computational resources are allocated to training, then model accuracy improves, but training time and resource consumption increase
Solution Approach 1:
The patent employs a lightweight recommendation model that is intentionally designed to be computationally inexpensive and easy to train. Rather than using complex heavy-duty models, the system uses a simplified model structure that can be quickly trained on aggregated data and then deployed for fast inference, sacrificing some model complexity for significantly improved training speed and resource efficiency.
3Productivity
If item popularity is used as a feature in recommendations, then popular items are recommended, but artificially inflated popularity creates bias
Solution Approach 1:
The system incorporates feedback mechanisms that monitor actual user interactions with recommended items and adjust the model accordingly. By continuously learning from user behavior patterns and adjusting recommendations based on actual engagement rather than just initial popularity metrics, the system can correct for artificially inflated popularity and reduce recommendation bias.
Data Source
AI summary
Users interact with items, such as movies, music, and document templates, among others. Item recommendations based on these user interactions are determined and provided to a user. A binary matrix indicating what items users have interacted with is provided. A design matrix is determined from the binary matrix. In this format, the model can be processed in parallel by a computing device. Columns of the design matrix are processed by threads of one or more CPUs of a computing system, in which a least squares analysis is performed over each thread. The output of the processing is a trained model useable for outputting an item recommendation responsive to an input user interaction during inference.