Sparse Field Weights for Faster Content Selection Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for determining click probabilities and selecting content for presentation consume excessive computational resources and time due to their non-sparse nature, leading to inefficiencies in content selection processes.
Innovation Solution
Implement pruning operations to generate machine learning models with sparse field weights, reducing computational and memory requirements while maintaining accuracy in determining positive signal probabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are used for determining click probabilities and selecting content, then content selection accuracy is improved, but computational requirements and memory usage increase
Solution Approach 1:
The patent segments the machine learning model into multiple components: feature extraction module, field weight determination module, and click probability determination module. This segmentation allows each component to be optimized independently, reducing overall computational complexity while maintaining accuracy.
Solution Approach 2:
The patent extracts and pre-computes field weights from historical auction data separately from the main inference process. By taking out the weight determination step and storing it as pre-computed parameters, the real-time content selection process only needs to perform lightweight calculations using these pre-determined weights.
2Measurement precision
If machine learning models are used for determining click probabilities and selecting content, then content selection accuracy is improved, but memory and storage requirements increase
Solution Approach 1:
The patent extracts field weights as separate, compact parameters from the full machine learning model. These extracted weights are stored in a lightweight format that requires minimal memory, separating the essential decision-making parameters from the full model structure.
Solution Approach 2:
Instead of storing and processing full complex models in memory during inference, the patent inverts the approach by pre-computing and storing only the essential field weights, then using these compact representations for real-time predictions, thereby reducing memory requirements.
3Measurement precision
If extensive computations are performed for content selection, then selection accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary computation of field weights using historical auction data before the actual content selection process. By pre-computing these weights offline, the system eliminates the need for time-consuming computations during real-time content selection, significantly reducing processing time while maintaining accuracy.
Solution Approach 2:
The patent segments the computation into offline weight determination and online probability calculation phases. This segmentation allows computationally intensive tasks to be performed in advance, while the real-time phase only requires lightweight operations, thereby reducing processing time without sacrificing accuracy.
Data Source
AI summary
One or more computing devices, systems, and/or methods are provided. A machine learning model may be trained using a plurality of sets of information. One or more pruning operations may be performed in association with the training to generate a machine learning model with a sparse set of field weights associated with feature fields associated with features of the plurality of sets of auction information. A request for content associated with a client device may be received. A set of features associated with the request for content may be determined. Positive signal probabilities associated with a plurality of content items may be determined using the machine learning model based upon field weights, of the machine learning model, associated with the set of features. A content item may be selected from the plurality of content items for presentation via the client device based upon the positive signal probabilities.