Learning Apparatus Parameter Segmentation for Predictive Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing mechanisms for predicting user actions require a large number of parameters to be learned, leading to significant calculation resources being necessary for learning predictive models, which can be cumbersome and resource-intensive, especially when dealing with a large number of users and actions.
Innovation Solution
A learning apparatus that learns two parameter groups: one related to the membership rate of users in clusters and another related to the action tendency of clusters, instead of directly learning the probability of each user performing each action, thereby reducing the number of parameters and calculation resources needed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the probability that each user performs each action is learned directly, then the prediction accuracy for individual users is improved, but the number of parameters to be learned increases significantly (number of users × number of actions)
Solution Approach 1:
The patent segments the direct user-action parameter learning into two separate parameter groups: (1) user parameters that represent individual user characteristics, and (2) action parameters that represent action tendencies. This segmentation reduces the total number of parameters from the product of users and actions to the sum of user parameters and action parameters, making the model more scalable.
Solution Approach 2:
The patent introduces a new dimension by adding a parameter group related to actions as a separate entity. Instead of directly modeling the user-action matrix, the invention decomposes it into user-specific parameters and action-specific parameters, effectively changing the dimensional structure from a dense user-action matrix to a sparse parameter decomposition that reduces computational complexity.
2Adaptability or versatility
If the number of users and actions increases, then the coverage and applicability of the predictive model is improved, but the calculation resources required for learning become enormous
Solution Approach 1:
The patent segments the learning task into two independent parameter learning processes: learning user parameters from user behavior data and learning action parameters from action context data. This segmentation allows the model to scale to large numbers of users and actions without requiring proportional increases in computational resources, as each parameter group can be learned independently or with reduced complexity.
Solution Approach 2:
The patent changes the parameter representation from a dense user-action probability matrix to a sparse decomposition into user parameters and action parameters. This parameter transformation reduces the computational burden from O(n×m) to O(n+m) where n is the number of users and m is the number of actions, enabling the model to handle large-scale data efficiently.
Data Source
AI summary
A learning apparatus 10 of one embodiment includes an acquisition unit 11 for acquiring action history data indicating action history for each of a plurality of users, and a learning unit 13 for learning parameter groups PC, C included in a predictive model M for predicting an action of each of the plurality of users by using the action history data as training data. The parameter group PC is a parameter group related to a membership rate of each user for each of a plurality of clusters. The parameter group C is a parameter group related to an action tendency of each cluster for each of a plurality of actions.