Collaborative Filtering for Sparse Domain Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing collaborative filtering machine learning models face challenges in sparse, highly dimensional prediction domains, such as health insurance plan recommendations, due to complex relationships between input attributes and the lack of sufficient data, making them inefficient and unreliable for predictive data analysis.
Innovation Solution
The approach generates entity representations and rating features to facilitate predictive data analysis by training collaborative filtering machine learning models on initialization datasets with sparse data, using model-predicted ratings to improve prediction accuracy and reduce computational resource intensity, enabling efficient training and retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If collaborative filtering machine learning models are used in sparse, highly dimensional prediction domains, then prediction-based recommendations can be generated, but the models become inefficient and unreliable due to lack of sufficient data
Solution Approach 1:
The patent applies preliminary action by generating model-predicted ratings for null value fields before final training. The collaborative filtering model first generates predictions for missing data fields in the initialization dataset, creating a model-predicted dataset that is then used for training. This preliminary generation of predicted values allows the model to learn from incomplete data by filling in missing information before the actual training process.
2Productivity
If collaborative filtering models are trained on initialization datasets with null values, then training can proceed with sparse data, but prediction accuracy is reduced
Solution Approach 1:
The patent implements feedback by using model-predicted ratings to retrain the collaborative filtering model. The model first generates predictions for null values, creates a model-predicted dataset, and then uses this predicted data to retrain itself. This feedback loop allows the model to iteratively improve its accuracy by learning from its own predictions, effectively addressing the sparsity issue through multiple training iterations.
3Ease of operation
If existing collaborative filtering approaches are used, then recommendations can be generated, but computational resource intensity is high
Solution Approach 1:
The patent reduces computational resource intensity by performing preliminary actions to pre-process the data. The model generates model-predicted ratings for all null value fields in advance, creating a complete model-predicted dataset before final training. This pre-computation of predicted values allows the subsequent training process to be more efficient, as the model doesn't need to handle missing data during training iterations.
Data Source
AI summary
Various embodiments of the present disclosure provide methods, apparatus, systems, computing devices, computing entities, and/or the like for generating predicted recommendations by using an input entity representation, a reference entity representation, and collaborative filtering machine learning model.