Machine-Learned Networks for Visitation Prediction Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in predicting visitation to a location based on viewing content, due to the complexity of individual decision-making processes and the scarcity of data on personal preferences.
Innovation Solution
A machine-learned network embeds visitation and metadata information, using a first network to generate complete trace data for unobserved portions, a second network to model the causal effect of content on visits, and a third network to recommend content placement based on graph and metadata analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine-learned networks are used to predict visitation patterns, then prediction accuracy is improved, but system complexity increases
Solution Approach 1:
The system segments the prediction task into three distinct machine-learned networks: (1) generating complete trace data from sparse observations, (2) modeling causal effects of content on visits, and (3) recommending content placement. This segmentation allows each network to specialize in a specific aspect of the prediction process, improving overall accuracy while making the complex system more manageable and interpretable
Solution Approach 2:
The patent introduces intermediate representations including device trace data, content exposure data, and graph embeddings as mediators between raw input data and final predictions. These intermediaries transform complex raw data into structured formats that the prediction networks can process effectively, bridging the gap between diverse input sources and prediction outputs
2Loss of information
If complete trace data is generated for unobserved portions, then data completeness is improved, but data processing time increases
Solution Approach 1:
The system performs preliminary actions by pre-processing device trace data into structured formats and pre-computing graph embeddings before the actual prediction process. This preliminary structuring of data reduces the computational burden during prediction, allowing complete trace data generation without excessive processing delays
Solution Approach 2:
The patent uses copying by creating synthetic or imputed trace data entries for unobserved portions based on patterns from observed data. Instead of requiring actual complete observations, the system generates representative copies of trace data that capture the essential patterns, achieving data completeness without requiring proportional increases in data collection time
3Reliability
If causal effects of content on visits are modeled, then prediction reliability is improved, but computational resources increase
Solution Approach 1:
The system applies local quality by focusing computational resources on modeling causal effects only for relevant content-location pairs identified through graph analysis. Rather than computing causal effects for all possible combinations, the system identifies and processes only the locally relevant relationships, improving prediction reliability for key relationships while reducing overall computational resource consumption
Data Source
AI summary
For machine-based prediction of visitation, a machine-learned network embeds the visitation and metadata information. Since the trace data used to show access may be sparse, another machine-learned network completes the route data. Another machine-learned network recommends effectiveness of content based on routes, the graph, metadata, and/or other information. The recommendation is based on training using counterfactual and/or other causal modeling.