User Activity Assessment Using Exponential Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining user activity levels on online services are inaccurate, particularly for infrequent users, as they rely on frequent interactions and may miss device usage or provide misleading information.
Innovation Solution
Collecting device information, access information, and interaction data to calculate historical daily-count values and time-spent values using exponentially weighted sums, and reweighting these values based on time-spent information to determine usage probabilities, which accurately reflect user engagement regardless of frequency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing methods rely on frequent interactions to determine user activity levels, then the measurement appears straightforward, but the accuracy deteriorates for infrequent users
Solution Approach 1:
The patent changes the parameters used for measurement from simple interaction frequency counts to exponentially weighted sums of daily count values and time-spent values. This transformation allows the system to accurately measure user activity levels for both frequent and infrequent users by giving appropriate weight to different interaction patterns over time, rather than relying solely on raw frequency data.
Solution Approach 2:
The patent implements feedback mechanisms by continuously updating historical daily-count values and historical time-spent values using exponentially weighted sums. The system incorporates new interaction data while maintaining awareness of historical patterns, allowing the measurement to adapt and improve its accuracy over time for each user's specific usage behavior.
2Measurement precision
If the system collects detailed device information and interaction data, then the accuracy of user activity assessment improves, but the data collection and processing complexity increases
Solution Approach 1:
The patent extracts and processes only the essential features from the collected data - specifically daily count values and time-spent values - rather than processing all raw interaction data. By extracting these key metrics and computing their exponentially weighted historical sums, the system achieves accurate user engagement measurement while reducing processing complexity compared to analyzing every individual interaction detail.
Solution Approach 2:
The patent performs preliminary processing by pre-computing historical daily-count values and historical time-spent values using exponentially weighted sums before final usage probability calculations. This preliminary action aggregates and prepares the data in advance, reducing the complexity of real-time processing while maintaining measurement accuracy.
3Reliability
If the system uses simple interaction counts, then the processing is fast and simple, but the reliability deteriorates for infrequent users
Solution Approach 1:
The patent transforms simple interaction counts into exponentially weighted historical daily-count values and historical time-spent values. This parameter transformation maintains computational efficiency while significantly improving reliability for infrequent users, as the weighted historical approach can detect and accurately represent sporadic usage patterns that simple count methods miss.
Solution Approach 2:
The patent introduces dynamic weighting through exponentially weighted sums, where recent interactions are weighted more heavily than historical ones. This dynamic approach allows the system to adapt to changing user behaviors and maintain high reliability across different usage scenarios - from frequent to infrequent users - without requiring a complete redesign of the processing architecture.
Data Source
AI summary
In one embodiment, a method includes, for each of multiple interactions by a user with an online service, collecting: device information describing a device used by the user for the interaction; access information describing how the user accessed the online service for the interaction; and a count value of the interaction. The method further includes determining based on the collected information a historical count value for each of one or more particular combinations of device and access information.