Asynchronous Media Embedding Updates via Activity Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face inefficiencies and increased latency due to frequent updates of media item access history embeddings, especially when users access short-form media items, leading to wasted computing resources and less accurate predictions of user interest.
Innovation Solution
Implementing asynchronous updates for media item access history embeddings by determining whether embedding relevance criteria are satisfied, such as differences in engagement data or time periods, to decide when to generate updated embeddings, thereby reducing unnecessary updates and optimizing resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If frequent updates of media item access history embeddings are performed, then prediction accuracy of user interest is improved, but computing resource consumption increases and latency increases
Solution Approach 1:
The system dynamically adjusts the embedding update frequency based on user activity patterns. When users access many media items, embeddings are updated more frequently; when access is minimal, updates are reduced or deferred. This dynamic adaptation resolves the contradiction by optimizing prediction accuracy only when necessary, thereby reducing unnecessary computing resource consumption while maintaining effective user interest modeling.
Solution Approach 2:
The patent changes the parameter of update frequency from a fixed value to a variable determined by activity thresholds. By monitoring the number of media items accessed and comparing against thresholds, the system adjusts whether to perform full embedding updates, partial updates, or no updates. This parameter change enables the system to maintain prediction accuracy when needed while minimizing computing resource usage during low-activity periods.
2Measurement precision
If frequent updates of media item access history embeddings are performed, then prediction accuracy of user interest is improved, but system throughput decreases
Solution Approach 1:
The system dynamically controls update frequency based on user activity levels. During periods of low user activity, embedding updates are reduced or skipped entirely, allowing the system to maintain higher throughput for other operations. When user activity increases and crosses activity thresholds, the system increases update frequency to maintain prediction accuracy. This dynamic approach resolves the contradiction between accuracy and throughput by adapting update intensity to actual user engagement levels.
Solution Approach 2:
Instead of continuous or fixed-periodic updates, the system implements conditional periodic updates triggered by activity thresholds. Updates occur periodically only when user activity metrics indicate they are necessary. This approach maintains prediction accuracy when users are actively engaging with content while reducing update frequency during inactive periods, thereby preserving system throughput for other critical operations.
3Measurement precision
If embeddings are updated after each media item access, then user interest prediction is more accurate, but latency increases
Solution Approach 1:
The system changes the update timing parameter from immediate (after each access) to conditional (based on activity thresholds). By comparing the number of accessed media items against defined thresholds, the system determines whether to perform updates immediately or defer them. This parameter change reduces latency by avoiding unnecessary immediate updates while still maintaining accurate user interest predictions when user activity levels indicate they are warranted.
Solution Approach 2:
The update frequency is made dynamic rather than static. The system adapts update timing based on real-time user activity patterns. When users access media items at a rate that crosses activity thresholds, embeddings are updated promptly to maintain prediction accuracy. When activity is low, updates are deferred, reducing latency. This dynamic timing strategy resolves the contradiction between accuracy and latency by updating only when user behavior patterns indicate it will improve prediction quality.
Data Source
AI summary
Methods and systems for asynchronous updates for media item access history embeddings are provided herein. An embedding that represents a media item access history associated with a client device with respect to a first set of media items previously accessed by the client device is identified. A determination is made of whether one or more embedding relevance criteria are satisfied with respect to the media item access history of the client device. Responsive to a determination that the one or more embedding relevance criteria are satisfied, a media item is selected of a second set of media items not yet accessed by the client device based on the embedding. The client device is provided with access to the selected media item.