Feed Ranking Model for Social Networking Services
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Social networking services face challenges in determining which items to display in user feeds and their optimal order, as existing algorithms fail to adequately capture the value of downstream effects, such as viral actions, which impact both users and the platform.
Innovation Solution
A machine learned model is employed to predict the probability of viral actions on feed items, incorporating their potential impact on user interaction and the platform's value, to select and rank items for display.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional algorithms are used to select and rank feed items based on immediate user interaction likelihood, then implementation simplicity is maintained, but the ability to capture downstream effects and viral actions is insufficient
Solution Approach 1:
The ranking algorithm is divided into multiple independent components: a base ranking model that handles immediate user interaction likelihood, and separate downstream effect models that calculate viral potential and long-term value. Each component operates independently and their results are combined, allowing complex downstream effects to be captured without making the entire system不可管理
Solution Approach 2:
A downstream effect model is introduced as an intermediary layer between the base ranking model and the final feed generation. This intermediary calculates viral actions and downstream user interactions separately, then integrates these effects into the overall ranking score, enabling precise measurement of downstream effects without directly complicating the core ranking logic
2Productivity
If feed optimization focuses only on immediate user interaction, then computational resources are conserved, but long-term platform value and viral potential are not maximized
Solution Approach 1:
The system performs preliminary calculations of downstream effects and viral potential during the ranking phase, before feed items are actually presented to users. By pre-computing these metrics using the downstream effect model and incorporating them into the ranking scores, the system maximizes long-term platform value without requiring additional computational resources during feed delivery
Solution Approach 2:
The ranking system incorporates additional parameters beyond immediate interaction likelihood, specifically downstream effect scores and viral action probabilities. These new parameters are weighted and integrated into the overall ranking formula, allowing the system to optimize for long-term platform value while maintaining efficient computation through parameter-based modeling rather than complex simulations
Data Source
AI summary
In an example, a plurality of potential feed objects are obtained. An identification of a user performing a navigation command in a user interface is also obtained, the navigation command causing a feed to be displayed or updated. The identification of the user and the plurality of potential feed objects are fed to a machine learned feed object ranking model, the feed object ranking model having been trained via a machine learning algorithm to calculate a score for each of the potential feed objects, the score being based on a combination of a likelihood that the user will perform an interaction, via the user interface, on the potential feed object, likelihoods that the user's interaction will cause one or more downstream events by other users, and a value of the one or more downstream events to a social networking service. The plurality of feed objects are ranked by their scores.