Event Feed Recommendation via User-to-Space Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large organizations face difficulties in tracking and managing digital items created or updated over time, as existing systems struggle to effectively utilize system interaction events and other activity to provide relevant recommendations to users in content collaboration platforms.
Innovation Solution
A computer-implemented method for generating event feeds that includes recommending document spaces to users based on user-to-space classifiers, which are computed using past interaction events, allowing for personalized and contextually relevant feed items to be displayed in a graphical user interface, enabling users to subscribe to recommended document spaces and receive updates on associated events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the system tracks all digital items created and updated over time, then the quantity of tracked information increases, but the difficulty of tracking and managing these items increases
Solution Approach 1:
The system extracts only the most relevant digital items and events from the vast quantity of tracked data, using classification models to identify and present only those items that are likely to be of interest to each user, rather than attempting to manage or display all tracked items
Solution Approach 2:
The system uses user interactions with recommended items (clicks, follows, engagements) as feedback to continuously refine and retrain the classification models, improving the accuracy of future recommendations and making the tracking system more efficient at identifying relevant items
2Ease of operation
If the system provides personalized recommendations to users, then user engagement improves, but the complexity of the recommendation algorithm increases
Solution Approach 1:
The system applies different classification strategies and model complexities to different users based on their individual characteristics, interaction histories, and preferences, rather than using a single uniform algorithm for all users
Solution Approach 2:
The system pre-computes user profiles, item embeddings, and classification models in advance before users need recommendations, reducing the computational complexity required at the moment of interaction and enabling personalized recommendations without real-time computational burden
3Loss of information
If the system displays more feed items to users, then the information availability increases, but the relevance of individual feed items decreases
Solution Approach 1:
The system displays a limited number of highly relevant feed items rather than attempting to show all available items, using classification models to ensure that the subset of displayed items has maximum relevance to each user's interests and context
Data Source
AI summary
A method for recommending feed sources in an event feed includes generating an event feed comprising a plurality of feed items associated with a user. The event feed includes a recommendation feed item comprising one or more feed item sources, which may include a document space not currently followed by the user. The method further includes causing at least a portion of the event feed to be displayed to the user in the event feed. In accordance with a determination that the user is viewing a graphical user interface associated with a first software application, the recommendation feed item includes feed item sources associated with the first software application. In accordance with a determination that the user is viewing a graphical user interface associated with a second software application different from the first, the recommendation feed item includes feed item sources associated with the second software application.