Multitask Deep Learning Model for Cohort Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning technologies for online networks lack the ability to provide personalized, contextual, and cross-pillar holistic recommendations for next best actions, failing to optimize multiple objectives and scale effectively within online platforms.
Innovation Solution
A multitask deep machine learning model is employed, integrated with an artificial intelligence engine and backend service, utilizing user intent and action sequences to provide personalized recommendations across different portions of the online network, optimizing various objectives and scaling easily, while considering user interactions and context.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models are used for action recommendations, then implementation is simpler, but personalization and contextual understanding are insufficient
Solution Approach 1:
The system segments the recommendation task into multiple independent components: viewer feature extraction, viewee feature extraction, relationship feature extraction, and cohort ranking. Each component processes specific aspects of the data independently, allowing the complex personalization task to be divided into manageable segments that can be trained and optimized separately while maintaining high accuracy.
Solution Approach 2:
The deep machine learning model is designed to perform multiple functions simultaneously: it ranks cohorts based on viewer-viewee relationships, infers user intent from action sequences, and adapts to different contexts across various portions of the online network. This multi-functional approach enables a single model to handle diverse recommendation scenarios without requiring separate models for each case.
2Adaptability or versatility
If deep machine learning models are used for comprehensive recommendations, then personalization and contextual understanding improve, but computational resources and training complexity increase
Solution Approach 1:
The system performs preliminary actions by pre-processing and extracting features from viewer profiles, viewee profiles, and action sequences before the actual recommendation task. Viewer features, viewee features, and relationship features are extracted and stored in advance, allowing the deep learning model to receive pre-processed input that reduces computational complexity during inference while maintaining high contextual adaptability.
Solution Approach 2:
The system transforms the recommendation problem from a simple ranking task into a multi-dimensional analysis by incorporating viewer features, viewee features, relationship features, and action sequences. This dimensional expansion allows the model to capture complex contextual relationships and adapt to different scenarios, while the modular feature extraction architecture keeps the overall system complexity manageable.
3Reliability
If multiple objectives are optimized simultaneously, then recommendation quality improves, but training difficulty and computational cost increase
Solution Approach 1:
The system merges multiple objective functions into a unified training framework for the deep machine learning model. Instead of training separate models for different objectives (cohort ranking, intent inference, contextual adaptation), the model is trained simultaneously on all objectives with a combined loss function. This merging approach improves recommendation reliability by considering multiple factors together, while the shared model architecture reduces overall training complexity compared to maintaining separate models.
Data Source
AI summary
In an example embodiment, a deep machine learning model ranks cohorts of users as well as cohorts of products in a single ranking. When utilized to determine which cohort members to display to a user, the system selects one user cohort and one product cohort as the “best” (e.g., the top ranked user cohort and the top ranked product cohort). This ranking may be based on a number of contextual and non-contextual features, including viewer features (characteristics of the user operating the user interface), viewee features (characteristics of or related to the litem that the user is viewing, such as the characteristics of another user whose profile the user is viewing), and viewer-viewee relationship features (indications about how the viewer and viewee are related, such as common schools, locations, places of employment, etc.).