Mobile Content Recommendation via Server-Side Model Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating personalized content on mobile devices are inefficient due to high computational requirements and limited processor power, and traditional recommendation systems fail to consider user-specific factors like user groups, time, and location, leading to low accuracy in content recommendations.
Innovation Solution
A system using a wide and deep machine learning model with transfer learning to generate user-specific content classification models, which captures historical usage patterns and seasonality, allowing for resource-efficient and dynamic content recommendations on mobile devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex machine learning models are used for personalized content recommendation, then recommendation accuracy is improved, but computational resource requirements and processing time increase significantly
Solution Approach 1:
The patent divides the recommendation system into two segments: a server-side component that performs complex model training and generation, and a mobile device component that executes the generated model. This segmentation allows the mobile device to use lightweight models while the server handles computationally intensive tasks, resolving the contradiction between accuracy and processing power.
Solution Approach 2:
The system performs preliminary actions by pre-training complex recommendation models on the server using large datasets, then generating simplified versions of these models for deployment on mobile devices. This preliminary training approach allows accurate recommendations to be achieved without requiring the mobile device to perform heavy computational work in real-time.
2Measurement precision
If complex machine learning models are trained and re-trained frequently, then recommendation accuracy is improved, but time consumption and processing overhead increase
Solution Approach 1:
The patent separates model training operations from model execution operations. The server performs model training and generation as a preliminary action, while the mobile device only performs inference using the pre-generated model. This segmentation eliminates the need for frequent re-training on the mobile device, reducing time consumption while maintaining accuracy.
Solution Approach 2:
The system performs model training and generation as preliminary actions on the server before deployment to mobile devices. This approach allows the model to be ready for use without requiring time-consuming training on the mobile device, thus reducing the time loss associated with frequent re-training while maintaining recommendation accuracy.
3Speed
If traditional recommendation algorithms are used, then processing speed is maintained, but user-specific factors like time, location, and usage patterns are not considered leading to lower recommendation quality
Solution Approach 1:
The patent segments the recommendation system so that the mobile device handles fast local processing using pre-trained models, while the server handles complex model generation and updates. This allows the system to maintain processing speed on the mobile device while incorporating sophisticated user-specific factors in the server-side model generation, thus resolving the contradiction between speed and recommendation quality.
4Measurement precision
If large datasets are used for model training, then recommendation accuracy is improved, but memory storage and data transmission requirements on mobile devices increase
Solution Approach 1:
The patent segments the data handling responsibilities between server and mobile device. The server stores and processes large training datasets, while the mobile device receives only the trained model parameters. This segmentation allows high recommendation accuracy to be achieved using large datasets without requiring the mobile device to store or transmit large amounts of raw data, thus resolving the contradiction between accuracy and data quantity requirements.
Data Source
AI summary
Methods and apparatuses are described for generating and displaying custom-selected content for a mobile application. A mobile device captures a request to launch an application and determines content cards eligible to be displayed. A user-specific content classification model is generated for selecting one of the content cards to be displayed, where the mobile device identifies historical usage characteristics and transmits the historical usage characteristics to a server, which generates the user-specific content classification model based upon the historical usage characteristics and transmits the user-specific content classification model to the mobile device. The mobile device executes the user-specific content classification model to select one or more content cards to be displayed and launches the application to display the selected content cards.