ML Tokenization for Recommendation Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Enterprise organizations face challenges in identifying customer needs and managing budgets effectively for small business clients, particularly in generating personalized recommendations based on user account information and media platform activities.
Innovation Solution
A computing platform utilizing multiple machine learning models processes user account data and media data to generate recommendations by creating a fully connected graph structure, connecting tokens representing account and media features, and deleting non-matching tokens to produce a recommendation score, which is then sent to the user's device as a text message, email, or push notification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple machine learning models process user account data and media data to generate personalized recommendations, then recommendation accuracy and personalization improve, but system complexity and computational resources increase
Solution Approach 1:
The system segments the recommendation task into multiple specialized machine learning models: a user behavior analysis model that processes account data, a media content analysis model that processes media data, and a recommendation generation model that combines their outputs. Each model focuses on specific aspects of the recommendation problem, improving overall accuracy while managing complexity through functional decomposition
Solution Approach 2:
The patent introduces intermediate processing layers including feature extraction modules that transform raw account and media data into meaningful features, and a scoring mechanism that mediates between model outputs and final recommendations. These intermediaries simplify the interaction between complex models and the rest of the system, managing computational complexity while maintaining recommendation precision
2Adaptability or versatility
If the system processes historical account data, user-defined rules, and unstructured media data through multiple ML models, then recommendation personalization improves, but processing time and computational cost increase
Solution Approach 1:
The system performs preliminary processing of user account data and media data before the recommendation generation phase. User behavior patterns are pre-analyzed and stored as processed features, and media content is pre-tagged and categorized. When a recommendation is needed, the system retrieves these pre-processed features rather than re-analyzing raw data, significantly reducing processing time while maintaining personalized recommendation quality
Solution Approach 2:
The patent implements a hybrid approach where only certain aspects of user data and media data are fully processed through complex machine learning models, while other aspects use simpler rule-based processing or pre-computed features. This selective processing reduces computational time and resources while still achieving high personalization for the most important recommendation factors
3Measurement precision
If the recommendation system incorporates external events and location information to modify recommendations, then recommendation relevance improves, but data processing complexity and system requirements increase
Solution Approach 1:
The system applies different processing strategies to different types of contextual data based on their importance and characteristics. Location information receives localized processing that considers geographic context, while external events are filtered and weighted based on their relevance to the user's situation. This differentiated approach improves recommendation relevance by focusing computational resources on the most impactful contextual factors without uniformly increasing complexity across all data types
Data Source
AI summary
Aspects of the disclosure relate to generating recommendations for a user based on the customer's account information and the customer's activity on one or more media platforms using multiple machine learning (ML) models. A computing platform may determine a plurality of account features based on the account information via a user ML model. The computing platform may determine a plurality of media features based on unstructured media data via a media ML model. A recommendation ML model generates tokens representing each of the plurality of account features and each of the plurality of media features in a fully connected graph structure. The recommendation ML model processes and outputs a recommendation score based on the tokens in the fully connected graph structure. A recommendation is generated by the computing platform based on the recommendation score.