Transformer Hybrid Recommendation Model Contextual Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation engines struggle to provide dynamically relevant recommendations as user needs change over time, leading to decreased accuracy and user engagement.
Innovation Solution
A two-tower model that incorporates contextual features through natural language processing (NLP) transformers to dynamically infer user embeddings, enabling context-sensitive recommendations by combining user, item, and interaction features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If state of the art recommendation engines are used, then recommendations are generated efficiently, but recommendation accuracy decreases over time as user needs change
Solution Approach 1:
The patent implements dynamic user embeddings that are updated in real-time based on current context signals (device type, location, time, network conditions) rather than using static historical profiles. This allows the recommendation system to adapt dynamically to changing user needs and environmental conditions, resolving the contradiction between maintaining reliability and improving adaptability
Solution Approach 2:
The system changes the parameters used for recommendation by incorporating contextual features (device, location, time, network) into the user embedding vector. These parameter changes enable the system to capture evolving user preferences and contextual factors, improving both accuracy and adaptability simultaneously
2Reliability
If contextual features are incorporated into the recommendation model, then recommendation accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the recommendation model into two independent towers: a user tower that processes user features and context to generate user embeddings, and an item tower that processes item features to generate item embeddings. This segmentation allows contextual features to be incorporated systematically without overwhelming computational complexity, as each tower can be optimized independently
Solution Approach 2:
The user tower is designed to be multi-functional, handling both user historical behavior analysis and current context interpretation within a single unified model. This universal approach allows the system to process diverse contextual features (device, location, time, network) through a common architecture, improving accuracy without proportionally increasing complexity
3Productivity
If user embeddings are dynamically inferred with context, then user engagement improves, but processing time increases
Solution Approach 1:
The patent pre-computes and stores item embeddings in the item tower during an offline phase, so that during online recommendation generation, only user embeddings need to be computed in real-time. This preliminary action significantly reduces online processing time while maintaining dynamic contextual adaptation, thus improving user engagement without excessive time loss
Data Source
AI summary
In a computer-implemented embodiment, an interaction machine learning model is trained based on many interactions on many resources. A context lexical token is inferred that represents a current operational context of a user. The context lexical token is inserted into a sequence of other inferred lexical tokens. From the context lexical token within the sequence of tokens, the interaction machine learning model infers a predicted resource that will be accessed next. In an embodiment, accelerated matchmaking entails suitability measurement by a dot product of a) a dynamically inferred user embedding that is based on the context lexical token and b) a statically inferred item embedding.