Transformer-Based Recall Data Ranking with Contextual Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing intelligent search methods fail to accurately rank recall data due to the neglect of contextual information between data sets, resulting in poor ranking accuracy.
Innovation Solution
A method that determines a recommendation degree for each recall data based on its similarity to other recall data, using a transformer module to fuse feature vectors and consider contextual information, thereby improving ranking quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional ranking methods are used that only consider individual data characteristics, then the ranking process is simple and fast, but the ranking accuracy is poor due to neglect of contextual information between data sets
Solution Approach 1:
The patent merges individual data characteristics with contextual relationships between multiple recall data items. The transformer module combines feature vectors from multiple recall data to generate comprehensive embedding vectors that incorporate both individual attributes and contextual information, thereby improving ranking accuracy while maintaining computational efficiency through unified processing.
Solution Approach 2:
The patent introduces embedding vectors as an intermediary representation that captures contextual relationships between recall data. These embedding vectors serve as mediators that translate complex contextual information into a format suitable for ranking, enabling the system to consider relationships between data sets without significantly increasing computational complexity.
2Measurement precision
If contextual information between recall data is considered, then ranking accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary encoding of individual recall data into feature vectors and embedding vectors before the ranking process. This preliminary action prepares the data in an optimized format that facilitates efficient contextual analysis during ranking, reducing the computational energy required during the actual ranking operation by avoiding redundant processing.
Solution Approach 2:
The patent replaces traditional mechanical ranking approaches with a neural network-based transformer module that uses learned embeddings to capture contextual relationships. This substitution enables the system to efficiently process contextual information through differentiable operations rather than computationally expensive traditional methods, reducing overall energy consumption while improving accuracy.
Data Source
AI summary
A method is provided that includes: determining a plurality of recall data associated with data to be searched; determining, for each recall data of the plurality of recall data, a recommendation degree of the recall data based on a similarity between the recall data and each recall data of the plurality of recall data; and ranking the recall data in the plurality of recall data based on the recommendation degree of each recall data of the plurality of recall data.