Graph Embedding Vector Update via Online Training Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning methods for updating embedding vectors in graph data require high processing loads, especially when new data is added, as they often necessitate recalculating all vectors, which is inefficient for large datasets or frequent updates.
Innovation Solution
Implementing online training that updates only the specific embedding vectors related to the new data, with a determined online training range based on the distance from the new triple, reducing the processing load by limiting the scope of updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all embedding vectors are recalculated when new data is added to graph data, then the accuracy of the machine learning model is maintained, but the processing load increases significantly
Solution Approach 1:
The patent segments the graph data into multiple blocks and processes each block independently. When new data is added, only the specific block containing the new data and its related blocks are updated, rather than recalculating all embedding vectors for the entire graph. This segmentation approach maintains model accuracy while significantly reducing the processing load.
Solution Approach 2:
The patent applies local quality by updating only the embedding vectors in the affected block and its neighboring blocks when new data is added, rather than uniformly updating all blocks. This localized update strategy ensures that the accuracy is maintained in the relevant areas while avoiding unnecessary computation in unaffected areas, thus reducing overall processing load.
2Adaptability or versatility
If all embedding vectors are recalculated frequently to respond to data changes, then the model remains up-to-date, but the time consumption increases
Solution Approach 1:
By dividing the graph data into blocks, the patent enables selective updates of only those blocks that contain or are related to new data. This segmentation allows the system to respond quickly to data changes without the time penalty of recalculating all embedding vectors, thus maintaining adaptability while reducing time consumption.
Solution Approach 2:
The patent performs partial updates by recalculating only the embedding vectors in the affected block and its neighboring blocks, rather than performing a complete recalculation of all blocks. This partial action approach ensures the model remains sufficiently up-to-date with minimal time investment, balancing adaptability with time efficiency.
3Reliability
If the entire graph data is processed for machine learning, then comprehensive accuracy is achieved, but the complexity of the system increases
Solution Approach 1:
The patent simplifies system complexity by segmenting the large-scale graph data into manageable blocks that can be processed independently. Each block can be handled with simpler computational operations, and the overall system complexity is reduced through this modular approach while still achieving comprehensive accuracy through the collective processing of all blocks.
Data Source
AI summary
A storage medium storing a machine learning program that causes at least one computer to execute a process, the process includes acquiring information regarding a first triple that includes a first node, a second node, and a first edge that indicates a relationship between the first node and the second node; and updating a vector of a third node of a plurality of vectors and a vector of a third edge of the plurality of vectors, based on the information and the plurality of vectors, each of the plurality of vectors representing each of a plurality of nodes and each of a plurality of edges that indicate relationships between the plurality of nodes, the plurality of vectors being generated by machine learning that uses the plurality of nodes and the plurality of edges, the third node and the third edge being coupled to the first triple under a certain condition.