Similarity-Based Negative Sampling for Knowledge Base Completion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing negative sampling techniques for knowledge base completion generate nonsensical negative samples due to their random nature, which are ill-suited for training machine learning technologies, as they often replace entities or relations in a way that results in incompatible contexts.
Innovation Solution
The system employs similarity-based negative sampling, using vector space models to determine similarity metrics for entities and relations, allowing for the generation of meaningful negative samples that are distributionally similar to true assertions but false, thereby facilitating the training of machine learning technologies to distinguish between true and false assertions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If random negative sampling is used, then the process is simple and fast, but the generated negative samples are nonsensical and incompatible
Solution Approach 1:
The patent changes the parameter of sampling from random selection to similarity-based selection. By computing similarity metrics between entities and using these metrics to guide sampling, the system generates negative samples that are distributionally similar to true assertions but factually false, thereby improving sample quality while maintaining efficiency
Solution Approach 2:
The patent introduces similarity metrics as an intermediary between the knowledge base and the negative sampling process. These metrics serve as a mediator that evaluates entity relationships and guides the selection of appropriate negative samples, ensuring they are both plausible and useful for training
2Ease of manufacture
If random negative sampling is used, then the implementation is simple, but the training effectiveness is poor
Solution Approach 1:
The patent modifies the sampling parameters from uniform random distribution to similarity-weighted distribution. By incorporating similarity metrics into the sampling process, the system generates negative samples that better reflect real-world relationships, significantly improving training accuracy without complicating the overall implementation
3Reliability
If similarity-based negative sampling is used, then the negative samples are meaningful and distributionally similar, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary computation of similarity metrics between entities and stores these pre-computed values for use during the negative sampling process. This preliminary action reduces the computational burden during actual sampling by avoiding redundant similarity calculations, thereby improving sample quality while managing computational complexity
Data Source
AI summary
Techniques regarding similarity based negative sample analysis are provided. For example, one or more embodiments described herein can comprise a system, which can comprise a memory that can store computer executable components. The system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can comprise a similarity component that can determine similarity metrics for respective entities based on a vector space model. The respective entities can be represented by a dataset. Also, the computer executable components can comprise a sampling component that can perform a negative sampling analysis on the dataset based on the similarity metrics.