Document Classification for Data Loss Prevention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional data loss prevention (DLP) systems face challenges with computational intensity, latency, and high rates of false positives or false negatives due to the inclusion of irrelevant documents in training sets, which hampers efficient and accurate data classification.
Innovation Solution
The method involves performing semantic analysis, such as latent Dirichlet allocation, on prospective training documents to identify and exclude irrelevant documents based on similarity and predominance thresholds, ensuring that only relevant data is used to train machine learning classifiers for DLP, thereby improving classification accuracy and reducing false positives/negatives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional DLP systems use heuristics and formulas to detect sensitive data, then data protection coverage is improved, but computational resource consumption increases and latency increases
Solution Approach 1:
The system performs semantic analysis and topic modeling on training documents before the actual DLP classification task. By pre-processing training data to identify and remove irrelevant documents, the system prepares optimized training sets in advance, reducing the computational burden during real-time data protection operations.
Solution Approach 2:
The system extracts and removes irrelevant documents from training sets by identifying documents where unrelated topics predominate. This extraction of harmful elements (irrelevant documents) from the training data improves the quality of training while reducing the computational resources needed for model training and inference.
2Reliability
If traditional DLP systems use heuristics and formulas to detect sensitive data, then data protection coverage is improved, but false positive rate increases and false negative rate increases
Solution Approach 1:
The system performs semantic analysis and topic modeling on training documents before the actual DLP classification task. By pre-processing training data to identify and remove irrelevant documents, the system prepares optimized training sets in advance, reducing the computational burden during real-time data protection operations.
Solution Approach 2:
The system extracts and removes irrelevant documents from training sets by identifying documents where unrelated topics predominate. This extraction of harmful elements (irrelevant documents) from the training data improves the quality of training while reducing the computational resources needed for model training and inference.
3Quantity of substance
If traditional DLP systems include all prospective training documents, then training data volume is improved, but classification accuracy deteriorates due to irrelevant documents
Solution Approach 1:
The system extracts and removes irrelevant documents from training sets by identifying documents where unrelated topics predominate. This extraction of harmful elements (irrelevant documents) from the training data improves the quality of training while reducing the computational resources needed for model training and inference.
Solution Approach 2:
The system changes the quality parameter of training data by removing irrelevant documents identified through semantic analysis. Instead of using all prospective training documents, the system filters the training set to include only relevant documents, transforming the training data from quantity-focused to quality-focused.
Data Source
AI summary
A computer-implemented method for classifying documents for data loss prevention may include 1) identifying training documents for a machine learning classifier configured for data loss prevention, 2) performing a semantic analysis on training documents to identify topics within the set training documents, 3) applying a similarity metric to the topics to identify at least one unrelated topic with a similarity to the other topics within the plurality of topics, as determined by the similarity metric, that falls below a similarity threshold, 4) identifying, based on the semantic analysis, at least one irrelevant training document within the set of training documents in which a predominance of the unrelated topic is above a predominance threshold, and 5) excluding the irrelevant training document from the set of training documents based on the predominance of the unrelated topic within the irrelevant training document. Various other methods, systems, and computer-readable media are also disclosed.