Log Entry Encoding via Trigonometric Transformations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current natural language processing (NLP) methods struggle to accurately handle numerical data in database query logs, leading to decreased accuracy in anomaly detection and other tasks due to the inability to capture semantics and handle out-of-vocabulary words.
Innovation Solution
A novel approach that uses dedicated transformations to extract information from log-specific fields, converting numbers to scientific notation and applying trigonometric transformations for timestamps, allowing NLP methods to treat log entries as natural language sentences for improved encoding and anomaly detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If NLP methods are used to encode log entries, then textual data encoding is effective, but accuracy deteriorates when numerical data is introduced
Solution Approach 1:
The patent transforms numerical parameters into angular representations using sine and cosine functions. Specifically, numeric values are converted to angle representations where the original number is encoded as sin(2πx) and cos(2πx), transforming the parameter space from linear numerical values to circular angular coordinates. This allows NLP models to capture numerical semantics including magnitude, order, and periodicity relationships.
Solution Approach 2:
The patent introduces angular representations as an intermediary between numerical data and NLP processing. Instead of directly feeding numerical values to the NLP model, the numbers are first transformed into angular embeddings that preserve numerical relationships. This intermediary representation enables the NLP model to understand numerical concepts without being sensitive to the specific numerical format.
2Ease of manufacture
If traditional encoding methods are used, then simplicity is maintained, but inability to capture semantics decreases ML performance
Solution Approach 1:
The patent transforms categorical parameters into angular representations similar to numerical parameters. Categorical values are mapped to angles around a circle, where semantically similar categories are positioned closer together in the angular space. This allows the NLP model to capture semantic relationships between categories while maintaining a unified encoding approach.
3Ease of operation
If standard NLP tokenization is used, then text processing is straightforward, but out-of-vocabulary words cannot be handled
Solution Approach 1:
The patent transforms discrete categorical values into continuous angular representations. Instead of relying on a fixed vocabulary where each category has a discrete index, the system maps categories to continuous angular positions. This allows the model to generalize to unseen categories by positioning them appropriately in the angular space based on their semantic relationships to known categories.
4Loss of information
If numerical values are directly fed to NLP models, then data representation is direct, but robustness of numerical representation is reduced
Solution Approach 1:
The patent transforms linear numerical parameters into angular parameters using trigonometric functions. The transformation x → (sin(2πx), cos(2πx)) maps numerical values to points on a unit circle, preserving the periodicity and relative distance relationships. This angular representation is more robust because it naturally handles numerical overflow, underflow, and maintains semantic relationships even when absolute values vary.
Data Source
AI summary
Herein is natural language processing (NLP) to detect an anomalous log entry using a language model that infers an encoding of the log entry from novel generation of numeric lexical tokens. In an embodiment, a computer extracts an original numeric lexical token from a variable sized log entry. Substitute numeric lexical token(s) that represent the original numeric lexical token are generated, such as with a numeric exponent or by trigonometry. The log entry does not contain the substitute numeric lexical token. A novel sequence of lexical tokens that represents the log entry and contains the substitute numeric lexical token is generated. The novel sequence of lexical tokens does not contain the original numeric lexical token. The computer hosts and operates a machine learning model that generates, based on the novel sequence of lexical tokens that represents the log entry, an inference that characterizes the log entry with unprecedented accuracy.