Bit Sequence Data Classification via Token Length Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data classification methods are computationally complex, inefficient, and prone to errors due to incomplete knowledge or misconceptions about expected classes, especially when dealing with large datasets or encrypted data, and they often require complex initialization and parsing of higher-order data objects.
Innovation Solution
A computer-implemented method for image classification that uses tokenization to generate bit sequences from multiple field values, analyzing these sequences to identify feature-like tokens based on length, and storing them in subset-separated form, allowing for fast and resource-efficient classification without the need for complex initialization or parsing of higher-order data objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing classification methods are used, then classification can be performed, but computational complexity increases and processing speed decreases
Solution Approach 1:
The patent segments the data processing into two distinct phases: a training phase where bit sequences are generated and stored, and a classification phase where only bit sequence length analysis is performed. This segmentation allows the complex tokenization work to be done once during training, while the actual classification operates on simplified bit length comparisons, dramatically reducing computational complexity during processing.
Solution Approach 2:
The patent extracts only the essential feature (bit sequence length) for classification purposes, discarding unnecessary semantic analysis and higher-order data object parsing. By taking out only the critical dimension for differentiation and ignoring redundant information, the system achieves fast classification without the computational burden of complete data object analysis.
2Productivity
If complex initialization and parsing of higher-order data objects is performed, then data can be processed, but time consumption increases
Solution Approach 1:
The patent performs all complex tokenization and bit sequence generation work during the training phase as a preliminary action. Once the bit sequences are generated and stored, the actual classification process only requires simple length comparisons, eliminating the need for repeated complex parsing and initialization during processing operations.
3Reliability
If semantic analysis of data content is performed, then classification accuracy can be improved, but the system becomes vulnerable to encrypted data and requires extensive knowledge
Solution Approach 1:
The patent introduces bit sequence length as an intermediary feature that bridges the gap between raw data and classification categories. Instead of directly analyzing semantic content (which fails on encrypted data), the system uses bit sequence length as a mediator that works uniformly on all data types including encrypted data, while still enabling meaningful classification through statistical patterns in the length distributions.
4Measurement precision
If specialized parsers are created for each data type, then parsing accuracy improves, but system complexity and maintenance difficulty increase
Solution Approach 1:
The patent creates a universal tokenization approach that handles multiple data types (text, numbers, dates, images, audio, video) through a single unified process. Instead of requiring specialized parsers for each data type, the system uses a general tokenization mechanism that extracts bit sequences from any data type, eliminating the need for multiple specialized parsing systems while maintaining accuracy through consistent bit-level processing.
Data Source
Figure 1a
Figure 1b~2
Figure 3~4
AI summary
The invention relates to a computer-implemented method for data classification. The method comprises: - providing (402) a set of tokens (153) containing tokens generated from multiple field values of multiple data records (DR1-DR8) by tokenization, wherein the tokens were generated from field values of at least two different field types (F1-F7), and wherein the tokens are stored in the form of a bit sequence; - analyzing (404) one or more features of the tokens at the bit sequence level to identify subsets (420, 422, 424) of feature-like tokens, wherein the features comprise the bit sequence of the tokens and/or the length of the bit sequence; - storing (406) a copy of each of the subsets of feature-like tokens in subset-separated form (426, 428, 430), wherein each subset copy represents a class of feature-like data.