Document Classification via Text Segments for Localized Concept Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document classification methods struggle with scalability, adaptability, and the seamless integration of multimodal data, particularly in handling diverse and evolving document types, and fail to identify weak signals and topic combinations effectively.
Innovation Solution
A computer-implemented method that segments text into multiple sequences, calculates sequence vectors, classifies each sequence into categories, constructs a document vector, and applies a classification model to identify localized concepts and improve interpretability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning architectures (CNNs, RNNs, transformers) are used for document classification, then classification accuracy is improved, but computational resource requirements and processing time increase significantly
Solution Approach 1:
The patent segments the document into multiple text sequences and processes each sequence independently through the classification model. This divides the large computational task into smaller, parallelizable units, reducing peak memory usage and enabling processing on devices with limited computational resources while maintaining accurate classification results.
Solution Approach 2:
The patent applies classification to text sequences rather than requiring complete processing of entire documents at once. By classifying partial segments (text sequences) independently and aggregating results, the system achieves effective document classification with reduced computational overhead compared to processing complete documents through heavy deep learning models.
2Reliability
If entire documents are processed at once for classification, then comprehensive analysis is achieved, but processing time and computational load increase
Solution Approach 1:
The document is divided into multiple text sequences that can be processed in parallel or sequentially with smaller computational batches. This segmentation maintains comprehensive analysis by covering all parts of the document while reducing processing time through efficient resource utilization and potential parallelization.
Solution Approach 2:
The classification process operates continuously on text sequences as they are extracted from the document, rather than waiting for complete document processing. This continuous action on segments enables faster turnaround while maintaining comprehensive coverage of the entire document through aggregation of segment classifications.
3Productivity
If standard classification methods are used, then processing speed is maintained, but weak signals and localized concepts are not detected
Solution Approach 1:
The patent applies classification at the text sequence level rather than only at the document level, enabling local quality assessment of different parts of the document. This localized classification approach detects weak signals and specific concepts in particular segments that would be diluted or missed in whole-document classification, while maintaining efficient processing speeds.
Data Source
Figure 1
Figure 2
AI summary
The invention relates to a computer-implemented method (100) for classifying a digital document comprising text, the method comprising the following steps: - Segmenting (130), by one or more processors (10), the text of the digital document (Di) into a plurality of text sequences (Sj); - Calculating (140), by one or more processors (10), a sequence vector (VSj) for each text sequence (Sj); - Classifying (150), by one or more processors (10), each text sequence (Sj) into at least one of a plurality of categories (Ck); - Constructing (160), by one or more processors (10), an n-dimensional document vector (VDi) within which at least one dimension corresponds to each of said categories, s; and - Applying (170), by one or more processors (10), a classification model to the n-dimensional document vector (VDi)