Vector Space Alignment for Ambiguous Term Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text analysis methods face challenges in accurately classifying ambiguous terms that have different meanings in technical terminology versus general language, as they often fail to quantify the shift in meaning effectively.
Innovation Solution
The method involves creating vector space models using Word2Vec and fastText to represent words from both technical and general language texts, aligning these models through the orthogonal Procrustes problem to calculate vector distances, and utilizing the Personalized PageRank algorithm with a bias vector to analyze shifts in meaning, allowing for precise term extraction and search optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional rule-based and statistical methods are used for terminology extraction, then the analysis process is simple, but the accuracy in classifying ambiguous terms is insufficient
Solution Approach 1:
The patent transforms the text analysis problem into a vector space parameter comparison problem. By representing terms as vectors in multidimensional space and comparing their parameters (vector distances, angles) between general language and technical domains, the system achieves accurate classification of ambiguous terms through quantitative parameter analysis rather than traditional rule-based methods.
Solution Approach 2:
The patent introduces vector space models as an intermediary representation layer between raw text and classification results. Terms from both general language and technical domains are mapped into a shared vector space, where their semantic relationships are captured through vector operations. This intermediary representation enables precise comparison and classification of ambiguous terms.
2Loss of information
If vector space models are used to represent word meanings, then the representation of semantic similarity is improved, but the computational complexity increases
Solution Approach 1:
The patent merges the general language vector space and technical domain vector space into a unified comparison framework. By aligning corresponding vectors from both domains through Procrustes transformation and comparing them in a unified space, the system preserves semantic information from both sources while avoiding the need to maintain completely separate computational models.
Solution Approach 2:
The patent performs preliminary alignment of vector spaces using Procrustes transformation before conducting term classification. By pre-computing the optimal transformation matrix that aligns the general language vector space with the technical domain vector space, the system prepares the computational framework in advance, reducing the computational burden during actual term analysis and classification operations.
3Measurement precision
If multiple vector space models are aligned through Procrustes transformation, then the comparison accuracy between domains is improved, but the processing time increases
Solution Approach 1:
The patent changes the problem from comparing raw text or simple word embeddings to comparing aligned vectors in a transformed coordinate system. By applying Procrustes transformation that optimizes the alignment between general language and technical domain vector spaces, the system achieves precise measurement of meaning shifts through parameter comparison in the aligned space, making the measurement process more accurate and efficient.
Data Source
AI summary
Methods for analyzing elements of a first text collection, and devices and methods for searching for information in a set of digital data, in particular in a computer network.