Deep Learning Ontology Extraction for New Domain Knowledge
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current natural language processing techniques, such as Word2vec, struggle to effectively extract and classify new domain knowledge from unstructured text data, particularly in large volumes like vehicle service reports, due to limitations in capturing adaptive context and relying heavily on surface-level phrases, which hinders the detection of new technologies and fault signatures.
Innovation Solution
A transfer learning based deep learning model that incorporates part-of-speech (POS) information and position features to generate dynamic context vectors, combining Word2vec with POS2vec, and uses a hybrid approach for labeling data to improve the accuracy of ontology extraction and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional Word2vec model is used for text processing, then the model is simple to implement, but it fails to capture adaptive context and relies heavily on surface-level phrases, hindering detection of new technologies
Solution Approach 1:
The patent merges Word2vec with POS2vec to create a hybrid model that combines word embedding with part-of-speech information. This integration allows the system to capture both semantic meaning and grammatical context, significantly improving the detection of new domain knowledge while maintaining manageable complexity through modular architecture
Solution Approach 2:
The patent introduces dynamic context vectors that adapt based on the input text characteristics. The model dynamically adjusts its processing based on POS tags and contextual information, enabling it to adapt to new technologies and domains without requiring complete retraining, thus improving versatility while controlling complexity
2Measurement precision
If manual labeling is used for ontology extraction, then the quality of labeled data is high, but the time and resources required are excessive for large volumes of data
Solution Approach 1:
The patent performs preliminary automatic labeling using the trained hybrid model before final validation. This preliminary action processes large volumes of data quickly, and only requires manual intervention for verification and correction of ambiguous cases, dramatically reducing total labeling time while maintaining high accuracy
Solution Approach 2:
The system implements feedback loops where manually corrected labels are used to retrain and improve the automatic labeling model. This continuous feedback mechanism allows the system to learn from errors and improve over time, maintaining high labeling accuracy while minimizing manual intervention requirements
3Quantity of substance
If large volumes of unstructured data are processed, then the quantity of domain knowledge extracted increases, but the computational resources and processing time required increase significantly
Solution Approach 1:
The patent segments the processing into distinct stages: text preprocessing, POS tagging, feature extraction, and classification. Each stage processes data independently and can be optimized separately, allowing efficient handling of large volumes of unstructured data while controlling computational resource consumption through modular processing
Solution Approach 2:
The system dynamically adjusts processing parameters such as vector dimensionality, batch size, and model depth based on the characteristics of the input data. This allows the model to optimize computational resource usage while maintaining effective processing of large-scale industrial data across different domains
Data Source
AI summary
A system processes unstructured data to identify a plurality of subsets of text in a set of text in the unstructured data and determines, for a subset from the plurality of subsets, probabilities based on a position of the subset in the set of text, a part of speech (POS) of each word in the subset, and POSs of one or more words on left and right hand sides of the subset, a number of the one or more words being selected based on a length of the set of text. The system generates a feature vector for the subset, the feature vector including the probabilities and additional features of the subset; and classifies, using a classifier, the subset into one of a plurality of classes based on the feature vector for the subset, the plurality of classes representing an ontology of a domain of knowledge.