Document Categorization Model Using Predicate PMI Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing unsupervised learning methods for generating learned models suffer from low inference accuracy.
Innovation Solution
A method involving supervised learning to generate learned models by associating documents with categories, performing morphological analysis, extracting predicate words, and calculating pointwise mutual information to improve inference accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If unsupervised learning is used to generate learned models, then the learning process can be performed without labeled data, but the inference accuracy of the generated model is low
Solution Approach 1:
The patent introduces labeled data as an intermediary element to bridge the gap between raw data and model generation. By using labeled data to train a classifier that generates labels for unsupervised learning, the system achieves high inference accuracy while maintaining the benefits of unsupervised learning approaches.
Solution Approach 2:
The patent changes the parameter of learning approach from pure unsupervised learning to a hybrid approach that incorporates labeled data. This parameter change enables the system to achieve high inference accuracy by leveraging the supervisory signal from labeled data during the learning process.
2Reliability
If supervised learning with labeled data is used, then high inference accuracy can be achieved, but the requirement for large amounts of labeled data increases complexity and cost
Solution Approach 1:
The patent uses a classifier to generate synthetic labeled data by copying and transforming features from existing labeled examples. This allows the system to create additional training data that maintains the high inference accuracy while reducing the need for collecting and annotating large amounts of original labeled data.
Solution Approach 2:
The patent applies partial supervised learning by using labeled data only for training the initial classifier, while the subsequent unsupervised learning phase processes the majority of data without requiring labels. This partial application of supervision achieves high accuracy without requiring complete labeling of all training data.
Data Source
AI summary
An information processing device includes an acquisition unit that acquires multiple pieces of learning data in each of which a document and a category have been associated with each other, a morphological analysis performance unit that performs morphological analysis on each of the multiple pieces of learning data, an extraction unit that extracts words being predicates from among a plurality of words obtained by the morphological analysis, and a calculation generation unit that generates a learned model by calculating pointwise mutual information based on the plurality of words obtained by the morphological analysis, a plurality of extracted words, and a plurality of categories, the learned model being a learned model which outputs a category corresponding to data when the data is inputted.