Weakly Supervised Task Extraction from Documents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies do not effectively address the extraction of tasks from documents, which is crucial for applications like candidate shortlisting and common sense knowledge augmentation, due to the lack of explicit methods for identifying task mentions with considerations for expertise and volitionality.
Innovation Solution
A system and method for training a word-level data model using weakly supervised classification to extract tasks from documents, which involves preprocessing documents, labeling words as task or no-task headwords based on linguistic rules, and training a BERT-based classification model for task extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If weakly supervised classification is used for task extraction, then extraction accuracy is improved by capturing complete meaning of task mentions, but computational complexity increases due to preprocessing and model training requirements
Solution Approach 1:
The patent applies preliminary action by performing extensive preprocessing of documents before the actual task extraction. This includes tokenization, part-of-speech tagging, dependency parsing, and creating training datasets with linguistic rules. These preparatory steps are executed in advance to enable accurate weakly supervised classification during the actual extraction phase, resolving the contradiction between accuracy and complexity by front-loading the computational work.
2Measurement precision
If linguistic rules are applied for labeling words as task or no-task headwords, then extraction precision is improved by considering expertise and volitionality, but processing time increases due to multiple preprocessing steps
Solution Approach 1:
The patent applies segmentation by breaking down the document processing into distinct modular steps: tokenization, part-of-speech tagging, dependency parsing, and task headword identification. Each step processes specific linguistic features independently, allowing for optimized execution and parallel processing. This modular approach maintains high precision through comprehensive linguistic analysis while reducing overall processing time compared to monolithic processing methods.
3Adaptability or versatility
If BERT-based model is trained for task extraction, then ability to handle polysemy is improved, but training data requirements and computational resources increase
Solution Approach 1:
The patent introduces an intermediary approach by using weakly supervised classification with linguistically-rulelabeled training data as a mediator between raw documents and the BERT model. The linguistic rules and preprocessed features serve as intermediaries that guide the model's learning process, enabling it to handle polysemy effectively without requiring massive amounts of manually annotated training data. This intermediary layer transfers linguistic knowledge into the model training process.
Data Source
AI summary
This disclosure relates to extraction of tasks from documents based on a weakly supervised classification technique, wherein extraction of tasks is identification of mentions of tasks in a document. There are several prior arts addressing the problem of extraction of events, however due to crucial distinctions between events-tasks, task extraction stands as a separate problem. The disclosure explicitly defines specific characteristics of tasks, creates labelled data at a word-level based on a plurality of linguistic rules to train a word-level weakly supervised model for task extraction. The labelled data is created based on the plurality of linguistic rules for a non-negation aspect, a volitionality aspect, an expertise aspect and a plurality of generic aspects. Further the disclosure also includes a phrase expansion technique to capture the complete meaning expressed by the task instead of merely mentioning the task that may not capture the entire meaning of the sentence.