Non-factoid Question-Answering System Causal Relation Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current question-answering systems for why-type questions fail to effectively utilize various expressions of causal relations in documents, limiting their ability to provide accurate answers, as they primarily focus on causal relations between noun phrases within a single sentence and overlook relations between sentences or between noun and verb phrases.
Innovation Solution
A non-factoid question-answering system that generates features from morpheme and syntactic information, as well as causal relation expressions, to evaluate the plausibility of answer candidates by considering causal relations between sentences and between noun and verb phrases, using clue expressions and semantic polarity to determine relevance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the system focuses on causal relations between noun phrases within a single sentence, then the system complexity remains manageable, but the measurement precision of answer accuracy deteriorates
Solution Approach 1:
The patent segments causal relation extraction into multiple independent modules: (1) noun phrase causal relation extraction, (2) verb phrase causal relation extraction, and (3) inter-sentence causal relation extraction. Each module handles a specific type of causal relation independently, allowing the system to comprehensively analyze diverse causal expressions while maintaining manageable complexity through modular design.
Solution Approach 2:
The patent extends causal relation analysis from a single dimension (intra-sentence noun phrase relations) to multiple dimensions by adding: (1) different grammatical structures (noun-verb-obj patterns), and (2) different textual scopes (intra-sentence and inter-sentence relations). This multi-dimensional approach improves answer accuracy by capturing comprehensive causal information without proportionally increasing system complexity.
2Reliability
If the system analyzes only causal relations between noun phrases in the same sentence, then the processing speed is maintained, but the reliability of answer selection deteriorates
Solution Approach 1:
The patent performs preliminary classification of question types and pre-identification of causal relation patterns before full analysis. By pre-processing documents to identify potential causal structures and pre-classifying question types, the system prepares data in advance, enabling more reliable answer selection through comprehensive causal analysis without significantly impacting processing speed during actual query execution.
Solution Approach 2:
The patent introduces an intermediary causal relation representation layer that mediates between raw text and answer selection. This intermediary layer structures causal information from various sources (noun phrases, verb phrases, inter-sentence relations) into a unified format, allowing reliable answer selection based on comprehensive causal evidence while maintaining efficient processing through standardized intermediate representations.
3Loss of information
If the system overlooks causal relations between sentences and between noun and verb phrases, then the ease of operation is maintained, but the loss of information deteriorates
Solution Approach 1:
The patent creates a universal causal relation extraction framework that handles multiple types of causal expressions through a single integrated system. The system universally processes: (1) noun phrase causal relations, (2) verb phrase causal relations, and (3) inter-sentence causal relations, minimizing information loss while maintaining operational simplicity through unified processing rules and consistent analysis methods across all causal relation types.
Data Source
AI summary
A question-answering system capable of appropriately answering a non-factoid question. The question-answering system responsive to a question, for taking out answer candidates from an object document storage unit; extracting plausible ones from the answer candidates; recognizing causal relation expression included in extracted answer candidates; determining whether or not a recognized causal relation is relevant as an answer; generating a feature vector related to a combination of the question and the searched answer candidate; and for calculating, when a feature vector is given, a score indicating a degree of how correct an answer candidate is as an answer to the question, for the question and the answer candidate as a source of the feature vector, and for outputting, as the answer to the question, an answer candidate having the highest score.