Document Summary System Using Decision Network for Sentence Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing natural language processing techniques for generating summaries from text documents either result in summaries that are not fluent and coherent when using only extracted sentences or suffer from redundancy when using only abstracted sentences, failing to balance readability and accuracy.
Innovation Solution
A system that processes an initial summary of extracted sentences by generating vector representations and using a decision network to determine whether to add the extracted sentence, an abstracted version of the sentence, or discard it, thereby creating a balanced summary that combines the clarity of abstracted sentences with the preservation of key information from extracted sentences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only extracted sentences are used in the summary, then accuracy is improved, but fluency and coherence deteriorate
Solution Approach 1:
The patent merges extracted sentences (for accuracy) with abstracted sentences (for fluency) by using a decision network to select between them. The system combines both approaches rather than relying on one exclusively, allowing the summary to maintain both accuracy and readability.
Solution Approach 2:
The system changes the parameter of sentence representation by generating vector representations of both extracted and abstracted sentences. This allows the decision network to evaluate and select sentences based on multiple parameters including accuracy, fluency, and coherence, optimizing the balance between them.
2Ease of operation
If only abstracted sentences are used in the summary, then fluency and coherence are improved, but redundancy increases
Solution Approach 1:
The patent extracts key information from the original document using extracted sentences, then uses abstracted sentences to paraphrase them. The decision network selectively uses abstracted versions only when they improve fluency without adding redundant information, thus reducing overall redundancy while maintaining readability.
Solution Approach 2:
The system uses vector representations and the decision network as feedback mechanisms to evaluate each sentence before inclusion. This feedback loop ensures that abstracted sentences are only added when they improve the summary quality without introducing redundancy, dynamically adjusting the summary composition.
3Measurement precision
If a decision network with vector representations is used, then summary quality is improved, but device complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-processing the original document to extract sentences and generate their abstracted versions before the main summarization process. Vector representations are pre-computed, and the decision network is pre-trained, so that during actual summarization, the system only needs to make classification decisions rather than perform complex computations in real-time.
Data Source
AI summary
A system for generating a summary of a text document is disclosed. In some examples, the system includes a processor configured to generate an initial summary of an original document. The initial summary includes a selection of extracted sentences copied from the original document. For each extracted sentence of the initial summary, the processor processes the extracted sentence to generate an abstracted sentence, and generates vector representations of the extracted sentence, the abstracted sentence, the original document, and the current summary. The vector representations are then input to a decision network to compute an editing decision. The editing decision is selected from a group of possible decisions that includes a decision to add the extracted sentence and a decision to add the abstracted sentence. The processor also updates the current summary based on the editing decision.