Text Expansion via Sentence Pair Extraction and Sequence-to-Sequence Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional sequence-to-sequence learning algorithms for machine translation and conversation models require paired learning data, such as translated sentences or questions and answers, and lack a method to provide learning sentences for general text, limiting their ability to expand input text effectively.
Innovation Solution
A system and method utilizing a word chain network technique and sequence-to-sequence model to automatically expand input text by extracting sentence pairs from different documents, inputting them into an encoder, and generating expanded text through a decoder, with a text expansion system comprising a communication module, memory, and processor to execute this process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional sequence-to-sequence learning algorithms are used for machine translation and conversation models, then translation accuracy and conversation understanding are improved, but the requirement for paired learning data limits the ability to expand general text
Solution Approach 1:
The patent creates synthetic training data by copying and transforming existing parallel sentences from different languages. The system extracts sentence pairs from multilingual corpora, translates them bidirectionally, and uses these copied transformations as training examples for text expansion, eliminating the need for original paired learning data while maintaining translation accuracy.
Solution Approach 2:
The system enables the model to generate its own training data through self-service mechanisms. By using the encoder-decoder architecture to translate texts and then using those translations to generate expansion data, the system serves itself without requiring external paired datasets, allowing general text expansion while preserving translation capabilities.
2Reliability
If paired learning data such as translated sentences or questions and answers is required, then model training effectiveness is improved, but no method exists to provide learning sentences for general text
Solution Approach 1:
The patent performs preliminary data preparation by extracting and storing parallel sentence pairs from multilingual corpora before the actual text expansion task. This preliminary action creates a reusable dataset of transformed sentences that can be applied to general text expansion while maintaining the reliability benefits of structured learning data.
Solution Approach 2:
The system creates a universal training approach that works for both translation tasks and general text expansion. The encoder-decoder model trained on multilingual parallel sentences becomes multi-functional, serving both translation purposes and general text expansion, eliminating the need for separate paired datasets for different applications.
3Manufacturing precision
If word chain network technique and sequence-to-sequence model are used to generate expanded text, then text expansion quality is improved, but system complexity increases due to multiple modules including encoder, decoder, and communication modules
Solution Approach 1:
The patent merges the encoder and decoder into a unified sequence-to-sequence model architecture that processes text expansion tasks. By combining these functions into an integrated neural network system with shared parameters and coordinated processing, the system maintains high text expansion quality while reducing overall structural complexity compared to separate independent modules.
Data Source
AI summary
Provided is a method of automatically expanding input text. The method includes receiving input text composed of a plurality of documents, extracting a sentence pair that is present in different documents among the plurality of documents, setting the extracted sentence pair as an input of an encoder of a sequence-to-sequence model, setting an output of the encoder as an output of a decoder of the sequence-to-sequence model and generating a sentence corresponding to the input, and generating expanded text based on the generated sentence.