Neural Network Answer Scoring via Passage Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural network systems are inefficient in determining how well a text passage answers a question, requiring extensive data analysis and computational resources, and struggle to generate accurate answer scores for input passage-question pairs.
Innovation Solution
A neural network system comprising an encoder neural network, a decoder neural network, and a matching vector neural network is used to generate answer scores and output question strings, reducing the need for large datasets and computational resources by encoding a single passage and scoring multiple questions, rather than encoding multiple passages for a single question.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional neural network systems are used to determine how well a text passage answers a question, then answer score generation is achieved, but computational resources and data analysis requirements are extensive
Solution Approach 1:
The patent inverts the traditional question-answering approach by first encoding the passage once and then scoring multiple questions against that encoded representation. Instead of encoding each passage-question pair separately, the system encodes the passage once and reuses the encoding to score multiple questions, significantly reducing computational resources while maintaining answer score accuracy
Solution Approach 2:
The encoded passage representation serves multiple functions: it acts as a universal feature representation that can be compared against multiple different questions simultaneously. This multi-functional encoding allows the system to score multiple passage-question pairs without re-encoding the passage, reducing overall computational requirements
2Measurement precision
If traditional neural network systems analyze extensive data to generate answer scores, then measurement precision is improved, but processing time increases
Solution Approach 1:
The system performs preliminary encoding of the passage before scoring questions. By pre-computing the encoded representation of the passage, the system avoids re-encoding the passage for each question, thereby reducing processing time while maintaining the precision of answer score generation through the use of this pre-computed encoding
3Measurement precision
If the system encodes multiple passages for a single question, then answer score accuracy is improved, but device complexity and data requirements increase
Solution Approach 1:
The patent inverts the traditional approach by encoding the passage once and then scoring multiple questions against that encoding, rather than encoding multiple passages for each question. This inversion reduces data analysis complexity while maintaining answer score accuracy by using the single encoded passage representation as a universal comparator for multiple questions
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying answers to questions using neural networks. One of the methods includes receiving an input text passage and an input question string; processing the input text passage using an encoder neural network to generate a respective encoded representation for each passage token in the input text passage; at each time step: processing a decoder input using a decoder neural network to update the internal state of the decoder neural network; and processing the respective encoded representations and a preceding output of the decoder neural network using a matching vector neural network to generate a matching vector for the time step; and generating an answer score that indicates how well the input text passage answers a question posed by the input question string.