Neural Network Term Matching for Question Answering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing question answering systems rely on exact matches of surface forms for term matching, failing to utilize similarity scores based on vector representations, which limits their ability to capture contextual information and relevance.
Innovation Solution
A method that generates similarity scores by combining vector representations of question and passage terms using an artificial neural network, trained to identify similar dimensions, allowing for relevance determination and answer generation based on these scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If exact matches of surface forms are used for term matching, then the system is simple to implement, but the ability to capture contextual information and relevance is limited
Solution Approach 1:
The patent transforms term matching from exact surface form comparison to vector representation comparison, changing the parameter space from discrete string matching to continuous vector space similarity. This allows the system to capture contextual information and semantic relevance by computing similarity scores between vector representations of question terms and passage terms, thereby improving measurement precision without excessive complexity increase
Solution Approach 2:
The patent replaces the mechanical string matching mechanism with a neural network-based vector similarity computation system. Instead of direct character-by-character comparison, the system uses trained neural networks to generate and compare vector representations, substituting a more sophisticated computational approach that better captures semantic meaning
2Reliability
If vector representations with similarity scores are used for term matching, then contextual information and relevance are better captured, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary action by pre-training neural networks to generate vector representations of terms before the actual question answering task. These pre-computed vector representations and their similarity scores are then reused during term matching, reducing the computational burden during runtime while maintaining high reliability in relevance determination
Solution Approach 2:
The patent introduces vector representations as an intermediary between raw text terms and relevance determination. Instead of directly comparing text strings or making binary match decisions, the system uses vector representations as a mediating layer that captures semantic information, enabling more accurate relevance assessment through similarity score computation
Data Source
AI summary
Software that generates an answer to an input question using a source document by performing the following operations: (i) receiving a question; (ii) generating a plurality of vectors including a first vector representation of a term in the question and a second vector representation of a term in a source document; (iii) providing each dimension of each of the first vector representation and the second vector representation into a respective input node of an artificial neural network; (iv) determining whether the source document is relevant to answering the question based, at least in part, on an output generated by the artificial neural network; and (v) in response to determining that the source document is relevant, generating an answer to the question utilizing the source document.