Bidirectional Probability Matching for Contextual Reply Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current human-computer interaction systems using sequence-to-sequence models for chatterbots fail to effectively consider the linguistic context and identity of respondents, resulting in poor intelligence and relevance of reply statements.
Innovation Solution
A method and apparatus that determine reply statements by matching statements to a preset lexicon using a neural network, calculating matching probabilities, and iteratively adding words to form potential reply statements until a terminator is reached, while considering both input and output probabilities to improve contextual relevance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If a sequence-to-sequence model is used to determine reply statements, then the system can automatically generate replies, but the intelligence and contextual relevance of the replies deteriorates
Solution Approach 1:
The patent applies feedback by calculating bidirectional matching probabilities: not only the probability of the reply being generated from the input statement, but also the probability of the input statement being generated from the reply. This feedback mechanism ensures that selected replies are contextually appropriate by verifying they could plausibly generate the original input, thereby improving reliability while maintaining automation
Solution Approach 2:
The patent changes the selection parameter from single-direction probability to a combination of bidirectional probabilities (first matching probability + second matching probability). This parameter transformation allows the system to evaluate replies based on mutual contextual consistency rather than simple generation likelihood, resolving the contradiction between automation and contextual relevance
2Reliability
If bidirectional matching probabilities are calculated for all potential reply statements, then contextual relevance improves, but computational complexity increases
Solution Approach 1:
The patent performs bidirectional matching probability calculations selectively rather than exhaustively for all possible replies. By calculating probabilities only for potential reply statements generated from the input statement, and then filtering based on these calculations, the system achieves sufficient contextual relevance without the excessive computational burden of evaluating all possible statement pairs
Solution Approach 2:
The patent segments the reply selection process into distinct stages: first generating potential replies from the input statement, then calculating matching probabilities for these candidates, and finally selecting the best match. This segmentation allows computational resources to be focused on evaluating only relevant candidates rather than performing exhaustive comparisons
Data Source
AI summary
Aspects of the present disclosure provide a method and an apparatus for determining a reply to a statement. The apparatus includes processing circuitry determining, based on a preset lexicon, potential reply statements in response to a statement, and first matching probabilities respectively corresponding to the potential reply statements. A first matching probability indicates a probability of the corresponding potential reply statement being output in response to the statement according to the preset lexicon. The processing circuitry also obtains second matching probabilities respectively corresponding to the potential reply statements. A second matching probability indicates a probability of words in the statement being output in response to the corresponding potential reply statement according to the preset lexicon. According to a combination of the first matching probabilities and the second matching probabilities, the processing circuitry selects one of the potential reply statements as a target reply statement.