Dialog Generation Using Latent Vectors to Suppress High-Frequency Replies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current dialog generation systems face challenges in maintaining high accuracy, particularly in multi-round dialog scenarios where information compression leads to low-quality replies and a high generation probability of high-frequency answers.
Innovation Solution
A method involving converting each term in a query sentence into word vectors, calculating positive and negative latent vectors, determining content topics, and using these vectors to generate replies, effectively suppressing high-frequency responses by incorporating topic information and bidirectional attention mechanisms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If contexts are directly spliced into a long query sentence in single-round modeling, then the system can process multi-round dialogs, but information compression chaos occurs causing low quality of generated reply sentences
Solution Approach 1:
The patent segments the long query sentence into multiple short sentences, where each sentence corresponds to one round of dialog. This segmentation prevents information compression chaos by processing each dialog round separately while maintaining the ability to handle multi-round dialogs through sequential processing of segmented sentences.
2Adaptability or versatility
If multi-round query and reply transfer process is modeled, then the system can handle context, but the model tends to generate high-frequency answers with low accuracy
Solution Approach 1:
The patent extracts and removes high-frequency answer patterns from the training data before model training. This extraction process eliminates the tendency to generate repetitive high-frequency answers, allowing the model to learn more diverse and accurate reply patterns while still handling dialog context effectively.
3Measurement precision
If rule-based dialog system is used, then structure is simple and accuracy is high, but generalization capability is poor
Solution Approach 1:
The patent merges rule-based dialog processing with generation-type dialog modeling. The system uses rules for structured processing to maintain accuracy while incorporating generative modeling capabilities to improve generalization, creating a hybrid approach that combines the strengths of both methods.
4Reliability
If searching-based dialog system is used, then recall can be improved, but a relatively large number of high quality corpuses are required
Solution Approach 1:
The patent performs preliminary processing of the corpus by segmenting dialog contexts into individual sentences and pre-processing training data to remove high-frequency patterns. This preliminary action reduces the complexity and requirements for the corpus, allowing the system to achieve good recall performance with less data while maintaining processing efficiency.
Data Source
AI summary
A dialog generation method, apparatus, and device, and a storage medium are provided. The method includes converting each term in a Kth round of a query sentence into a first word vector, and calculating a positive latent vector and a negative latent vector of each term according to the first word vector, where K denotes a positive integer greater than or equal to 2. A content topic of the Kth round of the query sentence is obtained, and converted into a second word vector. An initial latent vector output for the Kth round of the query sentence is determined according to the second word vector, the positive latent vector of a last term in the Kth round of the query sentence, a latent vector of a last term in a (K-1)th round of a reply sentence output for a (K-1)th round of the query sentence, and an initial latent vector of the (K-1)th round of the reply sentence output for the (K-1)th round of the query sentence. A reply sentence for the Kth round of the query sentence is generated according to the positive latent vector and the negative latent vector of each term in the Kth round of the query sentence and the initial latent vector output for the Kth round of the query sentence.