Chatbot Context Ellipsis Recovery via Candidate Text Buffer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for performing context completion in chatbots are limited by their inability to efficiently update candidate texts for context ellipsis and struggle with domain transfer, requiring extensive labeled training data and being time-consuming and costly.
Innovation Solution
An end-to-end architecture for context completion in chatbots that automatically determines context ellipsis and domain retention, continuously updates candidate texts, and selects complementary texts using CNN-based and regression-based ranking models, allowing for efficient domain transfer without the need for extensive retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing techniques are used for context completion, then the chatbot can handle context ellipsis, but the system requires extensive labeled training data and is time-consuming and costly to update and transfer domains
Solution Approach 1:
The system performs preliminary actions by maintaining a candidate text buffer that is continuously updated with texts from the current domain. This buffer is prepared in advance and can be quickly utilized when context completion is needed, eliminating the need for time-consuming retraining when transferring domains. The candidate text buffer acts as a pre-computed resource that adapts to domain changes automatically.
Solution Approach 2:
The system creates a copy of relevant domain information in the form of candidate texts stored in a buffer. Instead of retraining the entire model when domains change, the system copies and updates the candidate text buffer with new domain-specific texts, enabling fast domain transfer without losing the original model structure or requiring extensive labeled data.
2Reliability
If existing techniques are used for context completion, then the chatbot can recover context ellipsis, but the process is costly and time-consuming to update candidate texts
Solution Approach 1:
The system implements continuous update of the candidate text buffer through a sliding window mechanism that continuously incorporates new texts from the current domain while maintaining a rolling buffer of candidate texts. This continuous update process ensures the system maintains high reliability for context completion while improving productivity by automating the update process and eliminating batch retraining operations.
Solution Approach 2:
The candidate text buffer is designed as a dynamic structure that automatically adapts to domain changes. The buffer size, update frequency, and text selection criteria can be adjusted dynamically based on the current domain context. This dynamic approach allows the system to maintain high accuracy while efficiently adapting to different domains without fixed, rigid update schedules.
3Measurement precision
If extensive labeled training data is used, then the context completion model achieves high accuracy, but the training process becomes time-consuming and costly
Solution Approach 1:
The system extracts only the essential elements needed for context completion from large amounts of domain text, storing them as candidate texts in a buffer. Instead of using extensive labeled training data for every domain, the system extracts and stores unstructured candidate texts that can be quickly utilized, significantly reducing the need for voluminous labeled training data while maintaining high accuracy.
Solution Approach 2:
The system changes the parameter of training data requirements by shifting from needing extensive labeled data to using a compact candidate text buffer. The buffer stores pre-processed candidate texts that serve as the training resource, changing the data volume parameter from large-scale labeled datasets to a manageable buffer that can be quickly updated with unstructured text from the current domain.
Data Source
AI summary
Methods and apparatuses for performing context completion to messages in a session are provided in the present disclosure. A message may be obtained. It may be detected that there exists context ellipsis in the message. It may be determined whether the message is retained in the current domain of the session. In response to determining that the message is retained in the current domain, a complementary text for recovering the context ellipsis may be selected in the current domain. A completed message may be generated based on the message and the complementary text.