Autonomous Support Case Creation From Limited Case Histories
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated technical support systems face challenges in efficiently addressing issues for products with limited volumes of technical support cases, as fine-tuning Large Language Models (LLMs) with a large corpus of case information is not feasible, leading to inefficiencies and the need for significant human oversight.
Innovation Solution
A method utilizing a semantic analysis model trained on a limited volume of existing cases to identify related technical support issues, creating an initial communication template and question-answer pairings, enabling automated resolution without human intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If fine-tuning LLM with large corpus of case information is used to improve automated support accuracy, then support quality improves, but it becomes infeasible for products with limited case volumes and requires significant human oversight
Solution Approach 1:
The patent segments the training process into two distinct phases: unsupervised pre-training on available case data to learn basic patterns, followed by targeted supervised fine-tuning only on identified common issues. This segmentation allows the system to achieve competent performance with limited data without requiring exhaustive fine-tuning across all possible cases.
Solution Approach 2:
Instead of performing complete fine-tuning on the entire case corpus (excessive action), the system applies partial action by selectively fine-tuning only on the subset of common issues identified through clustering analysis. This partial approach reduces computational requirements and data needs while maintaining effectiveness for the most frequent problems.
2Reliability
If comprehensive fine-tuning is applied to improve automated case handling, then resolution accuracy improves, but human oversight requirements increase significantly
Solution Approach 1:
The system performs preliminary unsupervised pre-training on all available case data before deployment, establishing a baseline level of automated capability. This preliminary action reduces the need for ongoing human oversight by pre-learning patterns from the case corpus, allowing the system to handle routine cases autonomously.
Solution Approach 2:
The system implements feedback mechanisms where automated resolution outcomes are continuously monitored and used to identify common issues that require additional fine-tuning. This feedback loop enables the system to progressively improve accuracy while maintaining automation, as human reviewers only need to validate and annotate cases that fall into identified common patterns rather than reviewing all cases.
3Loss of time
If automated case handling is implemented without sufficient training data, then deployment speed improves, but support quality deteriorates
Solution Approach 1:
The system performs unsupervised pre-training on all available case data as a preliminary step before deployment, establishing baseline automated capabilities without requiring extensive fine-tuning. This preliminary action enables faster deployment while maintaining acceptable support quality for non-common cases, as the model learns general patterns from the available data corpus.
Solution Approach 2:
The system changes the training parameter from requiring large volumes of finely-tuned data to accepting smaller datasets with unsupervised pre-training followed by selective fine-tuning on common issues. This parameter change allows deployment with limited data while maintaining quality through the two-phase training approach, balancing speed and effectiveness.
Data Source
AI summary
A technical support system identifies a plurality of technical support cases that were previously resolved and identifies related technical support cases of the plurality of technical support cases based on problem descriptions and resolution summaries associated with the plurality of technical support cases. A Large Language Model (LLM) is used to create an initial contact template for potential technical support issues associated with the related technical support cases. Question and answer pairings associated with the potential technical support issues are identified based on communications in the related technical support cases. A machine learning model is trained to communicate with potential users associated with the potential technical support issues based on the question and answer pairings. The initial contact template and the machine learning model are used to resolve a subsequently received technical support issue associated with the related technical support cases.