Signal, Sentiments and Intentions
Mariana Carrero Rodrigues
Last Update 2달 전
When feedback data is ingested into Birdie, it undergoes an AI-powered enrichment process to extract deeper meaning and structure from unstructured text. This process classifies each feedback entry based on three complementary dimensions:
Signal – whether the feedback contains valuable, actionable content
Sentiment – the emotional tone expressed in the feedback
Intention – the purpose or intent behind what the user is saying
These classifications help filter out irrelevant inputs, highlight meaningful themes, and enable smarter analysis across teams.
Definition
The goal of Signal Detection is to filter out irrelevant or low-quality feedback (referred to as "noise") and retain only the valuable content (referred to as "signal") that contains meaningful information for analysis.
Purpose
This task ensures we focus our analysis only on feedback that contributes actionable insights, improving model performance and dashboard relevance.
Examples
Noise: “Good morning!” (greeting), “#@@#” (nonsense), “Good” (too short)
Signal: “The new feature is useful.”, “I need help to understand this…”
Birdie’s AI identifies and tags sentiments expressed in feedback. Each piece of feedback may include more than one sentiment, and Birdie classifies them accordingly.
| Sentiment | Sentiment Level | Description |
|---|---|---|
| Positive | Only positive | All feedback sentences contain only positive sentiments |
| Mostly positive | More than half of the feedback sentences contain positive sentiment, even if combined with other sentiments. | |
| Any Positive | There is some mention containing positive sentiment in the feedback sentences. | |
| Neutral | Only Neutral | All feedback sentences contain only neutral sentiment. |
| Mostly Neutral | More than half of the feedback sentences contain neutral sentiment, even if combined with other sentiments. | |
| Any Neutral | There is some mention containing neutral sentiment in the feedback sentences. | |
| Negative | Only Negative | All feedback sentences contain only negative sentiment. |
| Mostly Negative | More than half of the feedback sentences contain negative sentiment, even if combined with other sentiments. | |
| Any Negative | There is some mention containing negative sentiment in the feedback sentences. |
Intention classification captures the underlying purpose of what the user is expressing. A single feedback entry may include multiple intentions.
| Intention | Description |
|---|---|
| Compliment | Sentences that express admiration or positive recognition regarding some aspect of the product or service. |
| Problem | Sentences that express a problematic situation, difficulty, or challenge that the feedback author is facing. |
| Question | Sentences that seek specific information or clarifications and are generally formatted as a question. |
| Request | Sentences that express a request, desire, or need for something. |
| Information | Sentences that provide information or clarifications without necessarily expecting a response. Similar to a neutral description of a situation before mentioning a Praise or Problem. |
| Solution | Sentences that propose a response, solution, or method to resolve a Problem or situation presented by the feedback author. |