Message Classification Model
Updated
The Message Classification Model identifies and categorizes phrases within case conversations according to predefined quality parameters. Based on the set scoring logic, quality scores are then populated on the platform.
Supported Languages
The model supports English, French, Spanish, German, Arabic, and Hinglish languages for both digital and voice cases, and is applicable for global parameters only.
Standard QM Parameters: (OOTB model)
L1 Category | L2 Category |
Opening Quality | Greeting, Agent Introduction, Opening with Brand Mention |
Closing Quality | Gratitude, Further Assistance, Feedback, Closing with Brand Mention |
Attitude | Courtesy, Empathy, Accountability, Promptness, Lack of Clarity |
Active Listening | Active Listening, Efficiency |
Communication Skills | Profanity |
The classification model operates at the phrase level. For instance, when assessing an agent's performance in the Courtesy category, the pre-trained model scans for phrases like "Thank you for your time" or "Thank you for your patience, have a good day" and categorizes them accordingly. The scoring logic is configured to assign a score of 100 if messages are detected for positive sentiment categories.
Additionally, we have the capability to train the classification model for custom categories, provided it is feasible. This process involves collecting and cleaning data, annotating the data for the specific custom category, and training the model with the compiled data.
Example of Classification Model Scoring on Sprinklr Platform
In the image provided, the Quality Management (QM) model has identified one of the agent messages in the conversation as depicting Courtesy. The relevant messages are highlighted accordingly. Since Courtesy is considered a positive parameter, the quality score assigned for this specific parameter is 100.
