Chat-Bot Development
Building a system for automatic processing of incoming calls to the contact center assumes the existence of a given taxonomy, which will be categorized treatment with its subsequent processing. In collaboration with a large number of contact centers, a system for rapid analysis of dialogues is necessary, followed by the construction of a category taxonomy for a specific contact center.
Input Data
- Dialogues multiple contact centers
- Assessors markup of paraphrase for each dataset
Output Data
- Model soft hierarchical clustering dialogs
- Summary taxonomy of categories with description
- Customizable server application with the API Interface
Main Results
Integration into the BI department for the automation of client contact centers:
- Reducing the burden on analytics on the allocation of classes of automation 80%
Taxonomy harmonization:
- Reducing time to identify a new category by 60%
- The selection of new intents in the stream of calls with a quality of 70%
Selection of entities and analysis of interpretability:
- Selecting named entities and filling in a client card is 10% more accurate
- Increasing the proportion of interpreted intentions relative to the old model by 40%
Quality soft hierarchical clustering
| Quality criterion | Upper | Lower |
|---|---|---|
| Accuracy | 83 % | 70 % |
| Precision | 83 % | 78 % |
| Recall | 87% | 73 % |
| F1-score | 79 % | 71 % |
The final quality of interpretability and selection of entities:
| Quality criterion | Quality |
|---|---|
| The proportion of interpreted topics | 81 % |
| Quality of selection of named entities | 83 % |
