| # Custom BERT Model for Intent Recognition | |
| This repository contains a custom fine-tuned BERT model for intent recognition. The model was trained to recognize a set of customer service-related intents, and it's based on the pre-trained BERT architecture (uncased_L-12_H-768_A-12). | |
| ## Python Version | |
| This project is compatible with **Python 3.7.4**. It is recommended to use this version for compatibility with the listed dependencies. | |
| ## Model Information | |
| - **Base Architecture**: BERT (uncased_L-12_H-768_A-12) | |
| - **Max Sequence Length**: 200 | |
| - **Number of Intents**: 15 | |
| ## Classes | |
| The model is trained to classify the following customer service-related intents: | |
| don't change the order while intializing | |
| 1. `service_availability_check` | |
| 2. `billing_inquiry` | |
| 3. `order_cancellation` | |
| 4. `address_verification` | |
| 5. `user_authentication` | |
| 6. `account_information_update` | |
| 7. `call_divert` | |
| 8. `customer_service_escalation` | |
| 9. `appointment_scheduling` | |
| 10. `order_status_inquiry` | |
| 11. `product_information_request` | |
| 12. `complaint_registration` | |
| 13. `call_disconnect` | |
| 14. `appointment_confirmation` | |
| 15. `appointment_cancellation` | |
| ## How to Use | |
| To use the model, load the configuration file (bert_config.json), the checkpoint files (bert_model.ckpt*), and the vocabulary file (vocab.txt). Along with these, load the saved fine-tuned model or weights (if you plan to modify layers or change the max_seq_len [the length of input sentences]). This ensures that the model is correctly configured and functions as expected for your custom use case. | |
| ## Intended Use | |
| This model is designed for intent recognition in customer service applications and supports a variety of queries such as billing inquiries, order cancellations, service availability checks, and more. | |
| . | |