Call tagging using machine learning model
Systems and methods disclosed relate to contextually tagging statements associated with calls. In particular, the contextual tagging is directed to training a call tagging model for predicting one or more categories associated with a statement for tagging. The disclosed technology generates training data for training the call tagging model based on a list of known phrases used in contacts in a contextual category and matching phrases and words in the list of known phrases against words and phrases used in statements in sample call transcripts. The call tagging model is fine-tuned by using sample statements that appear in contacts. Once trained, the call tagging model is used to determine a probability distribution of categories associated with statements in a contact and further determine contact-level category distributions using multi-dimensional vectors. The tagged contacts are used to determine contacts that are contextually similar to a given contact.
1. A computer-implemented method, comprising:
retrieving, for training a call tagging model, a set of known phrases associated with a category;
generating a list of phrases for training by expanding the set of known phrases using natural language techniques;
retrieving statement data;
generating a set of training data, wherein the set of training data comprises a first set of statement data with matching known phrases and a second set of statement data without matching known phrases;
training the call tagging model using the set of training data;
fine-tuning the call tagging model using an additional set of training data, wherein the additional set of training data is based on call transcripts associated with contacts in a topic area; and
deploying the call tagging model.
2. The computer-implemented method according to claim 1 , wherein the call tagging model comprises a transformer model.
3. The computer-implemented method according to claim 1 , further comprising:
masking the set of known phrases in the statement data to create masked statement data; and
updating, the set of training data with the masked statement data.
4. The computer-implemented method according to claim 1 , wherein the category represents a contextual category.
5. The computer-implemented method according to claim 1 , wherein the category comprises at least one of:
compliance, escalation, billing, or returns.
6. The computer-implemented method according to claim 1 , wherein the retrieved statement data comprises one or more sentences, and each of the one or more sentences include one or more words.
7. The computer-implemented method according to claim 1 , wherein the call tagging model is a single category model for predicting a single contextual category based on statement data.
8. The computer-implemented method according to claim 1 , wherein the call tagging model is a multi-category model for predicting one or more contextual categories based on statement data.
9. A computer-implemented method, comprising:
receiving a selection of a contact;
retrieving a call transcript data associated with the contact;
determining one or more categories associated with one or more statements in the call transcript data using a trained call tagging model, wherein the trained call tagging model generates a set of multi-dimensional statement-level category vectors;
determining a set of contact-level categories associated with the contact based on the set of multi-dimensional statement-level category vectors; and
determining another contact that is contextually similar to the selected contact.
10. The computer-implemented method according to claim 9 , wherein determining another contact that is contextually similar further comprises determining contextual similarity based on cosine similarity.
11. The computer-implemented method according to claim 9 , wherein the call tagging model comprises a transformer model.
12. The computer-implemented method according to claim 9 , wherein a category of the one or more categories represents a contextual category.
13. The computer-implemented method according to claim 9 , wherein the one or more categories comprise at least one of:
compliance, escalation, billing, or returns.
14. The computer-implemented method according to claim 9 , wherein the call tagging model is a single category model for predicting a single contextual category based on a statement data.
15. A system comprising a processor configured to execute a method comprising:
receiving a selection of a contact;
retrieving a call transcript data associated with the contact;
determining one or more categories associated with one or more statements in the call transcript data using a trained call tagging model, wherein the trained call tagging model generates a set of multi-dimensional statement-level category vectors;
determining a set of contact-level categories associated with the contact based on the set of multi-dimensional statement-level category vectors;
determining a second contact that is contextually similar to the selected contact; and
training, using the determined set of contact-level categories and data associated with the second contact as training data, the trained call tagging model.
16. The system according to claim 15 , wherein the call tagging model comprises a transformer model.
17. The system according to claim 15 , wherein a category of the one or more categories represents a contextual category.
18. The system according to claim 17 , wherein the category of the one or more categories comprises at least one of:
escalation, termination, compliance, customer, complaints, billing, or returns.
19. The system according to claim 15 , wherein the call tagging model is a single category model for predicting a single contextual category based on a statement data.
20. The system according to claim 15 , wherein the trained call tagging model is a transformer model.