Description:
in order to automate the process of constructing dialogue scenarios between call center operators and company clients, and also subsequently automatically evaluate the quality of work of contact center operators, it is necessary to categorize a set of dialogues between operators and clients, highlight the main topics, and also recognize sequences of topics in the dialogues of operators with subscribers . It is planned to use text segmentation into semantically related pieces using topic modeling models.
Context:
There are a large number of agents working in a contact center, it is necessary to understand the quality of their work, since their efficiency and sales effectiveness directly depends on their adherence to the script. Manual validation is a rather long and resource-intensive process; it is necessary to automate it and propose a quality assessment methodology.
Solution:
Based on the replicas inside the CC dialogues, many topics were built. A model was built that decomposes all new dialogue into replicas, and labels the replicas with appropriate topics. Then a “map” of dialogues with the highest quality sales was built - an ideal operator script. For new dialogues, the sequence of remarks was checked against an ideal map and the “quality” of the dialogue was measured.
Results:
To build the model we used:
Simulation results:
Customer: Finance, Banking
Technology stack: Topic modeling, Syntactic connections, Neural network models of text segmentation, BigARTM, PyTorch, gensim, nltk, Python.
in order to automate the process of constructing dialogue scenarios between call center operators and company clients, and also subsequently automatically evaluate the quality of work of contact center operators, it is necessary to categorize a set of dialogues between operators and clients, highlight the main topics, and also recognize sequences of topics in the dialogues of operators with subscribers . It is planned to use text segmentation into semantically related pieces using topic modeling models.
Context:
There are a large number of agents working in a contact center, it is necessary to understand the quality of their work, since their efficiency and sales effectiveness directly depends on their adherence to the script. Manual validation is a rather long and resource-intensive process; it is necessary to automate it and propose a quality assessment methodology.
Solution:
Based on the replicas inside the CC dialogues, many topics were built. A model was built that decomposes all new dialogue into replicas, and labels the replicas with appropriate topics. Then a “map” of dialogues with the highest quality sales was built - an ideal operator script. For new dialogues, the sequence of remarks was checked against an ideal map and the “quality” of the dialogue was measured.
Results:
- 41 topics were identified;
- Topic identification quality: 75% accuracy on average;
- Integration into the Bank's BI for assessing the quality of operators;
- Increasing conversion and the percentage of successful dialogues.
To build the model we used:
- Dialogues between call center operators and clients;
- Information about the success of the dialogue;
- A priori expert knowledge about topics;
- Assessor markup.
Simulation results:
- A set of topics for dialogue between operators and clients;
- Transition graph between topics for successful and unsuccessful dialogues;
- A tool for thematic segmentation of dialogue lines.
Customer: Finance, Banking
Technology stack: Topic modeling, Syntactic connections, Neural network models of text segmentation, BigARTM, PyTorch, gensim, nltk, Python.
