Translation. Region: Russian Federation –
Source: Moscow Government – Government of Moscow –
The citywide contact center (CCC) has begun using new domestically developed large language models (LLM) in pilot mode. This is a type of artificial intelligence that enables a voice assistant to quickly find the right option in the knowledge base, answer several questions from subscribers at once, and learn independently. Its use will help improve the quality and speed of providing consultations on hotlines, and will also give operators time to resolve more complex requests from residents.
“The use of language models is being piloted on the Moscow Government’s unified helpline for two important tasks: providing consultations on incoming calls and classifying feedback based on service results. In combination, this approach allows not only to train the voice assistant without the help of operators, but also to improve the quality of consultations: the language model classifies residents’ feedback based on call results, which will further help improve the knowledge base and consultation scenarios, and, if necessary, improve the qualifications of operators,” said Andrey Savitsky, head of the citywide contact center.
What are large language models?
Large language models are a type of deep learning based on a neural network with many parameters. The large amount of data allows them to search for answers to several questions at once in a single query.
For example, if a resident calls the line and asks how to obtain a Russian passport and which My Documents offices are near the desired metro station, and also asks to clarify the opening hours of the institution, then a large language model will be able to give a comprehensive answer to all questions at once, providing only the necessary information. In contrast, a regular voice assistant processes only one question, and the rest have to be repeated.
A Million More Calls: How Digital Technologies Make City Hotlines More Accessible to MuscovitesThe citywide contact center’s voice assistant has handled over 135 million calls in 10 years
How does this work
The use of language models on the Moscow Government’s unified information service line is invisible to residents. When a person calls the line and asks a question, the speech recognition system is activated. With its help, the message is recorded, converted from audio format to text and sent to the language model. The latter, in turn, using additional filters, finds the necessary articles with information in the knowledge base, highlights the essence and then generates an answer that the voice assistant tells the subscriber. At the same time, the most accurate comprehensive answer to all the resident’s questions is formed without unnecessary information.
Large language models can independently and in a split second find information on each new topic in the entire contact center knowledge base and voice it to the applicant. Unlike a standard voice assistant, which reads out the answer from the knowledge base, an assistant based on LLM can support a live dialogue, focusing on the intonation and manner of communication of the subscriber.
In addition, a large language model based on a neural network trains the voice assistant. If previously operators had to manually upload new topics and answers in various variations to the virtual assistant, now LLM helps the voice assistant independently find the necessary information in the general knowledge base.
The role of a human is not excluded. Some calls are still processed by operators, and the editors involved in the hotline work monitor the relevance of the information in the knowledge base. Thus, a neural network trains a neural network, but under the strict control of contact center specialists.
Constant learning and love for people: DIT Moscow on the work of city hotline operators
Feedback processing
Another important area for using language models is a new approach to classifying feedback after service. The voice assistant can already collect feedback from residents on contact center lines, but the comments were still processed by contact center specialists. Now, thanks to the implementation of language models, when receiving an assessment from a resident or a comment after a consultation, the neural network ranks them as positive, negative, and neutral, allowing quality control department employees to quickly identify shortcomings and make the consultations provided even faster and more accurate.
The use of large language models on the hotlines of the OKC to work on the quality of consultations complements the already implemented tools based on artificial intelligence. For example, digital audit, which has been operating since 2023, and speech analytics project, launched this year.
The hotline of the unified reference service of the Moscow Government of the citywide contact center has been operating since 2015. It is available at the number: 7 495 777-77-77. Most often, residents call the line to clarify the work schedule and addresses of the My Documents government service centers and city departments, find out information about the issuance and replacement of Russian passports, and contact the portal’s technical support mos.ruand get advice on receiving government services electronically.
The voice assistant has been working on the hotline for eight years. The virtual assistant takes 40 percent of incoming calls on almost 300 topics.
The use of digital technologies and artificial intelligence to improve the quality of life of city residents is in line with the objectives of the national program “Digital Economy of the Russian Federation” and the Moscow regional project “Digital Public Administration”. More information about this and other national projects implemented in Moscow can be found Here.
Help for residents and businesses: Moscow contact center voice assistant received almost 160 million callsHow domestic developments help the work of artificial intelligence in a citywide contact centerThe citywide contact center project received high praise from the jury of the Crystal Headset competition
Please note: This information is raw content directly from the source of the information. It is exactly what the source states and does not reflect the position of MIL-OSI or its clients.
Please note; This information is raw content directly from the information source. It is accurate to what the source is stating and does not reflect the position of MIL-OSI or its clients.
http://vvv.mos.ru/nevs/item/146187073/