Page 126 - AI for Good-Innovate for Impact Final Report 2024

AI for Good-Innovate for Impact

27�2�2 Future work:

Create new variations/extensions to the same use case

If the Metro Intelligent Customer Service Center case receives bonuses and resource support,
the intelligent customer service system will focus on the following aspects in its future work
plan to improve user experience and service efficiency.

Firstly, we will use the intelligent customer service system as the foundation to empower on-site
equipment at the station and provide proactive services for passengers.

Secondly, the system will combine generative large language models and multimodal
technology to provide passengers with more intelligent and diverse voice inquiry services.

Then, the system realizes cross channel integration, supports multiple channels such as App,
applet, official account, and establishes online and offline full scene intelligent customer service.

Finally, we will take remote customer service assistance as the main focus, strengthen
automated process processing, improve service efficiency, reduce on-site manual customer
service intervention, and gradually achieve unmanned stations.

In terms of technological research and development, we will continue to learn and improve,
continuously optimize algorithm models through data analysis and user feedback, adapt to
multiple languages and cultures, customize personalized solutions, and explore applications
in more scenarios.

Through the implementation of the above plan, the intelligent customer service system will
develop towards intelligence, emotion, personalization, and universality, in order to provide
a better customer service experience and promote the sustainable development of the
transportation industry.

27�3� Use case requirements

• UC36-REQ-01: Domain - Rail Transit Domain.
• UC36-REQ-02: Problem being addressed: excavate and enrich the corpus data in
the rail transit industry, construct a Large Language Model of Rail Transit Domain and
continuously optimize it, study the integration of Large Language Model technology with
existing small models to avoid AI hallucination issues in the public service field, deploy
the large model in a lightweight manner.
• UC36-REQ-03: Key solution - processing of multi-source heterogeneous data, fine-tuning
of large models, text classification and retrieval, integration of large models with industry
knowledge graphs.
• UC36-REQ-04: Technology keywords - Large Language Models, Fine-tuning, Lightweight,
Knowledge Graphs.
• UC36-REQ-05: Data available - private data.
• UC36-REQ-06: GPU - Nvidia A100 graphics cards.
• UC36-REQ-07: Metadata - The core algorithms of this project are focused in the field
of NLP (Natural Language Processing), hence the training data is predominantly textual,
although it also involves voice data. However, ultimately, the voice data is transformed
into text data through ASR (Automatic Speech Recognition) algorithms for NLP model
training and optimization.

110