Smart Personalized Customer Services by Federated Machine Learning and Evidential Reasoning

Background

According to the latest analysis by the International Air Transport Association (IATA), oil prices will be increasing, and global traffic growth is expected to slow down in the near future. Airline companies also face other severe challenges, such as security, environmental impacts, political unrest, tough competition from other types of transport (e.g. high-speed trains) and increasingly demanding customers. These trends will make it harder for the industry to maintain profitability. Airlines can no longer rely solely on ticket sales but must offer more personalised experiences to their customers. This requires analysis of big data that are often owned by different organisations and are unlikely to be shared among airline companies.

Project Description

Federated machine learning (FML) methods provide a means to build models of data that are distributed across multiple parties. In this project, we will explore their use in building conversion estimation models and improving the accuracy of personalised advertising strategies. The project will also use the evidential reasoning (ER) theory, which is rooted in probabilistic inference, to combine different estimations generated from various data sources for robust prediction and decision-making.

This project will combine FML and ER to optimise personalized marketing strategies for airlines to achieve customer “pull” and “drainage”. Estimation models will be built using FML methods, which will be applied to improve the effect of advertising strategies and accomplish the task of “recruiting” customers. Estimations generated by FML methods will then be combined with ER to analyse the values of customers and classify them accordingly. For different types of customers, the characteristics and needs of different customer groups will be analysed by focusing on the portrait description of these groups. The project will also investigate the efficacy of federated recommendation algorithms and tailor-made personalized service schemes. The following research questions are indicative and could be refined once the project starts:

  1. How can big data owned by different parties be analysed under the FML framework based on the ER theory, in particular likelihood data analysis for evidence acquisition?
  2. Is it possible to estimate advertising conversion rates using FML methods and provide decision support on how to adjust advertising strategies?
  3. Can a customer lifetime value (CLV) model be developed using multi-source data, and the ER theory be applied to analyse uncertainty and integrate FML methods for better CLV prediction?
  4. Is it feasible and beneficial to establish a FML-ER based framework to recommend personalised services and products to different customers?

Summary

This project involves research across a diverse mixture of topics including customer relationship management, revenue management, personalized customer services, multi-source big data analysis, federated machine learning, evidential reasoning and decision support. It will suit an ambitious student with an interest in multi-disciplinary work on challenging problems in service industries.

Project reference: MN41

Deadline: 14th April