Listen to this story
Consumer Demand is evolving. Earlier, while it was all about preferences, it was not all about “experiences on time”. Planning and forecasting what consumers need and want have been in place for a long time and several advances in digital, data sciences and applied AI have been made to solve problems in silos. Based on the problems over the years in terms of consulting for clients, the challenge to build a robust model continues to persist and the oft-asked question is what next and what more?
There is no ‘one size fits all’ and hence ‘plug-n-play’ algorithms or platforms fail to produce results as desired since they all try to ‘fit model to solution’ rather than ‘being data-centric’. In the below article, a blended method is introduced, which will ‘learn with time and domain experts’ as a system of evolutionary learning which is the need of the hour.
“If things are not so good, you may want to imagine something better” —Prof John F Nash
Introducing: Federated Knowledge Graph Learning System for Consumer Sciences
The Federated Knowledge Graph Learning System (FKGLS) is a new approach to demand planning and forecasting that leverages the power of knowledge graphs in near real-time. FKGLS uses a federated architecture, which allows it to learn from multiple data sources in a distributed manner.
This approach is designed to be highly explainable so that users can understand why the system made certain predictions. This is achieved through the use of knowledge graphs, which provide a clear representation of the relationships between different data elements. The system also provides users with tools for manipulating the knowledge graph so that they can investigate how changes to data inputs would impact projections.
The relationship between a brand and a consumer together needs to be collaborative; hence both parties need to be understanding.
How does the Federated Knowledge Graph Learning System Work
The Federated Knowledge Graph Learning System (FKGLS) is a machine learning system that employs a federated approach to demand planning and forecasting. The system is designed to provide explainable results by incorporating domain knowledge into the learning process.
The FKGLS system consists of five main components:
- A central server that hosts the federated knowledge graph.
- A set of distributed machines that each contain a local copy of the federated knowledge graph.
- A set of algorithms using Graph Sciences—currently an evolutionary mechanism—are executed on distributed machines in order to learn about the demand for products and services.
- The MLOps architecture for Graph Sciences is used to scale the Data as a Graph Model to improve the pipeline, including monitoring of the performance of data vis-à-vis performance of the models in the ‘best-fit model’ approach.
- A Visual Interface validates the predictions and tracks the decisions.
The FKGLS system operates as follows:
- The central server receives data from various sources, such as sensors, social media, and financial data providers.
- This data is used to update the federated knowledge graph hosted on the central server.
- The updated federated knowledge graph is then pushed out to the distributed machines.
- The algorithms contained in the distributed machines use this updated information to learn about demand patterns and future demands.
- These predictions are then fed back to the central server, which can use them to make decisions about pricing, production, and inventory management.
Benefits of the Federated Knowledge Graph Learning System
Federated learning is a new way of training machine learning models that can offer many benefits over the traditional approach. One of the key advantages of federated learning is that it allows training on data that is distributed across different devices or locations. This can be extremely beneficial when data is siloed within an organisation, as it can allow for models to be trained on a much larger dataset than would otherwise be possible.
Another advantage of federated learning is that it can help improve privacy and security by keeping data stored locally on devices rather than centrally on servers. This means that sensitive data does not need to be shared with third-party organisations in order to train models, which can help keep it safe from potential breaches.
Finally, federated learning can also be used to train models on edge devices, such as smartphones. This can be extremely beneficial for applications that need to run in real time, as the model can be updated very quickly without needing to send data back to a central server.
The FKGLS has been used in a number of different applications, including:
- Predicting consumer behaviour under changing macroeconomic landscape for Sales Boosting: The FKGLS can be used to predict the probability of how consumers will behave in the future and personalise by using macro-economic and extrinsic factors. This will not only tell what to sell but will tell “what not to sell given current conditions in near real-time”. This information can be used to make better decisions about product development, marketing and sales strategies.The FKGLS is a versatile tool that has a wide range of potential applications. It is an important tool for anyone who wants to make better decisions based on data.
- Elevating User Experience in Retail: We have data being considered from inventory, e-commerce websites, shipping info, logistics and geospatial insights. A knowledge graph-based retail e-commerce system can turn these siloes into an interconnected environment of communication platforms and transaction flows. Were it powered by a knowledge graph, the system interacted with—be it a chatbot or a website—would be providing real-time data with all the information needed.
The Federated Knowledge Graph Learning System improves scalability and privacy compared to centralised systems. It also increases the accuracy and robustness of knowledge graph representations through the combination of data and knowledge from multiple sources. It is still a relatively new and evolving area of research and there is much that remains to be explored and understood about its potential benefits and limitations. Nonetheless, it provides valuable insights into this important and rapidly-growing field and helps to advance understanding of how to effectively learn and use knowledge graphs in a decentralised and collaborative manner.
This article is written by a member of the AIM Leaders Council. AIM Leaders Council is an invitation-only forum of senior executives in the Data Science and Analytics industry. To check if you are eligible for a membership, please fill out the form here