Insights from our round table ‘The Power of Graphs’
This week, DataSeries, an OpenOcean led initiative hosted another Virtual Roundtable about “The Power of Graphs” together with Jennifer Schenker who published an article about this here:
How graphs will transform data management and business
INSIGHTS GATHERED
CHALLENGES:
Complexity of connections — Why knowledge graphs?
If you have a dataset that is very much biased towards the connection of the entities, more so than the entities themselves, then relational databases are not the best solution. Knowledge graphs apply semantics to give context and relationships to data, providing a framework for data integration, unification, analytics and sharing. Essentially enabling the discovery of new facts and relationships between people, processes, applications and data, in ways that give companies new insights into their businesses, create new services and improve R&D research.
Data integration and the challenge of connecting silos
Data silos, when combined with external models for glossaries, entity relationship diagrams, databases and metadata repositories, lead to incongruent data and, due to the explosion of uniquely labeled elements, it is nearly impossible to align these silos. impeding application development, data science, analytics, process automation, reporting and compliance.
Particularly when nuances matter — data integration is challenging. For example linking up a front office and back-office application. How do we connect these two elements especially for major corporations who’s data-management infrastructure is based on decades old technology? By not having the right technological foundation due to legacy systems, corporates end up with data that is hard to access, blend, analyse, automate, impeding application development, data science, analytics, process automation, reporting and compliance. There is often a lack of tooling and skills to effectively manage the data. Most of all, there is a mindset problem in order to change to the right kind of solution for the specific problem.
Graph thinking requires a mindset change
The need of knowledge graphs, unless there is an integrated graph thinking in the firm’s philosophy from day 1, usually just appears once the company is reaching a certain size that comes with complexities in the information system. Most tech giants have integrated a desire and agile methodology in the way they think about data/knowledge management and representation. Tech companies might not have the perfect schema in place, but there are people who are constantly introspecting and generate insights about their own data. For more traditional companies, acquiring this mindset is a true challenge!
New infrastructure and skills needed
Today making the leap from columns to context is achievable but not automatic. Most of the enterprise data is stored in a tables-columns format and it is accessed using the highly common SQL language, while knowledge graphs’ data is stored in a different format and is accessed using less known query languages. This means that in order to benefit from knowledge graphs, organizations must invest in new infrastructure, data must be transformed and new skills must be learned.
Knowledge graphs are still relatively expensive to implement and there is a need for data engineers, domain expertise, modelling knowledge and ontology expertise. Access to external knowledge is often the truly needed domain expertise. Finding a combination of a biologist/chemist with ontology knowledge, for example, is really tough. We have seen a similar issue at our resource optimisation roundtable, where participants were stressing the point of how rare it is to find a software engineer who has also got domain expertise in hardware and vice versa.
OPPORTUNITIES:
An integrated and linked representation of all the data
A shift away from conventional relational databases to knowledge graphs allows corporates to capture the meaning of data as well as how concepts are connected. Semantic modelling eliminates the problem of hard-coded assumptions, because it focuses on concepts, not specific applications. Users automatically understand what the data represents even when it moves across organisational boundaries, allowing efficient reuse across systems and processes. Instead of data silos we get data that is integrated and linked, and organisations become more efficient because ontologies are standardised and reusable.
Towards data federation
The future holds a federated setup for (enterprise) systems that allow us to start joining the graphical structure of data more easily. We will get to a point where the end-user is using tools without even realising that the data is coming from different data sources or even Knowledge Graphs.
Relationships at scale and reducing complexity
Knowledge graphs allow us to deal with high complexity that is not doable in other ways. We need technology that’s really capable to scale into billions or even trillions of relationships. On top of it we have to be able to make sense of it. To put things into perspective, already 2 years ago Microsoft’s knowledge graph had 2 billion entities and 55 billion connections.
User friendly tools for increased adoption by non tech people
While various solutions exist for placing virtual knowledge graphs on top of legacy data management systems, what is missing are user-friendly tools that help non-techies within the business to work with knowledge graphs. If you want to use knowledge graphs to its fullest potential, then your software’s design also has to be adapted accordingly.
Knowledge graphs to empower machine learning
Due to the focus on context, graphs can be extremely powerful for machine learning. It is argued that graphs are a perquisite for achieving smart, semantic AI-powered applications that can help you discover facts from your content, data and organizational knowledge, which would otherwise go unnoticed. They make data understandable in business terms rather than in symbols only understood by a handful of specialized personnel.
A Machine Learning training process is reviving every possible data-experience in order to generalise. If there is a knowledgeable instructor in place, then you will be able to create general rules. Knowledge Graphs are introducing this, but we are still not there. Using a Knowledge Graph as a guide and instructor for a machine learning process is an exciting prospect.
For the field to move further, certain aspects of deep learning for graphs should be prioritized. For instance, the challenge of formalizing different adaptive graph processing models under a unified framework.
Accelerated industry adoption
Sequential data delivery can’t fully satisfy today’s demanding deep learning models — and graphs are emerging as the solution. Graph learning can be advantageously applied in domains such as chemistry and drug design, NLP, spatio-temporal forecasting, security, social networks, recommender systems and much more. Examples include:
- Healthcare: GNNs can play an important role in drug discovery. DeepMind recently announced it can predict the structure of proteins, a breakthrough that could dramatically speed up the discovery of new drugs.
- Supply Chain, Logistics and Travel: A graph approach can enable precise information of deliveries and quality within the chain, or help in making audit of suppliers easy and more transparent for instance. Google maps uses GNNs to more accurately predict travel times.
- E-commerce: Some of the world’s large e-commerce companies are using GNN’s to suggest diversified complimentary purchases.
- Social and entertainment: Social networks create friend graphs while Pinterest is using them to suggest additional relevant content to its users.
- Manufacturing Process Optimization: Describing each physical component and device in one knowledge graph allows a very user friendly and complete process monitoring. Knowledge graphs also help in understanding which processes to optimize, increasing yield and reducing defects in production.
Next to industry use cases — or as an enabling factor for technologies such as virtual assistants, predictive maintenance and many others — knowledge graphs can also be used for all kinds of research, for instance market research. Using the knowledge graph approach offers potential in data integration scenarios for highly dynamic environments where entities and relationships of new types can be added without much additional engineering.
Graph neural networks to empower knowledge graphs
We are starting to build graph neural networks (GNNs) which have deep mathematical relationships to signal processing and convolutions. They are being trained, optimised and used to actually simulate logical observations — to do things like query and logical problem solving on top of the knowledge graph structure.
Combining these two worlds and harvesting the best of both is an exciting prospect. With the introduction of knowledge graphs we are now able to move from the reconciliation process and focus on the meaning of our data. We have achieved a huge leap from stitching data together to being able to have the freedom to understand the deeper meaning behind it.
The real opportunity, it is argued, is to effectively combine rule based and deep learning approaches where we stop having to trade-off between high performing, not very robust deep learning techniques, versus very stable, but brittle, easy to predict rule- based systems.
