Graph Data Science 101: Ten Tips with Resources for Successful Graph Data Science

5 min readNov 29, 2020

If you’re wondering if your project is “graphy” and how to get started with graph data science (GDS), this chapter can help. We give you some Neo4j resources to guide you to more information, and to help you explore your project’s opportunity and successfully move forward from concepts to production, we include these ten tips:

Investigate use cases and get comfortable with concepts.

Because graph technology is applied across industries and in various use cases, it can be hard to know where to start. To expand your knowledge and help you get comfortable with GDS, review these examples:

Review use cases. Get up to speed on the problems graph technology can solve. Visit neo4j.com/use-cases to read some use cases.
Watch talks. Find out how people use GDS. Watch presentations from Neo4j’s Connections for GDS digital event: go.neo4j.com/connections-graph-data- science-lp.html.
Expand your knowledge of key concepts. Review material that sets GDS in a larger context. Visit neo4j. com/whitepapers/artificial-intelligence-graph- technology for how graphs enhance AI.

Identify and engage a spearhead team.

Using graph technology in production can be new to many people, so don’t expect teams to understand how to evaluate or compare graph options to other solutions. Assemble a small team that can become your experts in translating business needs into technical requirements and the application of GDS. Make sure to have representation from key organizations, including business, IT, and data science teams.

Provide your developers and data scientists with more technical information. Your team will likely need time to familiarize itself with the technology so look for resources that allow an easy start. Some examples include

neo4j.com/graph-algorithms-book
neo4j.com/graph-databases-book
neo4j.com/graph-databases-for-dummies
neo4j.com/sandbox

Evaluate your “graphy” problem.

Graph technology is useful anywhere you have a lot of connected, interdependent information. But at some point, you need to look into what areas of your business to focus on and what kind of project to start with.

Start with an intersection of ideas between users, business, and technology. Consider hosting offsite or virtual innovation sessions with your cross-functional team to define your stakeholders’ needs, create connection-related questions, story-board possible solutions, and identify key challenges and opportunities. This collaboration may naturally lead to a prototype that you can share with executives for feedback, but the goal is to uncover promising target use cases.

Assess the current state.

After you have a target use case in mind, start with documenting your current state. Consider existing problems as well as how the various parts of your organization will have different experiences and issues. Find out how your business sponsors view this use case and any problems or opportunities. Be as specific as you can. For example, what is the impact per customer of improved online profiles? What’s the revenue implication of a half percent increase in recovered fraud? Also remember to consider external market factors such as customer or transaction growth, competitive factors, emerging opportunities such as new delivery platforms, or productization opportunities.

Map the value of the proposed state.

Although your first graph project may spawn many new ideas and future projects, make clear and direct mapping of features of the near-term graph project to business values. Consider the current state and pain points and how your graph target use case can help with business concerns such as cost savings, increased revenue, new market opportunities, time to market, risk mitigation, and the like. For example, uncovering similar customer journeys and using that information in a machine learning (ML) model may increase the accuracy of churn prediction so the business could take early preventative action and reduce revenue loss.

Measure ROI.

For each of your value areas, determine how you plan to measure your return on investment (ROI) or success. For example, will you use predictive accuracy or reduced financial loss to estimate the impact of your end state? Compare the soft and hard costs of maintaining existing processes to your graph project. If you’re unable to audit your existing state, be more conservative when estimating incremental saving or revenue opportunities. Likewise, it may be difficult to measure the value of net-new capabilities, such as answering previously intractable questions, so you may need to get creative or add qualitative analysis.

Align stakeholders.

Eventually, you need cross-functional agreement on the goals and requirements of your graph project. This process is iterative, not something you tackle at one point in time. Different teams may have alternative views on the project vision, key ROI, and even the role of graph technology. Getting alignment on the goals of the project and how success is measured are essential — and you may want to consider a process for dealing with conflict or dissenting opinions.

Get your project approved.

Taking advantage of new technologies like GDS requires your stakeholders and approvers to be comfortable trying something unfamiliar, so your work to target the right use case, map values, and estimate ROI needs to come together in a concise story that aligns with your company’s motivations.

Document stakeholder assumptions about business value. For example, customer churn may be an issue, but is it a priority, and why? You may be asked about the competitive landscape as well as alternatives and the costs or lost opportunity if you don’t proceed. Clearly document the interdependent system touchpoints that are part of current processes and the impact of your graph solution.

Conduct a POC and plan for production.

Larger projects, especially if the technology is new to a team, often require a proof of concept (POC) before approval and deployment. A POC can prepare your team for production and identify any gaps. This process may involve iterating on previous prototypes before you move into data modeling and testing specific workflows.

In GDS, your data model and algorithm choices are highly dependent on the questions you’re trying to answer. Your data scientists and subject matter experts should be involved to ensure the right assumptions are made. Also make sure that your IT teams are involved to raise any red flags and that your end-users are on hand to evaluate any usability concerns.

Vendors that provide POC services can help accelerate your project with their graph experience. Visit neo4j.com/ professional-services for more info.

Get connected and continue your journey.

Applying GDS is a journey. You may start with one focused project and find yourself answering questions you never knew you had. We highly recommend your team connect and engage with the graph community. Graph communities consist of active groups of users that share new ideas and help with specific, and sometimes unusual, questions. Getting involved in a rich active community with educational support and certifications helps your team be successful with its first graph project and expands the value of your graphs over time.

Graph Data Science 101: Ten Tips with Resources for Successful Graph Data Science

Read More…

Written by Tech First