Neo4j, the leader in graph technology, announced the availability of Neo4j for Graph Data Science, the first data science environment built to harness the predictive power of relationships for enterprise deployments.

The unpredictability of the current economic climate underscores the need for organizations to get more value out of existing datasets, continually improve predictive accuracy and meet rapidly changing business requirements. Neo4j for Graph Data Science helps data scientists leverage highly predictive, yet largely underutilized relationships and network structures to answer unwieldy problems. Examples include user disambiguation across multiple platforms and contact points, identifying early interventions for complicated patient journeys and predicting fraud through sequences of seemingly innocuous behavior.

Neo4j for Graph Data Science combines a native graph analytics workspace and graph database with scalable graph algorithms and graph visualization for a reliable, easy-to-use experience. This framework enables data scientists to confidently operationalize better analytics and machine learning models that infer behavior based on connected data and network structures.

Alicia Frame, Lead Product Manager and Data Scientist at Neo4j, explained why Neo4j for Graph Data Science is the most expeditious way to generate better predictions.

“A common misconception in data science is that more data increases accuracy and reduces false positives,” explained Frame. “In reality, many data science models overlook the most predictive elements within data – the connections and structures that lie within. Neo4j for Graph Data Science was conceived for this purpose – to improve the predictive accuracy of machine learning, or answer previously unanswerable analytics questions, using the relationships inherent within existing data.”

Take fraud analysis, such as detecting identity fraud and fraud rings, as an example that spans areas from financial services and insurance to the government sector and tax evasion. Even the smallest predictive improvement translates into millions of dollars of savings. Neo4j for Graph Data Science makes it easier to make those incremental improvements without altering existing machine learning pipelines. Below are some simple steps illustrating how Neo4j for Graph Data Science fits into a fraud prediction workflow:

1. A data scientist can reveal suspicious groups of transactions using community detection algorithms, like Connected Components, to analyze behavior.
2. They can then dive deeper by applying graph algorithms such as Betweenness Centrality or PageRank to uncover hidden structures such as accounts with unusual influence over the flow of money or information.
3. An analyst could explore these clusters in an intuitive way and collaborate with fraud experts using Neo4j Bloom to infer which elements (i.e., features) are most likely predictive of criminal behavior.
4. They can perform “what if” analyses or even chain “recipes” of graph algorithms together with a mutable in-memory workspace where their graphs are reshaped on-the-fly.
5. Once the algorithmic recipes have been validated and understood, they can be used for machine learning models that are operationalized to proactively prevent – and not merely detect – fraud.

Neo4j for Graph Data Science enables data scientists to answer questions that are only addressable through understanding relationships and data structures. Graph algorithms are a subset of data science tools that capitalize on network structure to infer meaning and make predictions such as:

• Cluster and neighbor identification through community detection and similarity algorithms
• Influencer identification through centrality algorithms
• Topological pattern matching through pathfinding and link prediction algorithms

With Neo4j for Graph Data Science, teams confidently deploy a proven solution at massive scale to run optimized graph algorithms over tens of billions of nodes with production features such as deterministic seeding, which provides starter values and consistent results for reproducible machine learning workflows. Through intelligent integration of network analytics and a database, Neo4j automates data transformations so users get maximum compute performance for analytics and native graph storage for persistence.

Ben Squire, Senior Data Scientist at Meredith Corporation, a leading media and marketing services company with publications reaching 190 million unduplicated American consumers every month, including nearly 95 percent of U.S. women, across broadcast television, print, digital, mobile, voice and video, shared his experience with Neo4j for Graph Data Science.

“Providing relevant content to online users, even those who don’t authenticate, is essential to our business,” said Squire. “We use the graph algorithms in Neo4j to transform billions of page views into millions of pseudonymous identifiers with rich browsing profiles. Instead of ‘advertising in the dark’, we now better understand our customers which translates into significant revenue gains and better-served consumers.”

Dr. Alexander Jarasch, the Head of Data and Knowledge Management at the German Center for Diabetes Research (DZD) and collaborator on COVIDgraph.org, explained how Neo4j for Graph Data Science offers an intuitive data science experience with logical parameters and Neo4j Bloom for comprehensive graph exploration.

“Nothing is more pressing today than understanding COVID-19,” said Jarasch. “Graphs give us the ability to bring together the salient information around this confounding disease and provide a synthesized view across heterogeneous data. Today’s understanding of this coronavirus is severely hampered by minimal peer-reviewed research and the absence of long-term clinical trials. Neo4j for Graph Data Science will help us to identify where we need to direct biomedical research, resources, and efforts.”

The post Introducing Neo4j for Graph Data Science, the First Enterprise Graph Framework for Data Scientists appeared first on Artificial Intelligence.

Big data exploration usually starts at a high level of data abstraction, and then gradually plumbs into the depths of the data as companies learn more from it.

The approach has worked well, and is operative in many different types of applications.

For instance, GIS and mapping systems use data to visualize a big picture map and then to focus in on a specific point or location. As the data analyst drills down to this location, they can then look at other related data that might be appended to the location such as the demographics of individuals who live at that location, or the number of traffic accidents at that location.

However, there is also another ground up approach that has the ability to unlock hidden values of big data. This approach actually starts at the lowest level of the data and then works its way up to more sophisticated data structures to deliver data insights that are helpful to management and staff.

Here is an example:

“A single pixel display can reveal the visible color of a point, but also the infrared value, which can be used to measure vegetative health,” said Layton Hobbs, research and development director and vice president at Woolpert, an architecture, engineering and geospatial solutions firm.

Hobbs is talking about the potential of agriculture and forestry companies to go beyond basic geospatial data that they collect and unlock hidden treasures that are buried in geospatial data such as data on topography, soil, ground cover, plant health, and tree canopies.

“Most geospatial data is created for one specific reason or need, but there is so much more information in geospatial data that is underutilized or not recognized,” added Woolpert’s associate and geospatial discipline leader, Joe Cantz. “Particularly with the newer technologies, the data-rich information is growing exponentially, but we are using only a small percentage at this point.”

According to Woolpert officials, geospatial data pixels are capable of storing a much wider range of values than the traditional 256 values of an 8-bit image. “These modern systems often store four bands of data (red, green, blue and infrared) at up to 12 bits or around 4,000 values for each band,” said Hobbs. “Combining those four bands for image interpretation creates 256 trillion possible combinations at one spatial location! This is definitely overkill for most applications but shows the potential for big-data applications of imagery.”

Why does this matter for company big data projects?

IoT data, such as data captured and emitted by sensors, immediately comes to mind.

With IoT, you can start with your own top-down big data initiatives and analytics when it comes to utilizing data and imagery that gets sent from sensors on board drones—but what if you looked into each individual pixel of data that the drone was sending back—and discovered that there was additional data value captured that could answer questions that you weren’t interested in today, but could be in the future?

Here’s how you can optimize data for both current and future use:

Analyze what is possible to extract from a given unit of data (e.g., a pixel), even though you may not care about all of this information today.

This can be easily done. Referencing Layton Hobbs’ example, maybe you don’t care about the health of the forest floor today, but if you one day want to restore this forest after a harvest, understanding something about forest health will help. At that point, knowing everything you can obtain from your big data under management becomes significant.

Catalog the information capture that is possible at the lowest unit of big data.

If you are dealing with a pixel and you know that forest health and topography is possible to analyze from this data and you make a record of it, it is much easier to remember the information potential of your data and to activate it if and when you need to.

Don’t lose yourself in the details

While it is important to catalogue the information potential of your big data at the lowest level of data, it is equally important not to lose yourself in the details. If your job today is simply to map a forest and to identify stands of harvestable timber, stick with that. Don’t get off course with other types of data explorations that aren’t relevant to the task at hand.

Anticipating lessons learned

When I was running a marketing department for a bank, we used demographics for one of our checking campaigns by identifying persons in certain locations by age group, and then linking checking products to the various life cycle stages that customers were in. Later, we wanted to improve results, and we added occupation as well as age for targeting our checking products.

This is a common scenario for companies. They want to go back to the data to see if they can add more information so they can improve results.

By assessing and cataloguing the potential information yield of big data at the lowest level of the data, data analysts can be poised to open up the data to more comprehensive analytics that can unlock the answers to questions that the company will want to ask next.

The post How to optimize your company’s big data for future use appeared first on Artificial Intelligence.