If You Work In These 8 Settings, You Need To Know About Knowledge Graphs
Knowledge graphs are starting to gain a bit of buzz. In 2020, Gartner placed knowledge graphs at the peak of their hype cycle. And while jumping on every hyped-up tech should be avoided, knowledge as a service industries have learnt a thing or two from DAAS and IAAS. Namely, rather than building a monolithic data structure, knowledge graphs should be built into knowledge workflows helping to provide data and context for particular roles in particular industries. The automation of tedium and knowledge gathering for specific individuals is a good enough goal. And could collectively change our lives at work.
Full disclosure: I work at Diffbot, developer of the world’s largest Knowledge Graph. With that said, this has afforded me with an inside view into innovative knowledge graph uses across a range of industries that I share below.
Not sure what a knowledge graph is? Knowledge graphs are like relationship-first databases. They’re comprised of nodes, often called entities, and edges (basically relationships). Depending on the type of entity, different types of facts and relationships may be attached. For example, an article entity may have topic tags, sentiment, an author, and a publisher. Meanwhile an organization entity may have employees, subsidiaries, industries, news mentions, and more. This interconnectedness between entity types allows for a rich queryable landscape of firmographic, demographic, product and news centered information.
Knowledge graphs may be sourced by internal data. But the largest knowledge graphs utilize external data sources like news or — in the case of Diffbot — the entire public web.
Using domain expertise (for smaller KGs) or NLP, machine vision, and AI (for larger knowledge graphs), knowledge graphs turn unstructured data into information + context (knowledge).
8 Work Settings Where KGs are Killing It
Ecommerce is aided by knowledge graphs in two ways. When you think of ecommerce, unstructured data may not come to mind. There are APIs galore for information from major sellers. If you’re an ecommerce brand, you should have a plethora of signals about what individuals do on your own site. But ecommerce isn’t only about internal data. Knowledge graphs help remedy this.
Historically disparate webs of data can pulled into context with a KG. This includes data on public perception (review data and sentiment), brand mentions, news monitoring and market intelligence on competitors, and global pricing and availability data.
A second way in which knowledge-centered workflows are killing it in ecommerce comes through the form of structuring specific unstructured data stores online. For example, you may want a nearly live web extraction of 50 competing domains. Or you may want to structure the data on your own website to determine duplicate products, fraudulent products, or pull in review data in a usable contextually-linked format.
Finance is a field where we’re starting to see knowledge graphs become more prevalent. There are a few reasons for this. The structured nature of organization entities allows for finance firms and investors to query rather than “search” public web data. For knowledge graphs with a wide range of facets, data provenance, and long tail organizational coverage this is a game changer.
The fact that knowledge graphs can interlink multiple data types can allow queries like “show me all companies in X industry with >Y market caps that have offices in city Z.” Or, “show me all home contractors in Canada with less than 100 employees who are growing quickly.” Or, “show me a break down of the skills of Ford’s employees… then compare that to Tesla’s.”
A second use case is in the form of news data that’s been pre-tagged as a given topic or with a certain sentiment and attached to an organization or individual. Whether in manual market research or news mentions that are consumed programmatically, there are no news feeds as comprehensive as the largest knowledge graphs.
Sales are a great setting for impactful data enrichment and exploratory analysis built directly into a workflow. We routinely see knowledge graph data used for two purposes within sales. First, the workflow of looking up an organization (or set of orgs) and then mining in to determine if they may be a good fit for outreach. While much of the information sales professionals need is publicly available on the web, doing this manually isn’t scalable. Knowledge graphs with long tail structured organization data can enable you to quickly survey entire industries or locations and then mine into individual org profiles.
For example, using a knowledge graph you could gain a summary view of the job titles within an organization. You could look at skill sets or a tech stack. You could also look at the velocity of funding rounds or positive news mentions or many other signals. Again, while you can perform these tasks manually, parsing through unstructured data by yourself. Decreased tedium and increased prospecting speed leaves sales teams able to focus on what they should be focusing on.
Secondly, once you already have individuals in your CRM, data enrichment via a knowledge graph can quickly flesh out your profile of individuals including skillsets, education, news mentions, and contact information. Revitalizing CRM data over time can save countless manual hours, false leads, duplicates, and provide re-actionable info.
Public Relations professionals have a range of media monitoring tools to choose from. But most news aggregation or tracking tools don’t go beyond a string match when linking organizations or individuals to a mention. Knowledge graphs employing NLP or machine vision can programmatically pair organizations or people to news mentions. In the case of Diffbot or GDELT’s knowledge graphs, news monitoring data also cover articles in every major language. This avoids the language-by-language silo imposed by most search engines. Additionally, previously unstructured articles in structured data format allow for parsing of sentiment, speakers, and the pulling out of important facts or entities.
Marketing use cases of knowledge graphs fit somewhere between sales and public relations. Account based marketing pairs well with rich firmographic data. Knowledge graphs that can provide industry codes, access to key events in the development of organizations, and links to employee profiles can lead to quality segmentation, personalization, and validation that you’re talking to the right audience.
Within Martech, growing nocode or lowcode movements enable marketers to connect knowledge graph data directly to the tools they use every day for outreach, automated emails, and ad spend.
Data Science has a long history of structuring public web data (through web scraping) for a wide range of applications. One way to think of web sourced knowledge graphs that provide information via API is that a knowledge graph can basically be an API for every site that doesn’t have one.
Knowledge graphs can be particularly useful for data science teams due to their scale and the structure of data. After all, a majority of data science lifecycles are comprised of simply finding and wrangling data. Additionally, while can knowledge graphs provide some additional processing on top of unstructured data (say, adding sentiment scores to articles), they can also provide data provenance and a glimpse at the source for a fact. This can be great for mining in to interesting trends or “sanity checking” your analysis.
Manufacturing relies on a complex set of regulatory, supply chain, and talent sourcing issues that can all be aided with knowledge graph data. Use cases we’ve seen for knowledge graphs related to manufacturing include the monitoring of long tail regulatory bodies for announcements, as well as the monitoring of news mentions related to key material production or supply chain issues. Pairing sentiment with articles around materials, markets, or competitors can cast a wide net for many potential issues within manufacturing.
For knowledge graphs that contain geopolitical events, organizations, and or individuals, linked data between these types can be used to keep tabs on competition or partners.
Human resources can utilize knowledge graphs in much the same way as sales can, to locate key individuals they may try to recruit. Additionally, staying on top of the distribution of skills in competing industries may help HR professionals to preempt talent shortages. For knowledge graphs that contain educational data on individuals, cohorts from prestigious programs can be tracked and new potential talent pipelines can be identified.
