Integrating Siloed Personal Knowledge Graphs

Critical steps in improving the productivity of enterprise note-taking

Figure 1: A dataflow diagram for integrating personal knowledge graph applications such as Roam and Obsidian. Image by the author.

In prior blogs, we covered the key concepts in Personal Knowledge Graphs and introduced the Integrated Personal Knowledge Graph (IPKG). This blog will look at the steps to build the integrated PKG system in a large organization. We will review each step and see how they can be part of a holistic enterprise knowledge management strategy.

The IPKG Hypothesis

The basis for much of the logic around these Integrated Personal Knowledge Graph (IPKG) efforts is the following hypothesis:

Integrated Personal Knowledge Graphs will allow users of PKGs to create more valuable and helpful knowledge than if note-taking is done with siloed tools.

The key to this hypothesis is to quantify what valuable and helpful means in metrics that even a finance person with no formal strategic knowledge management background can understand.

Align Your IPKG Benefits with Organization Strategy

There will undoubtedly be many individuals that will be ultra-enthusiastic about finally having an excellent note-taking tool. Integrating a siloed PKG with companies’ complete entity-type systems has many benefits, and our guess is your users will love the experience. However, companies rarely spend millions of dollars only to make their employees happy. Employee happiness is a pleasant side effect but should never be considered a primary driver for corporate IPKG initiatives.

Here are just a few of the benefits IPKGs can bring to an organization

1. sharing knowledge
2. avoiding duplication of knowledge
3. making knowledge consistent
4. improving collaboration between teams within a department
5. improving collaboration between departments or business units
6. streamlining communication
7. increasing knowledge worker productivity
8. improving search results and search ranking

The key is not just to state these benefits but to specifically tie these objectives to your organization’s annual business strategy and priorities.

Find and Educate an Executive Sponsor

My experience is that most successful knowledge management and Integrated PKG projects are driven by empowered staff with formal training in enterprise knowledge management strategies. Not all organizations have these teams, or if they do have them, they don’t have the enterprise mandate or sponsorship of an influential executive. To be successful, you will need long-term sponsorship and funding. Integration takes time and money. You will need to find staff with the right skills and the ability to get access to essential data services. Your business alignment document will help you convince executives that your Integrated PKG project will help them achieve their strategic objectives.

Define Clear Metrics for Success

One of the critical challenges of integrating organization knowledge with any text editor is to agree on standard measures of productivity improvement. Let’s take a simple example: adding a list of industry-specific terms or company product names to an individual spell checker.

We all know how annoying it is to “train” every new text editor you use to recognize company-specific terms for the correct spelling. The methods for modifying the dictionaries are often hidden or unavailable to the user. For example, the chat feature on your Zoom or Teams program might not allow you to add custom terms. Some people with dyslexia or where English is not their native language depend on good spelling checks. They might be reluctant to engage in any chat when they are not confident the tools will help them look professional.

For example, you open the “compose” window in your email editor. You start writing and using terms specific to your industry and organization are all marked as misspellings. These words can be the names of products you use or products you sell. For each word, you need to find a way to add this term to your dictionary. But what if the text editor could automatically be updated with words that your peers have already confirmed are correct spellings or valid acronyms for systems you work with daily?

Example: Metrics for Customise Spell Checking

The key is to capture the small productivity gains if your word processor is smart enough to gather terms from the IPKGs of your peers. It takes only a few seconds to add a word to the dictionary of a local word processor. But if you have thousands of terms and each of your 10,000 employees needs to add these words, then lost productivity is measured. You will need to estimate the cost/hour of your knowledge workers and use the best estimates of the increased productivity without the distractions of your text editor constantly telling you there are terms that it needs to understand. These are the types of metrics that can help you build a business case for IPKGs.

We can also extend this checking to include the correct use of the case in an official term. For example, we use “Facebook” but not “FaceBook” and “LinkedIn,” not “Linkedin.”

Example: Automatic HyperLinking

Let’s take this one step further. What if the first time we typed in a new technical term in a document, a wiki, or PKG editor, the text editor would automatically add a link to your company or department business glossary? Then your new employees could quickly jump to the right place to find that term and understand the definition of that term and related terms.

An excellent example is an ability to reference individuals directly on social media sites such as Facebook, Twitter, and LinkedIn. A confirmation list is presented by referencing a person starting with an “@” character. After you click on that item on the list, a link to that person’s home page is automatically inserted into the text.

The key to making these features useful is to limit the scope of these selection lists. You might only refer to people in your social network and rank-order the list based on criteria such as if they are on the current discussion or how often you have mentioned them in the past. The scope and sorting rules can be contextual and complex. The scope and sorting rules illustrate the effort required to build scalable IPKG autosuggest algorithms.

Integrating simple productivity features into any note-taking tool and tracking the measurable productivity gains is often the first step in your IPKG journey.

Once you have done this alignment and got a sign-off from an executive sponsor, you are ready to create a formal PKG integration plan.

Create an Inventory of Current Knowlege Repositories

One of the critical challenges of building good IPKGs is understanding where related information and knowledge exist within your organization today. Once we have an inventory of these knowledge stores, we can ask what autosuggest-related services could be designed using existing APIs. You will also need to estimate how frequently a note-taker will want to reference these concepts in their documents and prioritize the most commonly used terms.

Here is an example of some of the knowledge sources within a large organization:

1. Approved business terms from a centralized glossary of terms
2. List of industry-specific words or phrases from medical or legal dictionaries
3. Names of wiki pages that are the most frequently referenced in other wiki pages or tagged with a specific label
4. List of products and product attributes in a product catalog.
5. List of the approved desktop applications or tools that employees are using
6. List of computer systems or databases
7. List of employee names (both first and last names), including familiar person names from other countries and languages used in your organization

In summary, anytime your users add new terms to their local spell checkers, you can monitor them and look for trends.

Keeping terms up to date and checking quality levels high is key to building trust in your IPKG initiatives. Add a human-centric quality-checking process whenever you see quality issues enter a system. Sometimes common misspellings can creep into a list of approved terms. A subject-matter expert should review new terms. Ideally, any incorrect business terms are removed before other teams use them. Make sure that users can quickly provide feedback to the team if they dispute the correctness of a word or phrase.

Building a Reference Architecture Diagram

After your IPKG team starts to inventory knowledge sources, you should create a single-page “map” showing how knowledge flows into your IPKG system and its new services. These data flow diagrams start with a list of input sources on the left, your IPKG in the center, and consumers of this data on the right side. The top of this blog has an example of this diagram.

We must also keep track of the events that occur within each of the individual PKG editors related to the automatic completion and acceptance of suggestions. Any suggestions that fall below a 30% level should be carefully monitored. Users who find most automatic suggestions “unhelpful” will disable autosuggest features.

Monitoring You Progress

The IPKG reference architecture diagram includes a monitoring dashboard that accesses the list of terms and the user event log files. This dashboard pulls in data from multiple sources and allows different users to build a customizable single-page view of your progress.

A quality editing and review tool could involve tracking key metrics such as adoption rates, user satisfaction, and productivity improvements. Adjustments can be made to the implementation plan as needed to ensure the success of the note-taking tool in the organization.

Monitoring Broken and Obsolete Links

In addition to monitoring autocompletion events, the quality dashboards might also attempt to track when documents’ links point to concepts that are no longer relevant. For example, you might have documents describing an old computer system decommissioned last year. In many cases, links to decommissioned systems are of interest to historians and auditors and should remain the same. In other cases, links to decommissioned resources need to be updated with links to the systems that have replaced them.

Analysis of broken links is a dependency graph analysis and update problem. It is a task suited to enterprise knowledge graphs supporting two-way link analysis. The key is to decide what broken links need to be updated, how the updates should be done, and who will monitor the overall link quality.

Remember that documents that are accessed frequently have more value to an organization. Updating broken links on documents that have not been referenced in years adds little business value. Knowing how often a document is read is another reason that document access event logs be part of the event log stream and that current documents without broken links be ranked higher than documents that contain obsolete and incorrect knowledge.

Conclusion

Many word processing and editing tools today allow you to easily add custom dictionaries for basic tasks such as spell checking that leverage your organization’s shared knowledge. However, a new generation of personal note-taking tools, such as Roam Research, and Obsidian, allows personal note-taking autocomplete functions to reach far beyond simple term lists.

As the APIs for these tools improve, many organizations will move from a limited-scope collection of personal silos of knowledge to a proper network of integrated personal knowledge graphs that are consistent and complement enterprise knowledge repositories. Content from these IPKGs can be shared with other groups and be conditionally added to centralized stores of knowledge. Feedback systems can ensure those subject matter experts know that users depend on these sources and will be more motivated to keep them up-to-date.