Client Success & Data Science: Turning query exhaust into categorization fuel

Policy Topics Vary Infinitely: Client Success built a finite dataset

The FiscalNote Client Success team takes an approach where for each of our clients who joins, we learn their specific issues during on-boarding and hand-craft their queries for them.
We have clients from all over the globe and representing all sorts of areas of policy focus. Not only that, but they represent areas of policy focus at varying levels of depth—in one case the client might be interested in the Infrastructure Spending issue, and in another, in Bridge Repair Labor Employment Contracts. We meet the client at what level of depth and create search queries to target bills of that nature with a high relevancy ratio. We often spend a few hours on each on-boarding.

Custom configurations leave a query exhaust trail that we can turn into fuel

Turning Data Exhaust into Categorization Fuel

The FN Data Science team has a goal of building a universal Topic Hierarchy that can apply to all policy documents of all types—US state and federal legislation and regulation and their global equivalents. This Topic Hierarchy will lead to better features for our end-users in the platform, such as filtering by these topics, and will also allow us to make recommendations of items that may be relevant to them that a search query alone did not capture.

The Methodology

The Data Science team took the topics classification schemes from multiple sources including those used by the US Congress and European Parliament as well as academic projects such as the Comparative Agendas Project to create some top level topic groups. We then also looked at the various ways in which our clients organized their data into FiscalNote Issues and Labels. We then combined some topic modeling techniques and manual QA tasks to develop a hierarchy of issues each with their own set of terms that help us unique identify those topics. Our hierarchy has over 35+ ‘top’ level topics and goes up to 6 levels deep with a total of 1,800 distinct topics with different levels of granularity. This has allowed the Data Science team to analyze our data at different levels of specificity and picking up on general trends of data over time. By ‘embedding’ all of our data into this topics ‘space’ we can build models that allow us to easily compare and recommend relevant data.

The Virtuous Cycle

Now that this Topic Hierarchy is created, Data Science has also been able to use the ‘terms’ associated with each topic to generate search queries associated with each. When a new client joins, we can reference these topics to help the client get a better sense of what information they need for an effective configuration, or if they aren’t sure yet, we can use one of the Topic Hierarchy queries to start them out. We’re able to get our users a much faster Time to Value and much more efficiently, as a result.