Introduction

Prerequisites: This guide assumes that you have an account with Search. You should also familiarize yourself with the Search Backend documentation and The Search Algorithm, for more context into how the search configuration impacts the results that are returned by the algorithm.

This guide will walk through how to (and how not to) optimally set up the configuration of Yext Search for healthcare applications such as Find-a-Doctor search functionality.

For each step, you will typically be working in and navigating between the Yext Knowledge Graph Configuration, Entities, and Search Configuration screens.

Step 1: Set Up the Healthcare Taxonomy

Why this step matters: When using Yext Search for healthcare-related search experiences like Find-a-Doctors, the search configuration benefits greatly from having healthcare-specific terms within the Yext Knowledge Graph. Ideally, these terms connect clinical terminology like "nephrolithiasis" to nonclinical terminology like "kidney stones". Installing a Healthcare Taxonomy (either Yext's or your own) will drastically improve the quality of your healthcare search experience.

Knowledge Graph

Configure a linked system that identifies a provider's specialties, conditions treated, procedures performed, and subspecialties. For convenience, Yext provides two options:

1. Configurable Yext-Provided Taxonomy (Recommended)

• Uses Yext Knowledge Graph capacity
• Provides a set of entities with medical terms
• To use: Consult this Install Healthcare Taxonomy Solution App doc

2. Configurable Client-Provided Taxonomy

• Uses Yext Knowledge Graph capacity.
• Clients will upload their own entities for medical terms (e.g. specialties, conditions, etc) via a file upload. You will have to add fields, link entities, and generally configure their own business logic so that the correct entities are returned for relevant searches.
• When implementing, we recommend using linked entities so that Find-a-Doc experiences can take advantage of Search features like entity relationships to connect doctors to the conditions they treat, procedures they conduct, etc., which may often appear in search queries.
• The most important reminder when implementing a custom taxonomy is to employ clear business logic so that the correct entities are returned for relevant searches.

Search Configuration

• For each vertical, apply Phrase Match to all relevant fields that contain healthcare terms. This could include fields from the Yext Healthcare Taxonomy or custom fields containing organization-specific healthcare terms. If you find that this configuration setting is too restrictive (i.e., it is not including desired results in the results set), then consider expanding the result set by adding keyword search in conjunction with Phrase Match to desired fields of the Healthcare Taxonomy.

Common Mistakes

• Relying on Synonym Sets as the primary source of healthcare-related taxonomy. Synonym Sets are not intended to replace a fully built Healthcare Taxonomy. They are much more tedious to manage and don't take advantage of the benefits of the Yext Knowledge Graph and the Search algorithms!
• Using keyword search on Healthcare Taxonomy searchable fields as this can prove to generate search results with too much "noise" due to matches on every token (e.g., a query for "back surgery" would provide matches for any provider that has any medical terms including "back" and "surgery").

Step 2: Choose Filter Search or Free-Form Search

Why this step matters: For Healthcare, Filter Search generally offers more predictable results and a higher-quality search experience for end-users than Free-Form Search. The abundance of overlapping terms in healthcare taxonomies (e.g., "cardiologist" and "pediatric cardiologist") makes inferring filters with Inferred challenging. Free-Form Search will be a preferred option if you are building a universal search bar where users can search for entities beyond Providers & Facilities all within a single search bar.

Choose Your Method

1. Filter Search (Only available for Search experiences with front-ends built in React)

• Users must pick a filter value from a list of options (i.e., they are not allowed to search with free-form text!)
• Prompts are directly powered by field values in the Yext Knowledge Graph (you get to decide which fields are used for the search bar; you can use one or multiple)
• Users don't apply keyword search, semantic search, Inferred Filter, synonyms, etc. to searchable fields
• Search results are directly filtered based on the selections made in the filter search bar

2. Free-Form Search

• Users can search using any free-form text query
• Prompts can be hardcoded, "madlib" from certain fields in the Yext Knowledge Graph, or popular queries
• Results are returned using the full combination of search algorithms (depending on how you've set up your search configuration)
• Filters are inferred from the query using Inferred Filter (when applicable)

Setup - Filter Search

React: Add the Filter Search React Component which enables your users to search from a list of values coming directly from the Knowledge Graph, and filter only to the results that fit the selected criteria.

Setup - Free-Form Search

Search Configuration:

• Configure fields with the Inferred Filter search that have a finite number of shared variations across all providers and are structured as list or option fields (e.g., use Inferred Filter on an "Insurances Accepted" field that lists out the different insurance options).
• Define an inferred filter order to prioritize the order in which Inferred Filters should trigger.
• Configure fields that use Inferred Filter search to also have Facets to allow the user to reject the Inferred Filter for a more expansive result set on their own.

Common Mistakes - Free-Form Search

• Configuring Inferred Filter on any people name fields to avoid results that are too exclusionary.
• Configuring Inferred Filter on fields that contain long tail text (4-6 words or longer).
• Configuring Inferred Filter on fields that have a low completion rate (~<90% completion rate is a good rule of thumb) as that could cause providers to be filtered out inadvertently (e.g., if only 50% of providers have a "Secondary Specialty" field filled out due to missing data, providers may inadvertently be filtered out when an Inferred Filter is triggered on this field. Consider using Phrase Match in this situation). To derive completion rate:
  • Identify the total number of entities within a vertical or entity type using the Yext Knowledge Graph Advanced Filters functionality.
  • Identify the number of entities with the field filled out by applying the "Field Has Data" filter on the Field in question.
  • Divide the number of entities that have the field filled out by the total number of entities in the Advanced Filter.
• Configuring Inferred Filter on multiple fields that may have overlapping information (e.g., if a field "Keywords" and a field "Conditions Treated" includes overlapping information, it is not recommended to configure both with Inferred Filters).
• For the Yext Configurable Healthcare Taxonomy, do not enable Inferred Filter on anything other than specialty fields.

Step 3: Review Professionals' Names

Why this step matters: It is common for consumers in the Healthcare industry to search by provider names, and we often see consumers enter partial names, typos within names, nicknames, or even outdated maiden names. It is also common for providers to search themselves on their Healthcare System's Search and expect to come up first as a result when searched directly. For this reason, it is important to configure Search to be performant with these possible variations but also highly rank when a user conducts a full name search.

Knowledge Graph

• Ensure your professionals' names are current, accurate, and reflect the full name the professional expects to be discovered by in your Find-a-Doctor search experience.
• Include any nicknames or maiden names in separate custom fields.

Search Configuration

• Configure each name field (e.g., Name, First Name, Last Name, Nickname, Maiden Name, etc.) with a combination of Keyword Search and Phrase Match. We recommend using Keyword Search and Phrase Match for each name field because they are more inclusive of all terms within a user query.

Common Mistakes

• Using Inferred Filter on any people name fields. This is a mistake because this filter is typically too exclusionary and could inadvertently filter out the intended individual (e.g., a provider may have recently changed their name from "Jane Smith" to "Jane Roberts" and a query for "Jane Smith" could trigger an Inferred Filter for providers with the last name "Smith", filtering out "Jane Roberts").

Step 4: Add Location Search

Why this step matters: Consumers often search for providers in their area and are willing to go a small distance away to find care. It is important to configure Search in a way that returns providers not only in the exact geographical area, but also returns results nearby.

Search Configuration

• Enable builtin.location as Inferred Filter to handle Location type searches.
• If your health system serves a specific geographic area, consider implementing a Bounding Box that includes the system's region and surrounding areas. This will ensure that the Search Algorithm is most accurate for the geographical area your Healthcare System serves.

Common Mistakes

• Configuring additional Address (e.g., address.city, address.line1) or Geomodifier fields as Inferred Filter. If you would like to leverage those fields consider using Keyword Search and Phrase Match.

Step 5: Add Synonyms

Why this step matters: Users in the healthcare space can search in a variety of different ways, and will typically use "layman" or generalized versions of the proper terminology. It is important to configure Yext Search using best practices that account for this user behavior and create an efficient experience.

Knowledge Graph

• For the Yext Configurable Taxonomy, include Synonyms within the Synonym fields in the entity's profile. If additional synonyms are needed than the ones provided, add them to the fields on the appropriate Healthcare Taxonomy entity.

Search Configuration

• For the Yext Configurable Taxonomy, handle Synonyms within the Synonym fields in the entities of the Taxonomy by making them searchable with Phrase Match, not within the Synonyms feature of the Search Configuration.

Common Mistakes

• Using synonyms, especially Synonym Sets, within the search configuration (instead of the entity's profile in the Knowledge Graph) as a way to solve for a lack of Healthcare specific taxonomy.

Step 6: Add Custom Phrases

Why this step matters: Healthcare systems can have many multi-word concepts that should be treated as a singular concept, such as facility names and certain medical terminology (e.g., we want a search for "Smith Sleep Center" to avoid matching with "Jane Smith" or "Cardiovascular Center"). Yext Search allows healthcare systems to configure these brand-specific Custom Phrases to ensure the most relevant results are returned for your organization.

Search Configuration

• Configure Custom Phrases for multiple-word medical terms that should be treated as one singular concept (e.g., Sleep Medicine, Physical Therapy, Breast Cancer, etc.).
• If you are operating an experience that includes Facility and Professionals verticals, and the facility names include one or more medical terms (e.g., Springfield Cardiology Clinic, Smallville Pediatric, and Sleep Center, etc.), configure the entirety of the facility names as Custom Phrases to ensure searches which include the facility name do not trigger unintended matches on the medical terms.

Step 7: Add User Interface Elements

Why this step matters: It is critical for the user's experience that the interface of the search experience displays data points that match the expected results for queries. For this reason, it is important that there is a correlation between the fields indexed via the Search Configuration and the fields displayed in the front-end Search experience.

Search Configuration

• Configure fields you display on the experience's user interface as searchable in the Search Configuration.
• Configure Facets on fields in which inferred Filters are used to allow a user to deselect that filter if they wish to expand their search via the User Interface.

Common Mistakes

• Configuring inferred Filters or Facets on fields that may contain data that may differ from the front end (e.g., if a provider has a "Primary Specialty" field that is "Hip Replacement" displayed on the User Interface, but an "All Specialties" field configured with inferred Filter which includes "Hip Injuries" and "Shoulder Issues" a user might be confused as to why a provider with a card that states "Hip Replacement" is showing up for their shoulder related search).

Step 8: Configure Sort Order

Why this step matters: It is important to configure the Search Algorithm to return results in a manner that heavily weighs relevance to the query, while also taking into consideration distance.

Search Configuration

• Rely on "Relevance" as the first sort order, followed by "Entity Distance" to ensure medical term searches provide the most accurate results. Read more on how to set entity sorting here.
• Randomize the list of relevant professionals that return for a query by adding "Random" to the end of the sort order (following "relevance" and any other criteria). This will be helpful in cases where you have multiple doctors with identical specialties working at or near the same location.