How to enrich your data with the knowledge graph?

This article describes how to create a Knowledge Graph and use it to enhanance the quality of search results in your Data Hub. The article covers the following topics:

• Overview of the knowledge graph capabilities
• Building a knowledge graph through the concept map
• Semantic atlas: user-facing interface of the knowledge graph

Overview of the Knowledge Graph capabilities

The platform allows to build an additional metadata layer on the top of your data to enhance its descovery through better recommendations. This layer, which is essentially a Knowledge Graph, represents a mapping between data (time-series) and search terms (concepts). Powered by the Knowledge Graph, search engine is able to generate high-quality instant answers and suggest recommendations as shown below.

The knowledge graph also facilitates the data extraction from unstructured text through the Data Finder for MS Word and Google Docs. An example of unstructured text analysis by the Data Finder for Word is presented below.

Building a knowledge graph through the concept map

The knowledge graph represents a system of interrelated concepts (search terms) organized in groups and mapped to datasets. Each concept in the knowledge graph can also be associated with a number of synonyms. There are global and community-specific knowledge graphs in the platform:

• Global knowledge graph is build on the top of public data and is platform-wide meaning that it is used by search engine in all Data Hubs by default.
• Community-specific knowledge graph can be created by Data Hub managers on individual Data Hubs so that it is tailored to the exact needs of its users.

The tool allowing to build and edit a knowledge graph in the platform is called the concept map. The concept map of a Data Hub can be accessed through Admin > Data Finder > Open.

To create knowledge graph in the concept map editor, do the following steps:

1. Create a new concept. Click button New at the top panel. It will require to provide the following details:
   • Name - a term that people use to refer to a concept;
   • Type - defines a type of the concept to:
     • Indicator - a concept associated with numerical data such as production, sales, price, etc.
     • Named entity - categorical-type concepts like company, commodity, asset, etc.
     • Topic - a concept used to group other concepts, e.g. national accounts, demographics, ESG, etc.
     • Region - a concept describing a geographical area (note that you may use a checkmark "Include global regions to the Fact Finder results" to enable platform-wide catalog of geographical regions instead of adding regions manually)
     • Attributre - a concept used to refer to some property of an object, for example, logal, domestic, current, real, etc.
     • Unit - a concept used to refer to untis of measurement, like, dollars, tonnes, percent, etc.
   • Alternative names - list of synonyms separated by semicolon.
2. Map data items to the concept. Type "use <dataset ID>" (for example, "use FAOMCRST2018") in the command box and click "Execute". The dataset ID can be found in datraset URL after the DataHub name. Select one or several items of a dimension and click Bind data button at the top panel. Note that you may bind elements of only one dimension at once.
3. Link the concept to another one (optional). You can group similar concepts together by linking them to each other. Click Add link button at the top panel and select a concept that you want to be a parent. If you check "Break existing links" the concept will have only one parent, otherwise the concept will have several parents.
4. Verify the concept. Click Verify button at the top panel so that search engine indexes this concept and take it into account when processing search queries.

To edit an existing concept do the following:

1. Find a needed concept in the drop-down list at the top panel.
2. Click Edit button to rename a concept, change its type or add synonyms.
3. Click Add child button to create a related concept (for the "Add child" button to appear hover an arrow over concept name).
4. Click Multiselect mode button to apply the same action to several concepts or data elements at once. For example, you may select multiple data items to remove them from a concept. In a multiselect mode, there is a possibility to move selected data items to an existing or a new concept using the buttin Fork.
5. Click Delete data to remove all existing data mappings from a concept.
6. Click Delete to remove a concept. If you check "With data and links" the concept will be removed together with data mappings othetwise mapped data items will be moved to a parent concept. If you check "Recursive with childred" then not only the selected concept but also its children will be removed.

To export the concept map to or import from the file, do the following:

1. Click Download button at the top panel to save the concept map to a file.
2. Click Upload button to load a concept map from a file.

Semantic atlas: user-facing interface of the knowledge graph

A semantic atlas is a user-facing interface to navigate through the knowledge graph. It can be accessed through either a Data Altas or through Search. It loows as shown on the images below.

To enable the semantic atlas, you should have a concept map created in the Data Hub as described above. To access the semantic atlas directly, paste "/apps/data" into URL after the Data Hub name.