Testing Recommender Configuration on the Workbench
12/09/2025
This section will give you some insights on how to test Recommender configurations. In our case, we will test the previously set up configuration which includes three calls. We will run these tests on the Workbench. It will give you a complete walk-through the complete test procedure which also gives you the opportunity to fine tune the saved configuration, if required. Furthermore, you will see how to compare documents (i.e. your input and any of the recommended results), as well as how to export and, of course, import configurations. You will also see the three sample cURL request calls we will run later to obtain recommendations.
After logging in the first page you see is the Workbench start page.
Here you see the selected project(s) (1) as well as the search space specified in GraphSearch (2). Clicking on the pencil symbol or on the Configuration button (6) opens the Settings window where you can configure several parameters, which are explained in our guide on how to Configure the Recommender using the Workbench. Enter your input text in the text field (4) or use the buttons on the top right (3) to upload a text from an URL or a file from your computer. To start the test, click the Arrow icon (5).
At the top of this page you see the selected search space project(s) and search space, below the input text. If you want to rerun the query, simply click on the Arrow icon (1). Use the gear symbol (2) to review and modify the configuration. The "..." button (3) opens a menu allowing you to import and export the configuration. The "+" button (7) underneath the input field (2) can be used to add additional inputs.
If you want to add the additional input text, click on "+" to open a new dialog. Here you can enter a title for your input (1), the actual input text (2) and confirm with the Save button (3) or discard these inputs by clicking the Cancel button (3).
Note
Saving the configuration will always save all input texts.
Below the "+" button you can see the INFERENCE TAGGING (5) and RECOMMENDER (6) tabs. The INFERENCE TAGGING tab will open by default, displaying your input text with extracted concepts highlighted in green. Hover over a concept (7) to view the concept's details, such as its pref label, the project it was derived from, and it's concept score.
This section of our documentation will focus on the RECOMMENDER tab. For more information on inference tagging, visit our Inference Tagging Quick Start Guide.
Navigate to the RECOMMENDER tab (1) to view a list of document recommendations based on the selected configuration. Each document has color-coded concepts displayed beneath its title and abstract, as well as annotated concepts which can be added manually by clicking on them (3). In the example above, the concept "Artificial Intelligence" has been previously selected from the list of annotated concepts.
Clicking on the SHOW DOCUMENTS button (2) will reveal a panel listing similarly color-coded concepts: these are the extracted, expanded concepts generated by the input query and configuration, as well as any manually added concepts which may have been selected. The screenshot below illustrates what this panel of extracted, expanded, and manually added concepts might look like.
If the concept scoring was included in your queries, then this panel will reveal the score for each concept with a gray background. This scoring also determines the sequence in which the concepts are displayed. Concepts with the highest score are listed first. You can also manually add a concept to this list or remove any of the annotated concepts.
To see the respective details of such document you only need to click on its title. This opens a new page.
Tip
Use the Return to Search results button (1) on the top left of the document details screen to return to the list of all results.
First you see the title of the selected document along with its computed score (2); below a list of concepts (3); the matching concepts are shown in solid color whereas concepts with no match have white background; the corresponding legend is accessible by hovering over the "?" icon (6) at the end of the concept list. Underneath the list of concepts is an excerpt from the selected document (4). The Document comparison button is to the right of the concept list (5). Clicking this button opens a dialog where you can view a comparison between concepts annotated in your input text and in the selected result.
This comparison informs you on all matching and non-matching concepts, along with a color-coded legend indicating their type (1). On the left you have the concepts annotated in your input text (2), on the right - concepts existing in the selected recommendation result (3). All matching concepts between these two texts are listed (4) first - in our example there is only one match; below you see a list of concepts with no match (5). Their position on this list is based on their respective individual scoring. This comparison can be helpful when you need to decide whether a particular result is useful for you or not. It also makes it clear why a particular recommendation has been made.
Tip
The longer the list with matching concepts the better the quality of the recommendation result. Also the type of matching concepts provides you with information depending on whether the match involves annotated concepts from the extraction, or from the expansion, or such have been manually added by you, respectively whether such are shadow concepts.
On the results page you can find a "..." button (3) to the right of the Configurations button (2). Clicking on this "..." button opens a menu allowing you to import and export the configuration.
There is an option to import an existing configuration created on another Workbench instance, provided it has been saved in JSON format. Selecting the Import menu item opens the following dialog:
Here you can paste the configuration (1) in JSON formal. After pasting the configuration, confirm by clicking on Import and apply (2) and then click on Close (2) to exit this dialog.
Note
If you only click on the Close button without confirming with a click on Import and apply you will abort the import operation.
You can Export Configuration as JSON file (1) or Export cURL Request (2). The JSON file can be re-imported on this Workbench or any other PoolParty Workbench. Click the Export Configuration button (1) to open a new dialog.
The recommender configuration (1) is shown on this window. You can simply click the Copy to clipboard button (2) to copy it or click Close (3) to exit this dialog. You do not need to highlight or select the JSON configuration.
The cURL requests are API endpoint requests which are separately available for each of the extract, expand, and recommend calls.
These three request are shown within one window. Next to the title of the request (1) you see a "Copy to clipboard" symbol - click on it to copy the respective request part of the API call to clipboard. Underneath you see the actual respective request (2) in cURL format. To close this window either click the Close (3) button at the very bottom of this window or the "x" symbol in the top right corner.
After having configured the recommender and having copied the individual requests we are ready to switch over to the section on how to use these sample API calls.