knitr::opts_chunk$set(echo = TRUE, fig.width=9, fig.height=5)

Facepager is an open-source program that can be used to automatically retrieve data from the internet -- primarily via APIs (application programming interfaces). These are interfaces through which platforms such as YouTube, Instagram, Facebook, and even LLMs provide information automatically. A more detailed explanation of APIs and their use can be found here. When addressing the Databoard, the LLM (currently the LLaMA 3.1-70B model) is addressed. Various workflows are possible for this, with the coding workflow being discussed in more detail below. Facepager makes it possible to automatically send many requests in succession and store the responses in a structured manner in a database.

The summary workflow is suitable for qualitative content analysis. The resulting summaries can be formed from keywords on specific topics or categories, or consist of several sentences.

In Facepager, coding refers to the process of analyzing and categorizing data content. Predefined categories for a text are assigned by an LLM (as in quantitative content analysis). A schema with corresponding descriptions of the existing categories and examples must be defined for this purpose. Comments, posts, or other texts downloaded from Facepager via an API query can be used, as can data imported from your own sources. The Preset Databoard, whose use is explained below, is intended for this workflow.

Facepager can be used to send prompts to an LLM and save the returned responses. The latest Facepager (desktop) version must be downloaded for all applications to function. It can be found here.

This overview is extremely helpful for familiarizing yourself with the user interface and recognizing the terminology used below. For a more detailed look at the individual areas, we recommend reading the Facepager Wiki.

Select the preset: Here, you can click on the respective workflow between Coding and Summarize. Under Coding Submit task, you will find step-by-step instructions on how to import data. Then select Apply to set the default settings for the workflow. These are then visible in Query Setup.

Open new database: Next, open and name a new database (under New Database) to specify the storage location for the API requests and responses. Add Nodes: To add your own data, select a CSV file (Load CSV) containing the relevant data under Add Nodes. The nodes contain the rows of your own data collection and serve as the starting point for the query. These can be ID numbers, for example. These are listed in the Nodes View area after they have been imported.

Delete Nodes: Delete Nodes can be used to delete imported files if something has been imported incorrectly. However, since deletion takes a long time, it is advisable to open a new database instead.

Rules: Rules are necessary for coding tasks to determine the conditions under which coding takes place. You can either insert your own rules in the Payload window or upload the code book in the form of a JSON file (an explanation of this is also available in the preset). Payload Window: To display the Payload window, Posts must be selected as the method in Query Setup. Rules are required for the workflow to function and for data to be extracted or categorized. These can be entered in the payload. The syntax in the payload window must be correct, otherwise error messages will occur. For example, missing paragraphs or forgotten quotation marks can cause an error message.

In the case of the following entry in the payload window, the aim is to create a coding workflow in which text is coded in the form of the imported CSV file. The multi mode determines which system prompt Facepager addresses (this is not visible to users and is selected automatically) and thus which coding variant is used. When selecting multi, several gradations can be assigned to a category, provided that this applies or the absence is recognized, as in a manual content analysis. Under Rules, the Rules.json file is assigned as the set of rules for coding. These settings must be customized for your own implementation.

{ "task": "coding", "input": "<text|json>", "options": { "mode": "multi", "rules": <"rules.json"|file:txt> } }

The base path specifies the interface to which the query is sent. It is already filled in by the preset. Parameters can be set in query setup. If the result is available within ten seconds, it is returned. If this does not happen, select Presets Databoard > Coding > Get task results to initiate a new query. The necessity of this procedure is indicated by status 202, which is assigned the label Pending after the failed fetch. This label is visible in the query setup after sending a query.

In Query Setup, select “header” under Authorization. The password can be entered in the Access token field. This must be assigned in advance. Then click on Settings and enter the username under Client ID. You can then click on Login. You will be asked again for the username and password that you have already entered in the Access Token field. Without logging in, no query is possible.

Under Response in Query Setup, enter the file type you want to receive, e.g., a JSON file. Get results: To receive the results of the query, one or more nodes must be selected. The Fetch Data button is then clicked to initiate the start of the workflow. Facepager sends the query for each node to the LLM and displays the response in the Response field. Finally, the data can be exported (under Export Data).

Working with an outdated version of Facepager: In this case, not all functions will likely be available, meaning that the workflow will not function reliably. We recommend downloading the latest version.

The request error indicates that the query did not work. This problem can occur because you did not log in to the Databorad. This requires an access token, which must be assigned in advance.

For recognition, you must specify in the payload that reference must be made to the rules in the code book of the JSON file.

If the Run Pipeline command in the preset cannot be activated, Facepager must be restarted.

If the login input mask does not load, the window must be closed and reopened.