Augmented Deliberative Democracy (ADD-up): Enhancing Large-scale Public Arbitrations in Real Time


December 31, 2021

Valentin Gold. “Augmented Deliberative Democracy (ADD-up): Enhancing Large-scale Public Arbitrations in Real Time”, 2021.

As part of the ADD-up project, we have developed a rather flexible framwork consisting of separate modules. This requirement responds primarily to the needs of the political science discipline. With such a flexible framework, modules (such as input, analytics and visualizations) can be incorporated rather easily to the framework. Each module is wrapped within a docker container with (visual) results being deployed as an HTML-iframe. In principle, this framework can be made freely available and released under a CC-BY-SA 4.0 license which permits derivatives and adaptations but preserves free and open use of such derivatives and adaptations.

The framework is not only flexible with regard to possible extensions, but also with regard to programming languages (the software is written in R, Python, and JavaScript) and, more importantly, the social setting it is applied to. Before the software is started, the user is given the possibility to change the default settings to fit the social communicative setting, for instance to public consultations or to televised debates. One of the main important question to the user is whether (and how much) is known beforehand: If a list of topics is known before, users might want to include this list of topics to achieve better results of the appropriate analytics; if, however, users expect topics to evolve over time, another module is used for topic determination.

The first step either calls a module running speech-to-text API to automatically transfer spoken words to a written transcript (module speech2text) or simulates a stream of incoming words (module debug). Even though recent progress has been made in transferring speech to text, the transformation is far from being perfect – and might not be sufficient for in-depth linguistic analysis. Hence, we implemented an interface that allows real-time adaptations and corrections. Due to the fact that this service module highly depends on the software being used by the transcription service, we have refrained from programming such a service. If manual real-time transcription services are required, the framework can be extended to incorporate this service.

The most important module within our framework is a database (module mongoDB). The database stores the incoming chunks of text and all statistical calculations from the analytics modules. Each visualization module connects to the database, retrieves the required data, and generates an HTML-iframe to be displayed at large screens or to the wall. The backend module incorporates the settings and controls the access to the APIs. After the visualization iframes have passed the module nginx-proxy & nginx, access is simplified by a specific servername, e.g. . In many instances, we provide different visual approaches and have the user choose between those visualizations. The visualizations can then be combined to a specific collection of visualizations. Again, users can combine their preferred set of visualizations to fit the specific use case scenario.

Link to Framework