Published November 2019.
The core idea of the project is to develop a technology for the dynamic detection of so-called “intents” for dialog systems, which can be determined “online”, i.e. in this case during the operation of a service hotline. For example, a user may report that a certain service is no longer working, that more questions are being asked about current topics (e.g., deadlines for submitting documents), or that legal principles are changing (e.g., the issuance of new directives). Consequently, the intent is dynamically derived using Natural Language Understanding processes and possible triggerable actions are determined. This requires both the generalization of input and recognized patterns and the semi-automatic derivation of intents. If this technology can be developed successfully, a dynamic extension of the dialog system is possible, in which both new intents and training data extracted for them are proposed to the data scientist responsible for the dialog system. If the analysis, which is initially related to one domain, is successful, the technology can be extended to other domains and thus, in the long term, dynamic markets can be developed with semi-automated chatbots.
So far, no approaches exist that completely fulfill the characteristics envisioned here of dynamism, use of natural language, domain specificity, and the possibility of use in a service hotline. Thus, they mostly have cuts in at least one of these areas and none can guarantee the use according to German guidelines. The planned “Cognitive Assistant” should therefore not only combine all the strengths of previous approaches, but also allow use according to German guidelines, so that the project partner is also enabled to use such a service assistant in highly sensitive DSGVO areas.
The data set to be used is data from the publicly accessible Info-DB, which is provided by the 800 employees at DATEV from the areas of service, consulting and technical consulting. The necessary compliance with the DSGVO can also be ensured when processing the data by using the SDIL platform. For example, this allows the data to be stored and processed on secure servers in accordance with the specifications of the unsupported partner. In the short term, the newly recognized intents will be integrated into a chatbot engine by development teams (semi-automatically). Medium- and long-term exploitation is ensured by integrating the technology as a module in the SPEAKER platform (the executors in the project proposal are also part of the SPEAKER development team).
01.11.2019 – 30.06.2020
Prof. Dr. Jens Lehmann (Jens.Lehmann@iais.fraunhofer.de), Fraunhofer IAIS
Proff. Dr. Andreas Both (firstname.lastname@example.org), Head of Research, DATEV eGro
Translated with www.DeepL.com/Translator (free version)