Conversational AI

Feasibilty study: “Conversational AI for domain-independent learning dialogue”.

01.07.19 – 31.12.19

The project is based on the hypothesis that learning dialogues, independent of the learning content
a manageable finite number of identical intents (“intentions”) exist.

These intents could be something like this:

– to “continue” (“learn further”) in the learning material
– practice/repeat something you have already learned
– after asking something specific (“look up”)

To be able to process these requests, it is necessary to recognize the Intent, the
domain entities from the previous context and to know the semantic
Model relationships between the contents.

NER (Named Entity Recognition), Topic Models and network analysis are used to extract central domain entities. Entities are extracted from the pre-processed corpus texts and assigned to learning units. The central problem here is to determine the different importance of the entities for the learning unit with regard to the possible intents.

Based on the extracted entities, learning units are to be networked, semantic
relations such as similarity/equivalence (sameAs), but also didactic
Composition and dependencies (coursePrerequisits, isPartOf, hasPart,
educationalAlignment) can be modelled.