| title |
|---|
Home |
In the Conversational AI and Social Analytics (CAISA) Lab, we combine diverse expertise from areas such as natural language processing, machine learning, and computational social sciences, on a mission to understand people behind the language.
We incorporate personal and social context into our language understanding and language generation models, aiming at building more user-centric conversational agents as well as more accurate web discourse interpretation systems.
CAISA Lab is led by Prof. Lucie Flek and resides mainly at the Bonn-Aachen International Center for Information Technology (b-it) in Bonn Germany and Department of Mathematics and Computer Science of the Philipps Universität Marburg in Germany.
Below is a selection of our recent focus areas. See the project page or our news blog for more.
{% capture text %} How to represent users for social NLP tasks? Which subjective factors matter for interpreting a conversation? How to distinguish among users with varying preferences while preserving as much privacy as possible? {% endcapture %} {% include feature.html image="images/userwoman.png" link="projects" heading="User representation learning" text=text %}
{% capture text %} How to best model subjective human-machine dialogs, e.g. with opinionated agents? What are the user expectations on conversational AI in subjective areas, and how to evaluate these? {% endcapture %} {% include feature.html image="images/Conversational.png" link="projects" heading="Opinionated conversations" text=text %}
{% capture text %} Which personal and social factors influence opinion changes and fluctuations? Which factors play a role in accepting and spreading misinformation, and can these be affected by the quality of a conversation? {% endcapture %} {% include feature.html image="images/Network.png" link="projects" heading="Opinion formation and dynamics" text=text %}
{% capture text %} How can we transfer knowledge among conversational and social NLP tasks? How to best adapt existing NLP models to expert domains, such as medical or financial discourse? How can we effectively augment conversational tasks with synthetically generated data? {% endcapture %} {% include feature.html image="images/smalldata.jpg" link="projects" heading="Learning more with fewer resources" text=text %}
We are a diverse group of collaborators in areas such as natural language processing, machine learning, and computational social sciences. {%- include big-link.html icon="fas fa-user-friends" text="Team" link="team" -%}
If you are interested in joining our team, or are looking for a topic for your bachelor or master thesis, get in touch! {% include big-link.html icon="fas fa-hands-helping" text="Join us" link="contact" -%} {:.center}