Service Chatbot

Hello, my name is Lumos

and I am the new service chatbot at Trier University of Applied Sciences. My name is an acronym that stands for “LLM Support with Online Support for Students” (german: LLM-Unterstützung mit Online-Support für Studierende). I am beeing developed as part of the AI ​​Rollout (german: KI Rollout) project at Trier University of Applied Sciences, which is funded by the Federal Ministry of Education and Research and the Rhineland-Palatinate Ministry of Science and Health.

Further Information

Project idea

My job is to help you with your questions about everyday student life. For example, you may have questions about finding your way around the main campus or organisational questions about your thesis. Since I will be available online around the clock, you don't need to observe any service hours and can simply ask me on the go.

Student projects

Sometimes there are a few tasks that you can solve for me. If you are interested, you should check VAPS regularly. But remember that you can only access this system from the university network or via VPN.

Structure

I am based on an LLM, which stands for Large Language Model. You have probably heard this term in relation to ChatGPT. It is an artificial neural network that has been trained to understand natural language.

The LLM that forms my basis is SBERT [1]. SBERT stands for Sentence-BERT. BERT has a good understanding of context in natural language [2]. SBERT uses this understanding and provides me with a mathematical representation of your questions so that I can understand them better.

I then try to classify your questions to give you the most accurate answer possible.

Current state

At the moment, I am trained to answer about 150,000 questions from 70 different areas. I got the questions from surveys among students, prospective students, staff and professors as well as from committees, which were then expanded with word substitutions to the approx. 150,000 questions.

I will publish a list of the areas I already understand here soon.

If you can think of any other areas or questions you would like me to answer, just send an email to Kevin Tittebrandt.

Current challenges

Currently, I still have to overcome a few challenges before I can go online. Unfortunately, I only understand German at the moment and that's only if there are no typos, spelling mistakes or grammatical errors. Previous tests with me show that if there are small errors in the question I type, my accuracy means that I no longer fully understand the question.

With regard to international students, I would also like to be able to answer questions in English. Of course, it would be best if I am completely independent of the language entered so that I can really help everyone.

Differentiation from ChatGPT

As you surely know from the media, ChatGPT was released at the end of 2022. GPT stands for Generative Pre-trained Transformer. The focus here, as the name suggests, is on generating chat messages that should be as human-like as possible. The correctness of the answer is not the primary goal [3]. However, I want to provide you with answers that you can rely on.

References
  1. "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks", Nils Reimers and Iryna Gurevych, november 2019, http://arxiv.org/abs/1908.10084
  2. "BERT:Pre-trainingof DeepBidirectionalTransformersforLanguageUnderstanding", Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova, 2018, https://arxiv.org/abs/1810.04805v2
  3. OpenAI, november 2022, https://openai.com/blog/chatgpt
Prof. Dr. Hans-Peter Beise
Prof. Dr. Hans-Peter Beise
Professor FB Informatik

Contact

+49 651 8103-755

Location

Schneidershof | Building O | Room 203
Kevin Tittebrandt, M.Sc.
Projektmitarbeiter FB Informatik

Location

Schneidershof | Building O | Room 5
back-to-top nach oben