Large Language Models for Personalised and Interactive Explanations

Background

Large Language Models (LLMs) are transforming both personal and business interactions through their ability to enable natural language communication and process contextual information. In business settings, LLMs offer unique advantages by combining two critical capabilities: (1) the ability to analyze and interpret business data to extract meaningful patterns and insights, and (2) the capacity to communicate these insights through natural, personalized dialogue.

These capabilities are particularly valuable in the context of Recommender Systems and Decision Support Systems, where user trust and understanding are crucial for successful adoption. While these systems can generate sophisticated recommendations, their effectiveness ultimately depends on users' willingness to accept and implement them. Research has shown that explanations play a vital role in building this trust, with personalized explanations being particularly effective at increasing recommendation adherence rates.

However, traditional explanation methods often fall short in two key aspects: (1) They typically provide static, one-size-fits-all explanations that fail to account for individual user contexts and preferences. (2) They lack the interactive element that allows users to probe deeper or request clarification on specific aspects of the recommendations.

LLMs present a promising solution to these limitations through their ability to generate dynamic, contextually-aware explanations and engage in follow-up dialogue. While existing research has demonstrated the potential of personalized explanations in controlled environments, there is a critical gap in understanding how LLM-powered explanations perform in real-world business settings, where they must interpret pricing data and customer interactions to provide meaningful, actionable insights.

Research Goal

This thesis aims to investigate the effectiveness of LLM-powered explanations in a real-world business context, specifically focusing on the following research questions:

  1. How can LLMs be leveraged to analyze pricing data and generate comprehensive explanations that help users understand the rationale behind specific pricing recommendations?
  2. To what extent can LLMs tailor their explanations to individual users' needs and knowledge levels while maintaining consistency with the underlying pricing logic?
  3. How can an LLM-based explanation system be designed to effectively reduce routine customer support inquiries while identifying cases that require human intervention?

The research will be conducted in partnership with RoomPriceGenie, a startup specializing in dynamic pricing recommendations for hotels. Their platform presents an ideal testing ground for these questions, as it combines sophisticated pricing algorithms with a user base that frequently seeks explanations for pricing recommendations

This thesis will involve integrating an LLM with RoomPriceGenie’s database for customers, extract relevant information, and generate context-appropriate explanations for pricing recommendations. These explanations should be tailored to individual users, incorporating data such as previous customer interactions, sales trends, and specific business needs. Possible indicators to evaluate the system might include the reduction in customer support requests and the adherence to pricing recommendations from the startup.

Your Profile

We look forward to receiving your application because you:

  • Are currently enrolled in KIT master's program.
  • Have a strong foundation in machine learning and artificial intelligence.
  • Are proficient in programming languages such as Python and have experience with language modelling frameworks like HuggingFace, LangChain, and PyTorch.
  • Are passionate about AI research and eager to explore new frontiers in natural language modelling research.
  • Possess excellent analytical, problem-solving, and communication skills.

Details

  • Start: Immediately
  • Duration: 6 months
  • Language: English for communication and final thesis
  • Location: Flexible
  • Payment: Provided by the startup RoomPriceGenie

How to Apply

Interested students should submit the following:

  • A resume or CV highlighting relevant coursework, projects, and skills.
  • A brief statement of interest explaining why you are interested in this project and how your background and skills make you a suitable candidate.
  • Any relevant academic transcripts or references.

For more information or to submit your application, please contact us at:

Join us in pushing the boundaries of artificial intelligence and making contributions to language models in research. We look forward to working with talented and driven students who are ready to take on this exciting challenge.

References

[1]: Silva et al., 2024, „Leveraging ChatGPT for Automated Human-centered Explanations in Recommender Systems“.

[2]: RoomPriceGenie - The most intuitive room pricing solution ever](https://roompricegenie.com/)

[3]: Yang, Zhu, and Wang, 2023, „User Perception of Recommendation Explanation: Are Your Explanations What Users Need? | ACM Transactions on Information Systems“.

[4]: Yang u. a., „Fine-Tuning Large Language Model Based Explainable Recommendation with Explainable Quality Reward“.