Enhancing Recommender Systems by Adapting LLMs to Users' Values

Energy communities are a novel approach to cooperative energy management; however, because of the complex nature of user interactions, sophisticated tools that strike a balance between customization, openness, and flexibility are needed. The integration of large language models (LLMs) and recommender systems (RSs) is examined in this thesis, with a focus on the transformative potential of LLMs as generative engines. Properly calibrated to prototypical user archetypes and tuned to the energy domain, LLMs are involved in several stages of the interaction flow, from comprehending individual needs to producing contextualized explanations and personalized recommendations. LLMs' generative ability makes it possible to produce dynamic and adaptive content, which improves user interaction and communication. Additionally, their sophisticated semantic knowledge guarantees that user requirements and community resources are effectively mapped. This collaboration between LLM and RS not only increases system efficiency but also reconsiders the way humans and computers interact by providing decision support that actively adjusts to the priorities and values of community members. The findings demonstrate how the application of LLMs enhances the relevance, transparency, and motivation of recommendations to support cooperative and sustainable energy management.