Browsing by Issue Date, starting with "2025-06-05"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Enhancing Financial Product Recommendations through Chatbots and Large Language Models: A User-Centric ApproachPublication . Pereira, Pedro Manuel Noutel; Cunha, Carlos Augusto da SilvaIn an era where the intersection of artificial intelligence and digital finance is be coming increasingly relevant, the application of Large Language Models (LLM) to investor risk profiling is a revolutionary possibility. Traditional risk profiling meth ods rely too heavily on structured questionnaires, which fail to capture the nuances of investor behaviour, resulting in stereotyped categorization and more often than not, inaccurate representations of risk tolerance. This thesis explores how LLMs, specifically using OpenAI GPT series, can enhance the accuracy and personalisa tion of risk assessment models by leveraging natural language understanding and adaptive learning capabilities. This study explains the limitations of conventional profiling techniques and pro poses an LLM-based model that is dynamic in nature and adjusts according to in vestors’ responses, making the classification more precise with interactive discussion. To identify the effectiveness of the model proposed, the Morningstar risk profile is taken as a training and testing dataset for the LLM, while the traditional approach is analyzed, based on Google Forms, for comparing dynamic and static models. The results indicate that LLMs can significantly contribute to investor profiling by providing a conversational, real-time assessment process that complements the accuracy and user engagement, rendering the monitoring process more personalized. They also show that LLM achieves a significant improvement in investor profiling through a more flexible and interactive classification procedure. The model out performs conventional methods in cases where investor responses are ambiguous or subjective, ensuring better alignment with risk tolerance. In addition, it was ob served that the GPT-based classification tends to be conservative in the case of uncertain responses, which can serve to avoid excessive risk-taking. These results validate the potential of leveraging LLMs to augment financial advisory services by providing a more accurate and responsive risk assessment to individual investor profiles. By demonstrating the advantages and limits of AI-based investor profiling, the thesis contributes to the growing body of literature on LLM application for making financial decisions. The findings pinpoint the transition from rigid, questionnaire based questionnaires to adaptive, context-aware systems, with new potentials for augmenting financial advisory services in an increasingly digitalized economy.