Leveraging the capabilities of Large Language Models in Questionnaire Generation for Human Resource Management

My thesis explores the Survey Questionnaire Generation task in the Human Resource Management field. Although poorly explored and characterized by the lack of dedicated materials, this study overcame such limitations by proposing a comprehensive methodology. Therefore, we proposed a dataset collecting survey questionnaires for HRM, which was used to leverage the capabilities of LLMs, namely GPT-3.5-Turbo and GPT-4-Turbo, in such a task.

Computer Vision techniques for Violence Recognition

My thesis focused on the application of PoseConv3D to Violence Recognition, a sub-domain of Activity Recognition. We trained it onto four datasets to perform a binary classification. The videos collected in the involved datasets were pre-processed in order to extract the poses to reduce the bias induced by the backgroud of the scenes.