LARGE LANGUAGE MODEL FINE TUNING FOR MARKETING COPY GENERATION IN FASHION ADVERTISING

Fashion advertising is an essential component of the fashion industry, involving the marketing and communication strategies employed to promote products and brands. In recent times, the fashion sector has shifted from traditional communication channels to online platforms, leveraging social media to reach a broader audience and stimulate interest. However, this transformation has introduced new challenges for fashion companies. They must now produce creative and effective communication texts to stand out, adapt to different communication styles across diverse platforms, maintain brand consistency, and create authentic messages that resonate with their target audience. This thesis tackles these challenges by presenting a comprehensive framework that integrates methods and tools for generating advertising texts using Large Language Models (LLMs). LLMs are artificial intelligence models trained to understand and generate coherent texts. They leverage transformers, a deep neural network architecture, to learn linguistic rules and structures from the training texts. Within the scope of this thesis, a specific software architecture has been developed for generating fashion advertising texts. Information has been extracted from a fashion company's communication plan over the past five years, which has been subsequently used for fine-tuning the model. Various techniques for data cleaning, synthesis, and optimization have been evaluated and applied to achieve the desired outcome. Furthermore, a tool that allows non-technical personnel to use the model has been implemented. This platform integrates a React web interface and utilizes the FastAPI framework for constructing APIs. The validation and evaluation of the model have been conducted on test data and, whenever possible, through testing sessions involving marketing and communication experts to ensure the effectiveness and appropriateness of the generated texts.