Desarrollo e implementación de sistemas de recomendación

Abstract

This project was developed in collaboration with Compensar, one of the leading welfare and compensation fund entities in Colombia, with the aim of optimizing product selection for events through a content-based recommendation system. To achieve this, advanced natural language processing (NLP) techniques were implemented, utilizing the TF-IDF (Term Frequency-Inverse Document Frequency) model to extract and analyze key product attributes, such as category, name, and suggested event. Based on keyword analysis and textual similarities, the system identifies the most relevant products for each event. Additionally, a Multinomial Naive Bayes model was incorporated for product categorization, enabling a more efficient and precise organization of items. This model, trained with previously labeled data, improves the alignment between products and different event types, facilitating decision-making in planning and logistics. The primary objective of the system is to provide personalized and accurate recommendations, enhancing the user experience and optimizing event management at Compensar. The system was validated using key metrics, such as classification accuracy and user feedback, ensuring its effectiveness and scalability across different data scenarios.