Kodein-Penetration: Recommendations of Customer Personalization Level in A CRM using Deep Learning

Abstract

This study aims to develop a personalization-level recommendation model implemented in the Customer Relationship Management (CRM) system at PT Kodegiri, called KodeinPenetration. Personalization in CRM aims to improve customer interaction by providing more relevant recommendations based on their needs and preferences. To achieve this goal, this study tested several classification models using historical customer interaction data as the basis for analysis. The classification models tested included decision tree-based methods such as Random Forest, Gradient Boosting, and AdaBoosting, as well as deep learning models such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). In addition, two main feature extraction techniques were applied to process text data, namely TF-IDF (Term Frequency-Inverse Document Frequency) and Tokenizer Padding. TF-IDF is used to represent words as numeric vectors based on their frequency of occurrence. In contrast, Tokenizer Padding is used in deep learning models to convert text into a numeric format that neural networks can process. The test results showed that the decision tree-based method using the TF-IDF feature produced the best accuracy of up to 82%. On the other hand, the deep learning model with GRU architecture utilizing Tokenizer Padding achieved the highest accuracy of 88.23%. This shows that the deep learning model has greater potential in handling sequential data and providing more accurate results compared to traditional methods. This study provides an important contribution to the development of deep learning-based personalized recommendation systems in CRM. By leveraging historical customer interactions, this system can improve user experience by offering more relevant and targeted services.