AI-driven relevance finder for activations
Loading...
URL
Journal Title
Journal ISSN
Volume Title
School of Electrical Engineering |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Authors
Date
Department
Major/Subject
Mcode
Language
en
Pages
52
Series
Abstract
This thesis investigates the viability of synthetic customer profiles for predicting customer engagement with marketing activations in the electronics domain. The research addresses critical challenges in customer analytics, particularly data scarcity, privacy constraints, and the need for scalable personalization strategies. We developed a comprehensive methodology for generating synthetic customer profiles using large language models, engineering a multi-dimensional feature space capturing lexical, semantic, statistical, and sentiment-based characteristics through both handcrafted metrics and dense embedding representations, and evaluating multiple machine learning algorithms across diverse synthetic profile generation strategies. Experimental results demonstrate that synthetic data can effectively support customer engagement prediction. Models trained exclusively on synthetic customer profiles achieved strong performance when validated on authentic Amazon customer data, demonstrating successful synthetic-to-real transfer learning. Detailed synthetic profiles consistently outperformed simpler generation approaches across all evaluation metrics. Linear classifiers emerged as the most effective algorithms for this task, demonstrating superior generalization capabilities compared to ensemble methods. Cross-dataset validation revealed strong generalization performance, with models maintaining consistency across different real customer data partitions. The research provides empirical evidence that carefully engineered synthetic customer profiles can serve as viable alternatives to real customer data for training engagement prediction models, offering privacy-preserving, scalable, and cost-effective solutions for personalized marketing applications. These findings have significant implications for marketing analytics, suggesting that synthetic data approaches can democratize access to sophisticated customer modeling capabilities while respecting privacy constraints.Description
Supervisor
Juvela, LauriThesis advisor
Kuikkaniemi, KaiKosunen, Ilkka