Abstract

Abstract

Abstract

Abstract
Despite being one of the most cost-effective methods, email marketing remains challenging due to the low rate of opened emails and the high percentage of unsubscribed campaigns. Since the sender and the subject line are the only information that the recipient sees at first when receiving an email, the decision to open an email critically depends on these two factors, which should stand out and catch the recipient's attention. Therefore, the motivation behind this study is to support email campaign editors in choosing a subject line based on its potential quality. We propose and compare several models to measure the quality of a subject line, considering its potential to promote the email opening. The subject lines' structure and content are explored together with different machine learning techniques (Random Forest, Decision Trees, Neural Networks, Naive Bayes, Support Vector Machines, and Gradient Boosting). To validate the proposed model, a data set of 140,000 emails' subject lines was used. The results revealed that the models proposed are very promising to support the definition of the email marketing subject lines and show that the combination of data regarding the structure, the content of the subject lines, and senders characteristics leads to more accurate classifications of the potential of the subject line.

Abstract

Abstract
By carefully recommending selected items to users, recommender systems ought to increase profit from product sales. To achieve this, recommendations need to be relevant, novel and diverse. Many approaches to this problem exist, each with its own advantages and shortcomings. This paper proposes a novel way to combine model, memory and content-based approaches in a cascade-hybrid system, where each approach refines the previous one, sequentially. It is also proposed a straight-forward way to easily incorporate time-awareness into rating matrices. This approach focuses on being intuitive, flexible, robust, auditable and avoid heavy performance costs, as opposed to black-box fashion approaches. Evaluation metrics such as Novelty Score are also for-malized and computed, in conjunction with Catalog Coverage and mean recommendation price to better capture the recommender's performance.