Modeling User Email Preferences Using Biterm Topic Model

Email marketing is widely used by companies to communicate and maintain relationships with their existing customers. From a business perspective, email marketing is often personalized based on the customers’ historical transactions, which reflect their interest and could lead to repeat transactions. Unfortunately, not all industries have the luxury of using transactional data due to its limited and minimal amount of data. An example would be the hotel industry, which falls behind in each customer's transaction numbers compared to other sectors like retail and e-commerce. The challenge is how to use non-transactional data to find out the customers’ interests. This study proposes to model user email preferences based on the email marketing topics. The assumption is every user has one or more topics of emails as their interest and sending emails with the matching topics may result in more interaction. Biterm topic model for short texts is used to group emails with similar topics using its subject. This study validates the assumption by separating some of the latest campaigns to be observed with its historical email interaction. The result shows that around 69% of email openers within the last six months matched with their historical topic preferences. This finding proves that individuals have an interest and more likely to interact in some particular topic. Not only can these topics be used to give more relevance to the users, but they can also possibly help marketers achieve better marketing performance. CCS CONCEPTS • Information systems∼Information retrieval∼Document representation∼Document topic models