DIRAC: A Hybrid Approach to Customer Demographics Analysis for Advertising Campaigns

Digital advertising is a critical task for any business sectors to announce their new products and attract customers (audience). Digital advertising requires efficient market segmentation methods (i.e., customer clustering) to group customers into different categories based on age, gender, and other particulars so that ads can reach the appropriate target audience. However, such personal particulars are usually not available for advertisers to collect due to the privacy issue or the audience does not want to reveal it. To overcome this challenge, existing approaches analyze customer demographics by applying unsupervised machine learning algorithms to crawling data obtained when customers browse Internet websites. Crawling data includes customer browsing habit (e.g., browsing time) and hobby (e.g., frequently-visited websites). However, such algorithms are not robust against outliers or abnormal data. Further, they require a pre-defined number of clusters that may not be available in practice. To address the above challenges, in this paper, we develop DIRAC, a hybrid approach to customer demographics analysis for advertising campaigns. DIRAC consists of 4 phases that include data preprocessing, determination of the optimal number of clusters, clustering and clustering optimization. We carry out extensive experiments on a realistic dataset to evaluate the performance of DIRAC. The experimental results show that DIRAC provides useful insights on customer demographics and satisfies the expectation of advertisers while requiring 38% less execution time on the examined dataset.