In the era of personalized marketing, leveraging data analysis to recommend tailored products has become the key to building competitive advantage. As an essential aspect of customer information, family profiles can boost the inference of personalized needs. Combining positive and unlabeled learning and feature selection techniques, this paper designs a novel and flexible family profiling algorithm that tags target customers based on unlabeled transaction data. The empirical evaluation shows that our proposed algorithm outperforms other traditional algorithms regarding the recall rate of positive examples. This study also explores the balance between the effectiveness and efficiency of the proposed algorithm and compares the effects of different profile characteristics on the algorithm performances. The results of this study can be utilized to help retailers economically and efficiently predict customers’ family profiles, thereby reaching their target audiences and achieving better marketing performance.