CityCross: Transferring Attention-based Knowledge for Location-based Advertising Recommendation

With the rapid development of mobile networks and the widespread usage of mobile devices, Location-Based Adver-tising (LBA), which allows an advertiser to promote products or services to targeted customers in a suitable location, has drawn increasing attention. Recommending an optimal location by delivering appealing advertisements to potential customers is crucial for the advertiser. Existing recommendation models (such as collaborative filtering) are insufficient for solving the data sparsity and cold-start issue (e.g., no historical advertisement records in new domains) in LBA problems. To tackle the defi-ciency mentioned above, we propose a novel location-based ad-vertising recommendation framework: CityCross. The CityCross framework consists of a data extraction module and a learning module. The data extraction module conducts commercial and POI feature extractions from the LBA platform, and Gaode Map, respectively. The learning module is dedicated to learning the relevant knowledge of advertisement in a new domain by utilizing the attention-based semantic information, cross-city knowledge association, and the local neighbors' knowledge. The top-k locations are identified by a modified linear regression model based on the learned knowledge. Finally, we conduct extensive experiments on two real datasets to verify the superiority of the proposed approach.