Research on the recommendation method of urban location point of interest based on DTCN-EFFN-Transformer

In recent years, user travel has generated massive mobile data. Recommending the next location of interest not only greatly facilitates user travel, but also provides a corresponding opportunity for merchants to mine potential customers. In view of the fact that the existing urban location POI recommendation does not integrate multiple context information, leading to a series of issues, such as the inability of the recommendation results to accurately meet user needs. This research introduces a method for urban location POI recommendation utilizing the DTCN-EFFN-Transformer. This method combines user preferences, geographic location and popularity of POIs. Firstly, the deep convolution in DTCN simplifies the complexity of input data, so that the dilated causal convolution in TCN can focus more on extracting local features in user check-in records. Secondly, an expansion factor is added to the linear layer of EFFN. By extending the dimension of the middle layer, the nonlinear representation ability of the model is enhanced, thereby the long-distance dependence ability of the model in capturing user check-in records is improved. Experimental results indicate that using the FourSquare-NYC and FourSquare-TKY datasets, this newly introduced method improves Acc@1 index by 10.42% and 6.13%, respectively, compared with similar representative methods.