Deep Learning based Image-Text Bimodal Scenic Spot recommendation model

Abstract During the 14th Five-Year Plan period, tourism income and tourist number in Northwest China increased year by year. However, many tourists do not have a deep understanding of the tourism scenic spots in the five provinces of Northwest China. In the case of having travel intentions but not yet clear specific tourist scenic spots, an intelligent recommendation system is urgently needed to assist them in tourism decision-making. This paper proposes an image-text bimodal tourist scenic spot recommendation model T-ECBM based on deep learning to realize personalized recommendation of tourist scenic spots in five northwest provinces. Experimental results show that when using BERT with a single text modality, the Top−1 accuracy is 83.14%. When a single image modality EfficientNet-CA is used, the Top−1 accuracy is 85.94%. The Top−1 accuracy of T-ECBM model is 96.75%, the Top−5 accuracy is 99.64%, and the F1 value is 96.72%, which verifies the significant advantages of the dual-modal model compared with the single-modal scenic spot recommendation method. The T-ECBM model not only increases tourists' understanding of the scenic spots in the five northwest provinces, but also provides tourists with intelligent and efficient decision support and personalized scenic spot recommendation, which effectively solves the problem of selection difficulty.