Research On Technology Of Tourism Recommendation Based On Deep Learning

As the rapid development of tourism industry recently,the tourism information in cyberspace increases sharply.The huge and complicated tourism data often causes the overloaded information,which also makes it hard for users to acquire the information they desire promptly and exactly.For this issue,recommendation technology has provided a feasible solution.However,traditional recommendation algorithms,which construct models via the interaction between users and items mainly,generally possess the weak learning capability when the single rating data and the sparse rating matrix are applied.As a result,the quality of recommendation and experience of users are harmed severely.In this article,deep learning is exploited to explore the potential associations and the multiple complex relationships between users and items.With offering a reliable data source and relieving the sparse data,the recommendation accuracy is upgraded further.The research contents of this article are mainly as follows:(1)Aim to resolve the sparse interaction data between users and attractions as well as the underutilized information exist in traditional recommendation algorithms,an attributeaware graph neural network(AGNN)is proposed for tourist attraction recommendation.As the tourism plan of user is often decided jointly from various aspects and there are rich attributes for a tourist attraction,therefore this article explores the attractions users are interested in from the attributes of attractions to obtain the preferences of users accurately.First,the network among attractions is built with their attributes and another network between users and attractions is established with their interactions.The structures and information in these multiple networks are extracted by the GNN model.Next,since different interactions between users and attractions have different influences on user preferences,as well as various associations among attractions,attention mechanism is adopted to solve this problem.Then,the embedding vectors of users and attractions are fused via the multi-layer perceptron to accomplish subsequent recommendation mission.Last,the comparison between the AGNN model and other traditional and mainstream algorithms are conducted on the attractions dataset which is gained from tourist websites via the Scrapy framework.Experimental results reveal that the AGNN model all receive remarkable increases in both the RMSE and MAE metrics.(2)A gated graph neural network via multi-feature fusion(GGNNMF)for tourist attraction recommendation is proposed to relieve the limited recommendation accuracy caused by the variability of user interest.First,the attraction session graph and the corresponding attraction attributes session graph are built in chronological order.The GGNN model is utilized to extract the sequential information and the long short-term interests of users from the attraction session graph and the corresponding attraction attributes session graph,meanwhile,attention mechanism is adopted to distinguish the contribution of the local neighboring nodes.Next,the embedding vectors of users are obtained by the one-mode projection from the network between users and attractions to the relation graph among users.Then,both the embedding vectors of attractions and the embedding vectors of users are combined to acquire more accurate session representation.Last,the GGNNMF model is compared with the traditional and session recommendation algorithms on the tourism attractions session dataset.Experimental results show that the GGNNMF model achieves outstanding accuracy and recommendation performance comparing with the state-of-the-art model.