ZHU Zhiguo, SUN Yi, WANG Xiening, WAN Xiaoji
With the rise of social media such as “Mafengwo”, more and more tourists have become increasingly inclined toward the more flexible and freer self-organizing “tour group” rather than the traditional and standard “tour packages”. Different from traditional individual personalized recommendation, it has become a hot issue with important practical value that how to better aggregate the heterogeneous tour preferences of members for accurate tourist group recommendation. To this end, the method of Local Outlier Factor (LOF) is firstly adopted for data preprocessing to identify the outlier users with large differences in tour interests, and then the tourist groups can be preliminarily clustered. Next, the model ANC-TGR (attention-based neural collaborative tourist group recommendation) is proposed. In this model, the tour preference representation of a tourist group can be accurately aggregated through a well-designed two-layer attention network of “item-level” and “user-level”, and the representation vector is further input into a neural collaborative filtering recommendation framework for accurately recommending the Top-$N$ attractions for the tourist group. In the datasets of Mafengwo (with groups) and Foursquare (without groups), the experimental results confirm that the proposed model ANC-TGR, which further optimizes the preference representation of the fusion tourism group, compared with the optimal benchmark model, increased by 10.45%, 10.48%, and 10.07%, 10.87% on the metrics of HR@10 and NDCG@10, respectively. This paper provides technical support to improve the accuracy of attraction recommendations and travel satisfaction of tourism groups.
