Design And Implementation Of An Adaptive Tourist Attraction Recommendation System Based On User Interests And Risks

As a popular research direction of point of interest recommendation,scenic spot recommendation has been widely used in tourist travel activities.With the development of transportation,the range of travel for tourists has expanded significantly,and the negative impact of data sparsity on the scenic spot recommendation model is also increasing.With the development of complex social networks such as LBSN,how to more effectively utilize the increasingly rich travel trajectories of tourists and explore the links between tourist check-in behavior contexts in a more finegrained manner has become a hot research direction nowadays.In addition,the prevention and control of epidemics and other dangerous situations in recent years have also become the direction that the scenic spot recommendation system needs to be improved in the application field.Considering the above problems,this paper studies and designs an adaptive scenic spot recommendation system based on user interests and dangers.The main tasks are as follows:(1)A Logistic matrix factorization model based on the user’s local interest is proposed,which abstracts the local spatio-temporal context information into the user’s local interest,and integrates it into the matrix factorization model using a probabilistic method.Model the user’s preference probability for a scenic spot through the logic function,and then integrate the spatio-temporal information of the context to model the user’s global interest from local to global.The model demonstrates better performance than various benchmarks in the FourSquare and Go walla datasets,verifies the effectiveness of local interest through ablation experiments,and shows stable results under different sparsity tests in the sparsity experiments.(2)A social recommendation algorithm based on spatio-temporal activity center is proposed,which models the user’s spatio-temporal activity center according to the geographical aggregation characteristics of user activities,and on this basis,models the user’s social relationship,and jointly models the user’s explicit and invisible social interaction.social relationship.Finally,time,space and social relations are integrated into the matrix factorization method as context information.Through comparati ve experiments and ablation experiments,the effect of the social recommendation algorithm proposed in this paper on alleviating the data sparsity problem is verified.(3)A scenic spot recommendation model based on the user’s longterm and short-term preference adaptation is proposed.The time-space gate logic is added to the LSTM to strengthen the model’s learning of spacetime dependencies,and the hidden state of the user,the hidden state of the scenic spot and the hidden state of the trajectory sequence are combined to further A scheme for adaptive fusion of long-term and short-term preferences based on contextual information is proposed,and the user’s short-term and long-term preferences are dynamically combined to complete the adaptive recommendation task of scenic spots.The improvement effect of the model on personalized scenic spot recommendation is verified through comparative experiments,and the effectiveness of each optimization is verified through ablation experiments.(4)A cross-regional recommendation model based on NVDM-GSM is proposed,using the cyclic neural network model in(3)to obtain the known interest preferences reflected in the user’s existing check-in trajectory,and migrate to the off-site preference through the nonlinear mapping function of the multi-layer perceptron the vector space of On this basis,the neural topic model NVDM-GSM is used to encode and decode the travel records of a similar user set in the user’s remote area,and the variational reasoning is used to reveal the user’s travel intention in the area under unsupervised conditions.Finally,the user’s potential preferences in different regions are obtained by combining the known preferences of the above-mentioned users who have migrated with their travel intentions in different regions,and solve the cold start problem and interest drift problem of cross-regional recommendation.In comparative experiments on two real-world datasets,the proposed model outperforms each baseline model in recommendation results,which verifies its effectiveness in cross-regional recommendation scenarios.(5)Combined with the above recommendation model,this paper designs and implements an adaptive scenic spot recommendation system based on user interests and dangers,comprehensively considering the main problems of data sparsity,extracting user preferences from complex checkin trajectories,cold start and interest drift in cross-regional recommendations,etc.The recommendation results are adjusted in a timely manner according to user feedback and real-time danger conditions,which improves the effectiveness and real-time performance of the recommendation results.