Research Of Recommendation Model For Tourist Attractions Based On Sequence Information Mining

With the continuous development of Internet technology,people can enjoy the advantages brought by massive information,but there is also the disadvantage of information overload.With the emergence of personalized recommendation system,the recommendation system can help people to filter information and filter out useful information that meets users' preferences from the massive data,and it is widely used in different aspects such as life information category,traffic and travel category,video and movie recommendation category,etc.This thesis focuses on tourist attraction recommendation,using deep learning to analyze and model on users' historical attraction sequence data,which is a series of chronologically ordered attraction features that have been evaluated by users,containing not only information about the attraction itself and its category,but also rich hidden information in the resulting sequence relationships.Through the analysis of existing models,for tourist attraction recommendation,understanding the location relationship of users' historical sequences is crucial to understand the sequence of users' attraction preferences.By incorporating multi-head self-attention mechanism,not only can the feature information relationship of historical sequences be fully extracted,but also can solve the dependency problem between sequences when the sequences are too long,besides how to extract user explicit and implicit preference information from between sequences is also very important.Based on this,two recommendation algorithm models are proposed in the paper,and the algorithms are combined with the tourist attraction recommendation system,the main work is as follows:(1)The paper proposes the DCMPN(Deep Convolution and Multi-head self-attention Position Network)model,which uses the method of convolutional neural network to extract the user's historical interest feature information using horizontal and vertical filters,and then interacts the obtained information with the candidate attraction information to calculate the historical feature weight through the multi-head self-attention,the location information of historical interest sequences is extracted by the multi-head self-attention mechanism,and the historical features fused with the location information are then interacted with the candidate attractions for attention to calculate the historical feature weights.The final model realizes the deepening of user sequence interest and the fusion of location feature information to achieve the extraction of important preference degree and general preference of users and the distribution of feature weights among them.(2)The paper also proposes a model DIDFN(Deep Interest and Dual Features extraction Network)that considers the interaction between features,which can dynamically learn the historical feature importance and highlight the uniqueness of certain samples in historical sequences depending on the input target term.The model achieves explicit and implicit extraction of sequence features at two levels,vector-wise and bit-wise,and reassigns the extracted results as new inputs to the second-order feature interaction factor decomposer,and combines the historical sequence features with the candidate targets to do attention interaction mechanism together with the fully connected layer for connection leveling,activation function learning,and other operations,and finally realizes the recommendation of attractions that users may like.(3)Based on the two models proposed in the thesis,the validation is done on both private and public datasets and compared with some common advanced models.The experimental results show that both models proposed in the thesis can achieve the improvement of prediction effect to different degrees.Based on the proposed recommendation algorithm,a recommendation system for tourist attractions is designed and implemented,and the application point of the proposed model is shown through the page display of the system function.