Research On Point Of Interest Recommendation Algorithm Based On Neural Network

In recent years,the rapid development of the mobile Internet has made people's lives more and more convenient.Nowadays,people can connect to the Internet through terminals such as mobile phones and PCs,and use a variety of apps to learn,shopping and play.The rapid development of the Internet has spawned a large amount of information and the daily amount of information is still growing at an unimaginable rate,people are facing serious information explosion problems.In order to solve the information overload problem,related technologies have experienced three stages of catalog classification,search and recommendation.Nowadays,recommendation has become an extremely important information filtering technology.It has been widely studied in many fields such as machine learning and information retrieval,and has been widely used in many application scenarios such as music and video.It not only can enhance the user's experience but also helping the merchant to achieve intelligent goods and service push,thereby increasing the business revenue of the business.1)Recently,with the development of mobile Internet,cloud computing and artificial intelligence,location-based social networks have been welcomed by more and more people.Points of interest recommendations play an extremely important role in location-based social networks.The traditional method is to make recommendations based on the user's interests.This kind of interest is usually a general preference.These traditional methods rarely consider the continuity of the user's visit location and ignore the order relationship between the user tracks.Predicting the user's interest points that will be visited at the next moment can help the merchant to make visit predictions while helping the user make decisions at a specific time,which is convenient for the merchant.However,the current problem still faces some challenges,such as data sparsity and the difficulty of sequence modeling.Based on these opportunities and challenges,this paper has achieved the following research work and innovations:1)Similar self-joining problems have important applications in the fields of collaborative filtering and text scrutiny.Many algorithms have been proposed for this problem,but so far,similar self-joining algorithms for high-dimensional data are rareproposed.This paper designs a novel three-dimensional similar self-joining method based on diamond structure.It uses its spatial solid structure and regards the atomic nodes in the spatial structure as the computing node domain,combined with spatial node localization and coordinate filtering.The method of similarity matching on large-scale spatial datasets can quickly obtain all pairs of similar results.2)In recent years,algorithm models using recurrent neural networks(RNN)and convolutional neural networks(CNN)have emerged in an endless stream.However,both models have drawbacks when applied directly to recommended tasks.More specifically,they lack the ability to explicitly capture item-item interactions throughout the user history.The motivation to build item-item relationships in the user's context history is intuitive,as it is critical to understanding the fine-grained relationships between individual item pairs.This paper proposes a POI sequence recommendation algorithm based on the multi-head self-attention mechanism to capture the user visit order representation by learning the sequence of consecutive locations.In general,the user's short-term preferences are learned in a unit of attention-based and GRU,which explicitly invokes the location-location interaction in the user's historical access sequence.That is,it can learn short-term information on the relationship between L consecutive locations.Based on this output,we combine the user embedding vectors to form the final representation.Our experiments show that the proposed model can obtain better performance indicators and demonstrate the effectiveness of capturing both short-term interests and general interests.