Research On Multi-element Point Of Interest Recommendation Model Based On Deep Learning

Location-based social networks have become more incorporated into people's daily lives as Internet technology has advanced.By checking in and punching in,users will share their location information with friends as well as their favorite points of interest.With the rapid increase of these check-in contents,we can more correctly mine the features of users' interests over time from historical check-in information,and then recommend points of interest to users.However,accurately extracting user interests from a big amount of complicated data remains a difficulty.The majority point of interest recommendation algorithms have issues with sparse data,neglecting dynamic changes in user interest over time,and failing to integrate multivariate data for modeling.This study focuses on the following three factors after conducting extensive research on the aforementioned topics:(1)To solve the problem that users have different long-term and short-term interests,point of interest recommendation model integrating spatio-temporal preference features is proposed.To begin,the data set's user check-in records are split into working and rest days.The check-in sequence is then fed into the neural network,which extracts the user's long-and short-term goals.Finally,a Voronoi diagram was used to calculate geographical similarity,and the final recommendation result was obtained by combining spatio-temporal data from various scales.Experiments using available data sets reveal that the suggested algorithm has a higher accuracy and recall rate than similar algorithms.(2)Aiming at the problems of incomplete information extraction and insufficient coverage of interest point sequence,a new learning method about point of interest representation is proposed.From the initial point of interest sequence information,the Node2 vec technique is utilized to produce a more comprehensive interest point access sequence using a random walk strategy.(3)The problem of insufficient extraction of sequential feature information and spatiotemporal feature information of users' continuous check-in behavior.On the one hand,on the basis of the standard GRU,a module for processing temporal and spatial feature information is added,and on the other hand,an attention mechanism is introduced to the multi-strategy integrated point of interest recommendation model.Experiments using publicly available data reveal that the suggested model performs better in terms of recommendation.