Research And Application Of Location Recommendation Technology Based On Spark

In the era of information blowout,it is a heavy task to find out user's interesting information from mass data.The task of recommendation system is to discover the hidden value of data.In recent years,the application of location-based services has also developed rapidly.These applications sampled a large number of location check-in data through the device location interface.Thus,it provides a condition for the realization of user-oriented location recommendation.Recommendation algorithm is an algorithm worthy of further study in data mining.Place recommendation is also an indispensable function in well-known applications.Aiming at some problems of location recommendation,such as sparse data,cold start and low personalization,this paper builds an improved recommendation model based on Spark big data platform.This model not only uses a single recommendation algorithm,but also makes full use of collaborative filtering recommendation,taking into account the current needs of users and users' history preference.The data sparsity is greatly improved by using the data filling method of user's place-attribute preference,and the similarity calculation method which integrates user differentiation and location's own quality is used to improve the similarity calculation effect of user and location.At the same time,the system uses Spark distributed cluster to realize parallel computing of the model,which improves the modeling ability under massive data and shortens the training time of the model.The validation data in this paper comes from the user's check-in data on Foursquare social networking site.API interface was used to capture 1510 2513 check-in data.In order to evaluate the recommendation effect of the algorithm,not only the average absolute error(MAE)commonly used in the recommendation system is selected to test the prediction effect of a single recommendation target,but also the normalized cumulative loss gain(NDCG)is used as the evaluation index of multiple results.The experimental results show that the improved hybrid algorithm proposed in this paper can effectively improve data sparsity and better recommendation effect.