Personalized Recommendation Method Based On Prediction And Analysis Of User Interest Time Series Fluctuation

With the wide application of personalized recommendation technology in e-commerce,people pay more and more attention to the accurate collection and use of user's personalized interest information to improve the effectiveness of the existing recommendation system,which has a positive impact on improving user satisfaction and the final turnover of the platform.The core idea of the most widely used collaborative filtering recommendation algorithm is to filter out redundant and irrelevant information by establishing the relationship between users and information,and provide users with content more in line with their interests.In practice,due to the large number and variety of users and projects,traditional collaborative filtering has problems such as too much computation and data sparsity.To solve this problem,many scholars have improved the traditional collaborative filtering.At present,the relevant research on the use of user interest information mainly focuses on extracting user behavior information,and providing recommendations by analyzing the characteristics of user interest preferences.In the actual user operation,there are a lot of time series behaviors,such as browsing records,purchase records,and even user operation information contained in the mouse click.These time series feedback information amount,feedback from the user's natural behavior performance,is closer to the real preference index.However,many time series recommendations ignore the historical and dynamic characteristics of user preferences,and user interests often change with time,so when considering user feedback information,it is also necessary to combine the analysis of time series to analyze the trend of development with time,and predict and recommend the user's next behavior with the direction and extension of the trend.The time series recommendation method mainly aims at the time sequence evolution characteristics of different data clusters,and constructs the corresponding prediction methods for prediction.In recent years,time series model is widely used in life and production,such as financial forecasting,traffic forecasting and so on.Many time series models are improved by combining artificial intelligence technology.On the one hand,they enhance the generalization ability of the prediction model.On the other hand,they can integrate more heterogeneous time series information to improve the prediction accuracy.Although there is a lot of research on time series prediction,it can be effectively predicted in the data with stable changes.However,it is not effective to deal with the wave series with fuzzy data characteristics,for example,the effect of the existing traditional time series data prediction framework is poor.In addition,in the recommendation system,a large number of users have different feature sequences,which are complex and difficult to normalize,so it is difficult to describe all user features with one rule.Therefore,this paper proposes a hybrid time series prediction method based on neural network and fuzzy clustering(htsrf)for the recommendation system with complex environment.By improving the fuzzy clustering method,the time series data with no obvious characteristics are classified,and the fuzzy logic relation group is constructed for prediction.By analyzing the threshold value of the fluctuation amplitude of the time series data slice,the series can be divided into small fluctuation and large fluctuation according to the fluctuation amplitude of interest,and the improved neural network gradient descent learning and the hybrid method model of fuzzy clustering method are respectively used for processing,effectively combining the advantages of the two methods to process the time series,and further improving the accuracy of prediction.