Discovery and dynamic prediction of user's interest based on ARIMA

User's interest is changing over time in online social networks. How to make use of the user's historical data to forecast the user's interest in the future and then to make some individual recommendations with higher accuracy has become a particularly important research problem. To solve this problem, we propose an interesting model based on Auto Regressive Integrated Moving Average (ARIMA) to discover the user's preference dynamically and combine the Collaborative Filtering(CF) to recommend user's preference hashtags. In order to verify our method, we choose the real world data from Sina Microblog which is the biggest social network in China in two years as the experiment data set. More specifically, the data is divided into 24 periods by month average and extract interesting themes by Sina-users Latent Dirichlet Allocation(LDA) of every period. Then, we compute the users similarity based on Cosine similarity. Thus, we can get the time series of the user's interest for dynamic prediction by ARIMA. As shown in the experiment results, our designed method can not only predict the user's preference dynamically and more accurately than the previous work, but also can improve the sparsity slightly by making use of the content of Sina Microblog and user's hashtag.