Research On Mobile Recommender System Based On Spatial-temporal Context Awareness

The development of communication technology has made it possible for people to go into internet age. This also means people have more access to the internet regardless of the time and location by using mobile devices. In China, the great size of mobile phone users can reflect not only the potiential of mobile internet market, but also the bright research prospect of related field. Due to the large size of users' behavior records, it makes sense to prompt the market ecnomy and change the situation of the internet by doing research in mobile internet field.This paper is based on the true user-item records provided by Alibaba Group. It includes geographical information that can only be found in mobile age. Mining the user data helps to find the rich information hidden behind the data. Applying random forest and gradient descend regression tree to this data set allows people to acquire the specific things they need in specific time and location.This paper first introduces the background of mobile recommender systems and compares mobile recommender systems to traditional recommender systems in terms of techniques they exploit. Then the summarization is given to provide more information about the research frame of recommender systems and their applications to real life. The paper attempts to tackle the problem from two aspects.1) In order to improve the precision of recommender system, an outlier detection algorithm is proposed in preprocessing stage. The algorithm first utilizes rough set theory to analyze the impact between the attributes and abandoned less important ones. Then it divides data into different cubes according to the distribution of data on every attribute. Only the cubes with high possibility to contain outliers need to be focused on. At last, through the calculation of angle-based outlier factor, outliers gain more probability to be detected. Compared to conventional algorithms, the experimental results verify the feasibility of the proposed approach in terms of both efficiency and accuracy.2) Another contribution of this paper is to take spatial-temporal factors into considearation and construct features. The random forest and gradient descend regression tree are exploited to propose basic models and a complex model. The experiments are conducted by using real world dataset and the model can perform well in both precision rate and recall rate.