A Device-Similarity-Based Recommendation System in Mobile Terminals

Abstract—Smart Mobile device are becoming popular platforms for information accessing, especially when coupled with recommendation system technologies. They are also treated as key tools for mobile users both for leisure and business applications. Recommendation techniques can increase the usability of mobile systems by providing more personalized and interested content. In this paper, a novel personalized recommender system is proposed, focusing on Mobile Terminal (MT) similarities, such as brands, versions and types of Operating Systems. These similarities play a key role in filtering original recommendation data sets at the pre-processing stage. By calculating and comparing the Mean Absolute Error (MAE) values through 5-fold cross validation of the Slope One algorithm with/without optimizing data sets by device-similarity , the overall effectiveness and accuracy of the recommendation results are at least 20% improved in our experiment.

Main Activities:  The implanted system is consists of 2 modules, shown in Fig. 1 and throughout the design process for DSR, there are 3 vital parts, User-Item-Device model, MT similarity algorithm and the recommendation system. In order to calculate similarities between different devices, we collected some context information, such as brand, brand-version and Operating System (OS).

 Results: The DSR model could optimize and screen data by using context information of devices’ similarities. We also conducted experiments on two kinds of real data sets collected from mobile markets: ADS and WDS respectively. Then the experimental results are used to testify the feasibility of our model and verify its efficiency, compared with those of Slope One. Finally, the evaluation results of MAE indicate that our DSR system can improve the accuracy of recommending by at least 20%.

Leave a Reply