Clustering Rental Housing Information Using K-means On Mobile Platform

The mobile internet development accelerates the speed of data generated, promotes the advancement of data processing technology, like cloud computing, big data and so on. Various vertical search engine site develop rapidly, as general search engine gets massive amount of information and has a low query precision when searching mass data. Anjuke is a professional housing information search site, and provides a lot of housing information, specially, rental housing information is the largest and fastest update data among that. For the mass housing data, developer has been studying how to improve service quality and user visits. By counting user search, finding that user only interested in the first few pages for the numbers of result pages. It is worth researching how quickly and overall display mass rental housing information for user.It contributes to targeted browsing the location-based services that rental housing information scatters on the map with mobile platform map service. The screen presentting map is limited, especially mobile terminal is constrained by screen size, the single presentation data is more restricted. Exploring and solving the problem between rental housing density and map scale, proposed clustering rental housing information idea.It is need to do preprocessing for the data stored on the server before clustering. By data cleaning, data integration, data conversion and data reduction, the four steps process, reducing the amount of data processing computation on mobile terminal. Then making a detailed analysis at each stage of the rental housing information process, and making a experimental analysis on five million data’s preprocessing.Mobile terminal clustering the data coming from server, studying the related problem of central point and K value choice. And giving the steps of K-means clustering algorithm based on geographical location. Proposeing K-prototypes algorithm improved K-means for multidimensional mixed attributes under different map scale. For reducing calculated amount and data transmission, clustering new data again partially when map shift. Finally, analysis of the experimental performance of the algorithm. The effect of put the App in market also indicates this improve33%on user browsing.