Analysis Of Users' Network Behavior On Vehicular WiFi Hotspots

With the increasing improvement on mobile phone applications of user experience and the screen of mobile phone becoming larger, mobile phones have their popularity on the aspect of internet surfing, leading to the heavy needs of mobility and broadband of internet applications. Meanwhile, as one of the main technologies in wireless broadband access, WiFi has advantages on mature technology, low cost, wide application and so on. Nowadays, many companies have combined public transport system with WiFi technology, aiming at bringing more value-added services for passengers and further establishing a "portal + platform + service" ecosystem based on public transport WiFi network and big data. Obviously, large amount of network data can be generated by passengers using bus on-board WiFi network, consisting of great research value. By mining the data, it's beneficial for us to understand behavior patterns of collectives, thus providing theoretical and technical support for community labels, advertising recommendations and operational managements. As for the above backgrounds, this paper analyzes the behavior patterns of passengers using bus on-board WiFi network and then proposes a behavior pattern model, aiming at understanding their behavior features and mining their latent behavior patterns.The research background of user network behavior analysis and related research status at home and aboard have been firstly introduced in this paper. Based on the data of passengers using bus on-board WiFi network, related features of those passengers, such as click frequency, flow consumption, phone brand, preference of websites and time interval of internet surfing, have been analyzed by utilizing distributed platform of Hadoop and data flow engine of Pig.Based on the above statistical analysis, 1000 of the most important websites have been extracted and then are labeled by utilizing APP classification method, thus the features reflecting user behavior category can be obtained and analyzed. After that, this paper implements the task of user clustering by integrating distributed machine learning tools of Spark MLlib and then extracts new features by utilizing TF-IDF algorithm in the goal of improving clustering results. The analysis based on the results has been proposed to discover four mainstream type models. Finally, related system framework hidden from the above work has been introduced, consisting of data pre-processing, data analysis and data mining modules.