Research On Visual Recommendation Technology Oriented To Data Features And User Preferences

For data visualization has unique advantages in displaying data characteristics,data visualization analysis has become an indispensable part in the process of the mining and application of massive data.However,due to the explosive growth of data nowadays,with the increasing columns and complexity of data sets,traditional tools which require users to manually configure visualization,are hard to meet the needs of users to mine information from massive data effectively.Recently,Researchers have paid much attention to the visualization recommendation technology that automatically analyzes the visual feature of data sets,in order to improve efficiency and accuracy in data visualization analysis tasks.Our research focus on data features and user preferences,and pays attention to view interestingness measurement,personalized recommendation schemes,recommendation efficiency optimization.Our main contributions are as follows:(1)Based on analyzing the defects of the existing data-characteristics-oriented visualization recommendation technology from multiple cases,we propose three new dimensions to measure view interestingness,including 1)the difference between the trend of the view and the average trend;2)the fluctuation among the groups in the view;3)the easy-to-understand level of the view,which are quantified as outlier utility,variation utility,applicability utility respectively.Together with the improved deviation utility,they are linearly combined to form the multi-objective utility function to synthetically evaluate the interestingness of views.The experimental results show that the proposed utility function is able to dig up high-value results effectively by making up for the shortcomings of existing works.(2)In existing visualization recommendation systems,the weights corresponding to different utility are pre-allocated by the system or users' manual configuration,which leads to inaccuracy preference,poor flexibility and high barrier to use.To address various users' task target,we propose a weight fitting scheme that calculates weights according to user's feedback on a few view samples.Experimental results show that the proposed weight fitting scheme can make weight allocation effectively and accurately according to users' preferences,which meets the need of users' personalized recommendation while reducing the difficulty of the use of visual recommendation systems.(3)In order to improve the recommendation efficiency,a pruning strategy is proposed according to the cost distribution feature of the multi-objective utility function.The optimization strategy tries to avoid unnecessary high-cost operations while ensuring the quality of the recommendation result.In the experiment,the optimization effect of the pruning strategy is verified,and various factors affecting the optimization rate and the change trend of the optimization rate are studied and analyzed.