Assessing Information Quality Of UGC

UGC become more and more popular with the rise of Web2.0, which's main features is personalization, UGC application spawned from social networking, video sharing, blog to podcasts, UGC created a new profit model, and Formed a new business into market. But the UGC's information quality is uneven which can't be fully guaranteed, An objective evaluation of UGC by reasonable evaluation mechanism, can guide high-quality content to consumer, which can higher levels of consumer satisfaction, evaluation of UGC also good for development of UGC.This thesis established three-tier evaluation framework of the UGC information quality, evaluation model will be split into an object layer, dimension layer and measure layer. Model can evaluate the UGC by different forms of business model; ensure the UGC on the data collected in different ways can be compatible. Object layer's main tasks are the business forms and application background, business forms contain text/documents, pictures, video/audio, the program; application background contain two broad categories:"pull-style UGC " and "push-style UGC ". According to the definition of information quality, dimension layer split the IQ into the quality of the information in the form, content quality, information quality and effectiveness, and refinement indicators and quantitative problems of each category were established.In order to solve that each indicator is difficult to take in large-scale UGC, the evaluation measure layer's indicators should effectiveness and feasibility. Based on the interaction between users and content, Data Monitoring methods (automated monitoring methods, peer review Method, the user evaluation methods and so on), Measure level indicators include indicators of content, indicators of user interaction, web statistical indicators. All indicators are taken from the UGC user interactions model, and the Measure level indicators are Correspondence with dimension layer indicators; Measure level indicators and Data Mining Classification Algorithm form the UGC quality evaluation model together. UGC quality evaluation model is highly interactive and scientific operations.Base on Baidu Library, we analysis the UGC quality evaluation model, Experts quantify the dimension level indicators using AHP, and evaluation quality of content based on quality indicators. Experts assess the results as the decision attribute, and other conditions gathered together for classification mining property, both of them were papered for content evaluation.UGC quality evaluation model got more than 95% accuracy rate compare with expert evaluation. The ROC curves show that support vector machine is the optimal algorithm for UGC quality evaluation model. The merits order of other algorithms is decision trees, neural networks, Bayesian networks. Evaluation studies has concluded that the average quality of different types of content significant differences, the user evaluation scores and UGC are strong correlation.