Research On Quality Evaluation Method Of Online Commodity Evaluation Based On Bayesian Network

The online product review is a users' subjective or objective impression on online product service,which can assist users to make purchasing decisions,affecting their consuming behaviors,having Word-of-mouth marketing role.Meanwhile,owing to growing scale of online product reviews,the quality of which is uneven and users being difficult to filter useful information,with rapid development of e-commerce,the techniques to identify useful reviews become an important issue.However,there remain certain deficiency in this filed,such as lacking of further exploration of the effective characteristics of review quality,and for another instance,the deficiency of speculation of regression analysis and inflexibility of traditional classification methods,etc.Usefulness votes are annotations of network users to evaluate the quality of reviews,representing the standpoints of majority of network users,which are quantitative information,more objective.large amount of data,easy to obtain.so the quality of online product review was defined based on usefulness votes.The authors proposed evaluation model of online product review quality based on Bayesian network by counting multidimensional review characteristics such as meta data,appending review,grammar,emotion,readability,shop,and the quantity of product reviews,the quantity of usefulness votes of the product.We use BIC scoring function to evaluate the excellence of the structure,and use the K2 algorithm to find the structure of online product reviews quality evaluation model,and then,we calculate the model parameters by maximum likelihood estimation,and predicate the quality class of online product reviews with class weight,using clique tree inference algorithm.Through the real data sets,average precision rate,average recall rate and F-Measure index,the effectiveness of multi-classification model including two review characteristics,the total number of product reviews,the total number of usefulness votes of the product was verified.Then the efficiencies of modeling algorithm and inference algorithm are illustrated comparing the modeling time of different number of characteristics and evaluation time of different inference algorithms.At last,we designed and developed the Taobao product reviews crawling and processing prototype system to collect and process the necessary data set and the online product review quality evaluation prototype system based on Bayesian network to build the evaluation model of online product review quality to appraise the review information.