Study On Specialty Of Chinese Text Review In E-Commerce

With the prosperous development of e-commerce market recently, more and more people were willing to purchase online and give feedbacks. Meanwhile, most consumers would like to read these comments as references before shopping online. However, rapid development of e-commerce market resulted in an explosively growing of the review data. It was a problem that how to extract the valuable comments from vast amounts of text information, which was the basis for decision-making of consumers, merchants and online purveyors. Based on above background, this thesis analyzed the contemporary researches of commodity text reviews, and evaluates the reviews value from a particular point of view — “professional degree”.In this thesis, we analyzed the up-to-data researches of Chinese text reviews in e-commerce and summarized related technologies of Chinese text processing. Benefitting from the information retrieval and ontology, we proposed a review calculation model of professional degree based on a professional concept hierarchy tree.Using transaction data of one B2 C e-commerce website in 2012, this thesis proposed a feature hierarchy algorithm based on the improved Gini coefficient and the products’ categories(GCF for short). With the help of GCF algorithm, we constructed a professional concept hierarchy tree(PCH-Tree for short). The GCF algorithm calculated feature value of every professional concept extracted by some rules, and label them if they meet the threshold value. Experiment result showed that the precision of GCF was high, it was effective.After constructing the PCH-Tree, this thesis elaborated three professional factors from the aspect of breadth, depth and intensity of reviews, which was respectively effective length, the scope of reviews based on PCH-Tree and the cohesion. Combined with these three factors, a professional calculation model of commodity text review(RPC-Model, for short) was proposed. This model calculated the nodes’ depth and distance in PCH-Tree as well as the effective length of reviews. It reflected the degree of text reviews specialty comprehensively.We selected 18,415,146 reviews and 115 commodity categories of a B2 C e-commerce website in 2012 as experimental data set, and collected professional scores of reviews from almost 100 people as verified data. The comparative experiments results show that our RPC-Model outperforms the Length Model and other classified Models, which verified our model was more effective. At last, we showed a simulation platform included two parts: the professional score calculation of Chinese text reviews and the maintenance of PCH-Tree.