Chinese Multi-category Product Words Segmentation And Recognition Based On Electronic Commerce

With the rapid and vigorous development of the high technology, electronic commerce has played a dominant role on our working and life than ever before. Meanwhile, it provide an opportunity for offline payment to integrate with online payment, followed by the surging growth of transaction scale in the network payment industry. No matter which kind of media you use for shopping, online electronic commerce portal website or offline electronic commerce mobile application, it will absolutely relate to product search. In many cases, the segmentation for Chinese multi-category words is not accurate. Hence, the accuracy of part-of-speech tagging will directly affect recognition performance of Chinese multi-category words and results of analysis and processing.Through the study and comparison of traditional Chinese word segmentation technology, we determine to use the model of Conditional Random Field for the Chinese multi-category product words segmentation and recognition. The experiment find that the original feature template based on the conditional random fields is not totally applicable to the field of electronic commerce. Considering that the field of traditional Chinese word and electronic commerce attain the corpus from diversified characteristic and length of distance dependence, we take relative independence into account and add the unique characteristic related with electronic commerce. We also discovered that the distance dependence of feature template is approximated to normal distribution:the template get the extreme value is the optimal solution to the function. In this paper, we propose an effective method on how to recognize Chinese multi-category words in electronic commerce using Conditional Random Field and modified feature template.Experimental results show that our method remarkably enhances the accuracy of Chinese multi-category words recognition, especially the form of adjective, reduces the misunderstanding and improves the user experience of electronic commerce retrieval.