The Extraction Of Mobile Product Coments Based On The Hownet

Product characteristics and product reviews have important reference function to users and merchants. So to extract this kind of speech of a text automatically and accurately is very meaningful. In this paper, we used Hownet to extract compound phrases first; and then finished the extracting work combining with the characteristics of texts in the automobile field.Compound phrases, as a universal language syntax phenomenon, with its strong structural stability, integrity ideographic single and less ambiguity characteristics, in recent years is gradually to the field of natural language processing researchers attention. Currently, the scope of study on the compound phrase focused noun phrase recognition complex and compound phrases keyword identification and treatment.This paper mainly studies how to use Hownet to measure the semantic relatedness between words, and based on this, completed the digging and tagging of compound phrases. Finally, we finished the product characteristic and product reviews extraction work based on factors such as the frequency of words. In the first part, according to the matching characteristics of Chinese words, the paper did the segmentation work. After segmentation, the paper eliminates the noise. After the segmentation processing, the stop words, punctuation and the single words which cannot constitute compound words were removed. In the second part, the paper optimized the measure of semantic relatedness algorithm first, and established a potential compound phrases collection. Then based on the HMM algorithm completed compound phrase marking. In the third part, combining with the text feature of the automobile field, based on domain knowledge base, factors such as frequency of words and parts of speech, finished product characteristics and product reviews extraction.When the theoretical study has been completed, there is a system implementation based on this theory. After using the F value on the performance of the system was tested, from the test results, the system basically achieve the goals.