An Analysis Of Emotional Tendency Based On Food Comment

This paper, surrounding the key mission of emotion tendency analysis, firstly constructs distributed crawler system to crawl comment corpus of Dianping.com and Meituan.com; then compares and analyzes several performance factors such as feature selection, feature optimization and quantity determination of features etc to verify the superiority of analyzing comment emotional classification through regression algorithm of logistic; finally conducts measurement or quantization toward the catering comment intense degree through sentiment word, adjunct word, sentence structure and other factors. The specific research works are as follows:(1) Crawling method of catering comment corpus based on Hadoop. For the lower effectiveness of light engine crawling technology, the author prompts the distributed data crawling method based on Hadoop and constructs the distributed crawler system with combining of REDIS, HBase, Zookeeper and other relative technologies in the same ecosphere.(2) Catering emotional analysis and classification based on logistic regression. The thesis conducts optimization selection toward features and selects emotional title, verb, adjective and adverb as features. Then, the thesis compares and analyzes the performance difference among different methods applying in emotional classification through feature selection, feature optimization, feature number and other perspectives. The thesis mainly focuses on the feature optimization selection and feature number of emotional comment context, as well as the advantage of adopting regression classification algorithm in catering comment’s emotional classification.(3) The thesis, based on the combination analysis method of emotional thesaurus, ornament dictionary and sentence structure dictionary toward emotional analysis of catering comment, and through analyzing the classification method of catering comment emotion based on regression of logistic, finds that such methods which based on mechanic study can only conduct classification toward the emotion, namely positive emotion or negative emotion while having no measurement or quantization process that it would be hard to measure the intense degree of expressed emotion. Meanwhile, the thesis analyzes the limit of traditional method which is based on emotional thesaurus, as well as constructs more comprehensive positive dictionary, negative dictionary, punctuation mark dictionary, phrase dictionary, negation dictionary, adverb dictionary, conjunction dictionary and adversative dictionary. Ignoring such dictionaries can produce contrary polarity or can hardly reflect the emotional polarity of comment sentence accurately. As a result, the author regards that the traditional method which is based on mechanical study can only judge the emotional polarity but not the intense degree of emotional tendency and promotes emotion algorithm of emotional analysis based on the combination of emotional thesaurus, ornament dictionary and sentence structure dictionary. Then, the thesis adopts test to verify the effectiveness of emotional analysis method of catering comment based on the combination of emotional polarity thesaurus, ornament dictionary and sentence structure dictionary, as well as offers the score toward catering comment emotion, thus people can conduct ranking or searching through emotional tendency score, as well as can better serve the service information searching and emotional tendency analysis mission.