Information Mining Of Online Product Reviews' Emotion Based On Combination Of Multi-features For The Method Of SVM And Network

With the popularity of the Internet and the rapid development of electronic business,people increasingly prefer to shop online.Compared with offline shopping,online shopping has portability,help people save their time,was less affected by time and space,and other characteristics.Before consumers purchase goods online,a majority of them will browse the comments below commodities information generally,and they will evaluate this goods or service after they purchase some.The appearance of online product reviews has changed the time when enterprises improve their products.Traditionally,enterprises will improve their products before that products leave beltline.Now,after consumers use their products,they will get feedback information from consumers on their products,maybe before products are manufactured,they can understand the real needs of consumers in advance.Lastly,online product reviews help enterprises understand consumers,improve the quantity of products.Some researchers used machine learning to calculate the sentiment score of words,however,this paper paid more attention on the sentiment orientation of texts,namely,recognized classification of text sentiment orientation is positive or negative.The level of text data processed in this paper is clause.Finally,svm and probabilistic neural network were used to identify the sentiment orientation of clauses,then their results were compared.Probabilistic Neural network was used to forecast the sentiment orientation of clauses,product attributes were abstracted and classified.The distribution of emotion was obtained in the classification of product attributes.First,HUAWEI honor 4X mobile from Amazon website,for example,the rules which were used to capture online product reviews were set,then octopus data collector was used to crawl online reviews data.The data was processed by vectorization.Valid clause was extracted and divided.Then,stop words were removed from valid clause.According to some directories,emotional words,negative words,degree adverbs and special symbols were extracted.Then,text vectors were built by these features,svm and probabilistic neural network were used to build models,and were verified the performance on models.This paper judged neural network whether it could be used to recognize text sentiment orientation or not.In each method,according to different combination of features,there were five groups experiments,analyzed results and found the roles of features on text emotion recognition.Last,experimental results showed that emotional words and negative words which were from valid clause had great effect on text sentiment orientation,but degree adverbs and special symbols worked effectively little.Secondly,from the two aspects of accuracy and runtime of model,probabilistic neural network was fit to recognize text sentiment orientation.Then,the model of neural network was used to classify and predict experimental data,product attributes were abstracted and classified.The distribution of emotion was obtained in the classification of product attributes,these experimental results showed that this mobile phone had poor performance in camera and screen.Companies could improve the next generation of these products in the following two aspects.