The Research And Implementation Of The Opinion Mining Technology With Implicit Evaluation Content

In the era of increasing e-commerce websites,users'comments are constantly increasing.Users browse the data reviews of a product to get data support services to gain insight into the product and provide a reference when making user decisions.The merchants can guide the subsequent improvement of the quality of products and services by mining user feedback.The opinion mining technology about online commentary has been widely studied,and the extraction of evaluation objects in opinion mining is a concern of many researchers.The explicit evaluation object in which the comment attribute is clearly expressed has obtained more research.User insights with implicit evaluation content provide users and businesses with more comments.Therefore,this thesis combines the implicit evaluation content to research and apply the user opinion mining technology.The main work is as follows:1)An explicit evaluation object and an evaluation word extraction method are proposed.According to the different granularity of the evaluation object extraction task,it can be sorted into the extraction of the corpus-level and the sentence-level.Aiming at the corpus level extraction task,this thesis proposes a Biterm*Topic Model corpus-level evaluation object extraction model that combines semantic dependency and BTM.For the sentence-level extraction task,this thesis proposes a Weight~N-BiLSTM-CRF model based on BiLSTM-CRF combined with different weighted word embedding vectors and part-of-speech embedding vectors.Different extraction models are used to extract evaluation objects and evaluation words in different levels of text.2)Aiming at the prediction of implicit evaluation objects and the discrimination of emotional polarity in opinion mining,this thesis proposes a classification algorithm based on multi-input convolution for long short-term memory neural networks.In the proposed algorithm,the words and their corresponding part-of-speech tagging information transformed into a vector representation,and uses the vector as the input of the convolutional neural network.The convolutional neural network is used to extract the feature information and generate the feature matrix.The network performs feature learning on the feature matrix.The design is implemented to apply the algorithm to the emotional polarity discrimination and the implicit evaluation object prediction in opinion mining.The experimental results show that the Weight~N-BiLSTM-CRF model can improve the F-measure of explicit evaluation objects and evaluation word extraction tasks;a classification algorithm based on multi-input convolution for long short-term memory neural networks is better than the single-input convolution for long short-term memory neural networks.The prediction of the implicit evaluation object and the discrimination of the emotional polarity are improved in both the accuracy rate and the F-measure.