Research And Realization Of The Method Of Identifying Elements Of Explanatory Opinions Based On Multi-feature Fusion

More and more people are posting reviews of some products on the Internet platform,which we call User-Generated Content(UGC).How to effectively help people to identify useful information is one of the studies that are of increasing concern in academia and industry.The identification of Chinese explanatory opinions is to study how to accurately identify the elements of the user-generated content.The Chinese explanatory commentary element recognition task refers to the use of Nature Language Processing technology to accurately identify the opinion attribute,opinion comment,and opinion interpretation information in the comment statement.Based on the linguistic features of Chinese product domain reviews,this paper studies the identification of Chinese explanatory opinions based on the corpus of Chinese explanatory ideology.This article will study from the following three aspects:(1)LSTM-based explanatory opinion element recognition: LSTM-based methods can identify attributes,comments,and opinion comment interpretation segments,but do not utilize tags interactions,and the CRF framework can enhance the identification of adjacent tags by the current tag.In order to improve the accuracy of opinion element recognition prediction,this paper uses the combination of LSTM and CRF framework.Experiments have shown that elements components can be effectively identified.(2)Interpretative opinion element recognition based on pre-training model: CRF only uses the information between the prediction tags,while the Chinese commentary,the context information of the word,and the semantic information of the word in the sentence can be used as features.Pre-training has been validated in many studies,and features related to words in the constituent elements of the opinion can be obtained,and the accuracy of the prediction can be effectively improved.In this paper,the two pre-training methods of word2 vec and ELMo are used to extract the context information of the words in the feature components and the semantic information of the words in the feature components in the whole segment.Experiments show that adding these features separately can play a role in the effect of feature recognition.(3)Interpretative opinion element recognition with multiple features: The contextfeature and semantic feature can be used to achieve certain effects.In order to further improve the effect of feature recognition,the feature and word feature of the fusion element component and word are used.The features associated with the word are extracted by the pre-training model;the features of the word include the word as a simple feature and the stroke feature of the word,and the stroke feature of the word can be useful for identifying the element component in which the data is sparse.The fusion of multiple features is the combination of these features as the input of the framework of LSTM and CRF.Experiments show that the method of merging multiple features has certain effects.