Research On Fine-grained Sentiment Analysis Method Oriented To Scientific And Technical Resources Text Comments

The science and technology service industry is an important part of the modern service industry.Among others,professional science and technology resources are the prerequisite foundation and precious "gold mine" of the science and technology service industry."Review and evaluation" and "Scientific analysis and utilization" of scientific and technological resources are the key core tasks and inevitable trends of China's scientific and technological services.With the development of "Internet +" services,the willingness,breadth and depth of technology resource producers and users to participate in the evaluation of technology resource applications are increasing,and the value of their comment and behavioral data has also been continuously explored and utilized.Which has become a hot topic of sentiment analysis research.Compared with coarse-grained sentiment analysis,fine-grained sentiment analysis can not only analyze the sentiment polarity of user comments,but also obtain more finegrained and multi-faceted emotional content and effective information,and it is more in line with resource demanders' needs.Therefore,the fine-grained sentiment analysis method for "text comment data" of science and technology resources has also become a hot trend in the research and analysis of science and technology resource comment analysis methods.To this end,this article focuses on many tasks proposed in the national key R&D project topic "Distributed Resource Giant System and Resource Collaboration Theory"(Number: 2017YFB1400301),including: Aiming at the innovative "Internet +" environment for sharing technology resources,to study Science and technology resource sharing methods and cross-industry distributed technology resource search,analysis,matching,evaluation and optimization methods,etc.Based on the resource text review data of Wanfang Science and Technology Service Platform,Ningbo City Science and Technology Information Institute Public Service Platform and Internet Unstructured Technology Supported,focus on researching fine-grained sentiment analysis methods for text review of technology resources to support search,matching,and analysis of technology resources.The main research contents of this article are as follows:(1)Based on the unstructured science and technology resource text review data of Wanfang Science and Technology Service Platform,Ningbo Science and Technology Information Research Institute Science and Technology Resource Platform and the Internet,focus on the main features and problems of existing science and technology resource text review data,and The status and characteristics of sentiment analysis for comment texts.On this basis,focusing on the two parts of aspect word extraction and aspect-level sentiment classification,the realization plan of the aspect-level fine-grained sentiment analysis technology for text review of scientific and technological resources is designed.(2)According to the characteristics of multiple coverage,strong professionalism,and wide field coverage of the text review data of scientific and technological resources,and the problem of the lack of context semantic features of the input word vectors of the existing methods,considering the advantages of combining multi-task learning.This paper proposes an aspect word extraction method model based on multi-task language model learning.This method adds a language model learning layer on the BiLSTM-CRF network model in order to enable the model to express the semantic features of the generated word vectors and reduce the over-fitting phenomenon for specific tasks.Finally,this paper verifies and analyzes the validity of the model through experiments.(3)In view of the existing methods that use coarse-grained attention mechanism in aspect-level sentiment classification,which lacks the full mining of the interactive relationship between aspect words and comment sentences,resulting in the problem of information loss,This paper proposes a multi-level attention mechanism-based A neural network model for aspect-level sentiment classification.In addition,in view of the shortcomings of the existing two aspects of word position feature acquisition methods,This paper proposes a new method,the position symbol method(PS),which achieve a simpler and more convenient implementation process and good results.Finally,this paper verifies and analyzes the validity of the model through experiments.(4)On this basis,carry out the application verification of scientific and technological resource text reviews,and apply the method proposed in this article to the sentiment analysis of scientific and technological resource text reviews.By comparing with the existing methods,verify the effectiveness of the proposed method in practical application.