Method Of Opinion Target Extraction Combining Rule Template And Recommended Systematic Method Research

The main research topic of this thesis is the opinion target extraction,that is,the recognition of a text in the assessment of the goal.The opinion target extraction is fine-grained element extraction,and fine-grained is different from the traditional document-level emotion classification,the fine-grained aims to make subjective text data more In-depth opinion extraction and classification work,which can be widely used in product review mining,business intelligence,social public opinion analysis,and recommendation systems.There are roughly two broad categories of methods for evaluating objects:rule-based templates and statistical machine learning.The thesis combines these two methods to significantly improve the effect of the opinion target extraction.In order to make the opinion target extraction have better results,first construct a custom word segmentation dictionary in the preprocessing to solve the problem of inaccurate word segmentation.Five kinds of syntactic patterns are designed and formulated.According to the established rules,the potential opinion target in the corpus are extracted,and then the actual opinion target is filtered out through the dependency relationship in the syntactic analysis.Based on the selection of words,part-of-speech features,emotional word features,and sentimental words,there are features that depend on the relationship between the verbs and the verbs and the actual opinion target obtained by rule template extraction,the CRFs model is used for training and recognition.Making full use of the machine learning model has a very high accuracy and the rule template has the advantage of a high recall rate,thereby improving the evaluation effect of the opinion target.The main research contents of the opinion target extraction include the following four aspects:(1)In the preprocessing stage of this corpus,a custom user word segmentation dictionary,a sentiment dictionary,and a verb dictionary were constructed to prepare for follow-up experiments.(2)Firstly,the opinion target is extracted according to the rule template method.As a feature of the CRFs model,the rule template and the machine learning method are combined,so that the evaluation effect of the opinion target is significantly improved.(3)In order to find the best feature template window size under the CRFs model,this thesis selects different feature combinations on experimental data sets of different fields under 2-7 window sizes to see the experimental contrast effect and select the best Feature template.(4)Under the CRFs model,compare the different features in the data field of the mobile phone field,and observe the influence of each feature on the opinion target.(5)The method of this thesis is compared with the mainstream rule template method and machine learning method,which further shows that this method has a good effect on the opinion target extraction.This thesis also introduced a recommendation system method research,which is a specific application of the opinion target extraction.Based on the mutual information algorithm,the analysis of data from three aspects,including the user's historical behavior of goods,product titles and product label information,comprehensively consideration of three aspects,from the hit rate and coverage rate to determine the quality of the recommendation system.Experiments have proved that the recommendation system based on mutual information has achieved good results.The user's recommendation on the historical behavior of the product in this section includes the user's review information on the product.It is a specific application for the opinion target extraction.The opinion target and the opinion term are extracted through the review information,so as to gain insight into the user's demand for the product,then make a corresponding recommendation.