| In recent years,the identification and sentiment analysis of commodity evaluation has become a hot project in the field of Natural Language Processing.Compared with the general commodity evaluation,the fine grained commodity evaluation can reflect the experience of customers on purchasing goods more detailed and accurately.On one hand,the information can provide customers with the purchase of reference,on the other hand,it also can provide the basis for business to improve the quality of the goods and sale services to enhance the competitiveness of industry.The goal of identification and analysis on fine grained evaluation of commodity is to distinguish the emotional polarity of each side of the object in evaluation information and then reflect the commodity evaluation intention of the consumers precisely,therefore,this subject has higher practicability.The work of this paper was divided into two parts: synchronous identification of emotional objects and emotional words,and the tendency analysis based on the identification.In the identification stage of fine grained evaluation,we introduced the syntactic relation,the emotional factor and the clustering code information into the conditional random field model,on the basis of the simultaneous extraction of emotional objects and emotional words,we had experiments on three different data sets,the F1-score reached 94.16%,91.44%,91.41% respectively,on the mixed product evaluation data set,the experiment result showed that the accuracy rate was 95.43%,the recall rate was 91.61% and the F1-score was 93.48%,fully illustrated the effectiveness of the method.In the stage of tendency analysis of the fine grained evaluation,we used PMI algorithm to expand the HowNet emotion dictionary to solve the shortage of emotional vocabulary,and considering the tendency of the emotional factor pairs not only depend on the polarity of the emotional words,we introduced the negative information and the relationship between polarity of emotional words and polarity of emotional factor pairs into the support vector machine.The accuracy of mixed data in the bipolar analysis experiment of fine grained evaluation reached 98.15%,and in the unipolar analysis experiment,the F1-values of the positive and negative evaluation in mixed data were 98.86% and 91.69%,on one hand,it proved the rationality of the method,on the other hand,it showed the difference of processing mode of different polarities in the tendency analysis task.Finally,a system for evaluating the tendency of fine grained evaluation was designed and implemented,it can extract the elements of the commodity evaluation information which user input,obtaining the matching results of affective object and emotional words to do the tendency analysis and show it to the user. |
