The Extraction Of Micro-blog Evaluation Object Based On Semantic Feature

With the development of internet interactive technology,the network has become a new exchanging platform which produced massive text data. So the researching of sentiment analysis on these data develop rapidly.After ten years of development, sentiment analysis researching has become more detailed and deeper,while its study object is gradually transferred from chapter to sentence.In recent year, micro-blogging has become the most popular carrier of review information and the information is growing rapidly, and its value and urgency of research has become increasingly apparent.However,due to the micro-blogging sentence is irregular comparing to the traditional news statement,it makes research more difficult.Because of this,this paper select micro-blogging text as the study object.At the same time,select the evaluation object extraction which is the subtask of emotional elements extraction as researching task,then explore the new ideas of micro-blogging evaluation object extraction.Specific researching content includes the following two parts:1)For the problem that structure of micro-blogging sentence is irregular which leads to inaccurate of speech tagging and dependency analysis, then making the effect of micro-blogging evaluation object extraction affected.We propose a method that build a Maximum noun phrase (MNP) recognizer which used to simplify sentence structure so as to improve evaluation object extraction results. By identifying the Maximum sentence noun phrase and adding it to the user dictionary, makes sentence structure of micro-blogging corpus more concise after word segment. Comparing the experimental results of micro-blogging evaluated object extraction that add user dictionary and not, founding that the evaluation object extraction task can achieve better results in experiment that adding user dictionary. So micro-blogging maximum noun phrase(MNP) identification can contribute to the evaluation object extraction results.2)The effect of micro-blogging evaluated object extraction based on Conditions random field depends on the selecting of feature, while traditional lexical features only consider the word、speech, and ignore the semantic features of word, using semantic features to find evaluated object more in line with people’s thinking mode. Therefore, this paper presents four semantic features which are evaluation insulation word evaluation trigger word、evaluation digestion word and assessment point word. Through comparing the results of micro-blogging evaluation object extraction based on different combination of feature,found that some semantic features can greatly improve the results of micro-blogging evaluation object extraction. Finally find the optimal feature set through identify and combine the feature, and its F improve4.4percentage points comparing to the evaluation object extraction based on basic feature.