Based On The Semantics Of Emotional Tendencies Text Similarity Computation

Chinese sentence similarity computation is an essential task, which is widely used in the Chinese information processing such as info-classification, info-search, info-filter and info-management etc. Chinese is a special and complicated language. How to compute the sentence similarity is one of the most important problem, which is also a hotspot, and very difficulty that people study for a long time.Traditional Chinese text similarity computations were based on words statistic, or words semantics, whereas computation basing subjective sensation such as attitude, viewpoint, and affectivity of author are less. This thesis does research on the text similarity computation basing on affectivity.Affectivity of author represents as appraisive of sentence. Sentence is the smallest element, which computes similarity basing on affectivity. This thesis discusses Sentence appraisive orientation estimation, appraisive degree computation and affection of text similarity computation.The main innovation achievements of this thesis are as follows:Firstly, the importance of computation basing on affectivity and the text environment, which the affectivity method fits for is presented. Human use language to describe objects, exchange information and express emotion. Human language is full of emotion. So when we process Chinese text information, we should not ignore the influence of emotion. Affectivity will do great effects to the similarity computation of two texts, which has same or similar topic.Secondly, do research on the method to estimate the appraisive orientation and algorithm to compute of appraisive degree. We construct base appraisive dictionary and sentence construction template basing on HowNet. We use these resources to construct appraisive estimate regulation and compute appraisive degree of sentence by acquiring words appraisive degree and identifying appraisive properties of sentence construction. According to the appraisive degree of two sentences, we can acquire the appraisive similarity of sentence by appraisive similarity compute regulation. Thirdly, during the research of Chinese sentence similarity computation, the similarity computation that we have studied is focus on three levels: Word, Sentence and Paragraph.It is based on the property of Chinese, which is the sentence is composed of words and paragraph is composed of sentence. Although three levels are different, from the similarity computation to its applications, it is a gradually process with close relationship as a whole. The improvement of method which is used to compute similarity of sentences and paragraphs basing on the sememe of How-net is presented. This thesis studied the influence of many properties such as sentence length, words number and paragraph length to text similarity estimate.Fourthly, taking the computer forensics system as the examples, we show the importance roles that affectivity has been in practice. Then, we carried out a series of experiments, and acquired preferable effects.