Vocabulary Correlation Calculation Based On The Theory Of Resonance

The research of lexical relatedness is a basic subject in the field of Natural Language Processing. In the traditional work which concerns measuring lexical relatedness, evaluating semantic relatedness between a pair of words is viewed as main research topic. Furthermore, there is an assumption in related work: the words took into consideration to measure semantic relatedness always co-occur in the same text window. However, the events of the Internet are constantly changing. The movement of the content of the Internet news is thought to have occurred quickly or slowly. The words corresponding to different events may not appear in a same text.In this thesis, I put forward a kind of remote words remote correlation theory which is not based on co-occurrence. This kind of remote correlation is based on resonance theory, which is similar to the object of analogy in physics. We want to convert the vocabulary association study to the study of vocabulary fluctuation. Finally, we can found one or a few kinds of words which contain the remote correlation in some field by this theory. Related tools and method can be used to calculate the similar specific correlation between related words.According to word frequency statistics and its varying tendency with the time was obtained. The frequency-time waveform is displayed in the rectangular coordinate system. This thesis deals with frequency-time waveform by using Fourier transformation and wavelet analysis, which are two kinds of waveform processing tools. We get the results of Fourier components and the characteristics of the wavelet coefficients as the basis of lexical relatedness computation.After representing the words’ features as frequency vectors and words’ feature vector is clustered. The experimental results show that, our method can not only find out the traditional sense of the correlation between a pair of words, but also can find the remote words associated with fixed intervals. It makes sense to use the vocabulary of long-range correlations to predict the occurrence of some event, studies the correlation between events retrieval.