Change Detection of Text Documents Using Negative First-order Statistics

We present a probabilistic method for change detection in text documents based on the biologically motivated principle of negative selection. Compared to standard checksumbased analysis, our statistical approach is able to locate and approximate the magnitude of changes. Further, the detection process can be distributed to any number of independent nodes resulting in a fault tolerant system. The negative representation of information also makes it possible to protect the privacy of the analyzed data due to the difficulty of reversing the information of the non-self detectors. An experiment with a collection of Wikipedia articles is used to analyze the length of the required negative description compared to the length of the document.