Filtering operation in cross-document linking
Elena Mitocariu · 2014
In this paper a filtering operation for cross-document connections is presented. Different texts could be connected to each other if they refer to the same entity. Linking different documents only in terms of overlap leads to too many False Positives. The method proposed in this paper has as start point Centering Theory (CT). A list of Principal Centers (Cp) is created for each sentence in the document. This list filters the results in cross-document linking. A bigraph representation is proposed to highlight the connections between texts. A score for classifying the topics is also presented. The score is calculated based on entities occurrence frequency in the whole document. Such an approach eliminated some of the False Positive results (Fps).