In this era of textual data explosion on the World Wide Web, it may be very hard to find documents that are similar to the documents that are of interest to us. To overcome this problem we have developed a type of semantic signature that captures the semantics of target content (text). Semantic signatures from a text/document of interest are derived using the software package semantic signature mining tool (SSMinT). This software package has been developed as a part of this thesis work in collaboration with Sri Ramya Peddada. These semantic signatures are used to search and retrieve documents with similar semantic patterns. Effects of different representations of semantic signatures on the document classification outcomes are illustrated. Retrieved document classification accuracies of Euclidean and Spherical K-means clustering algorithms are compared. A Chi-square test is presented to prove that the observed and expected numbers of documents retrieved (from a corpus) are not significantly different. From this Chi-square test it is proved that the semantic signature concept is capable of retrieving documents of interest with high probability. Our findings indicate that this concept has potential for use in commercial text/document searching applications.
展开▼
