Abstract
The textual based information amount is rapidly accumulating Which stored electronically on our computers or Web. Any computer (laptop or desktop) is capable of accommodating enormous data amounts owing to the improvements in the storage devices.
Texts are included in text dataset and this dataset are unstructured. These unstructured data can be handled by text mining. The complexity and the considerable number for these data uncover numerous new capabilities to the analysts. Therefore, this work presents an enhancement of extracting useful patterns from text documents in the field of text mining using Pattern Taxonomy Model (PTM) and Levenshtein Distance Algorithm (LDA).
There are various methods to handle text documents. In this thesis, text mining system was suggested to overcome the problems that have occurred in term-based method and phrase-based method. The proposed system based on the behavior of LDA algorithm and PTM for determining the best accuracy of the extracted patterns with a short time and to prove that pattern based method is the best solution for text mining without any problems in the information extracted from the text.
The strength of the two algorithms (PTM, LDA) are tested using threshold values from 1 to 10 to get 1% to 10% of information in the text. The proposed system used "Openosis opinion dataset" and "Reuters 50_50 dataset" which stored in a file of ".txt" or text document
The results of this test obtained by comparing among values of four features which are (global probability, local probability, absolute support, relative support) for the text to get higher average accuracy.
The results of proposed system have been compared with other systems. The proposed system get (98.68%) average accuracy for Unigram grammar and (99.65%) average accuracy for Bigram grammar while a system that used the Levenshtein Edit Distance for automatic lemmatization for modern English achieved an accuracy of 96% for English language and the system that used the process of pattern evolving and pattern deploying get 62% of precision and 82% of recall. So, using LDA with PTM achieved a better results compared to other systems.
