Research On The Som Based Automatic Text Summarization Approach

Along with the improvement of natural language process techniques, automatic text summarization has made a great progess since the end of 1950s. With the development and popularization of the Internet, the great value of automatic text summarization is present to people absolutely. More and more scientists or researchers begin to research the automatic text summarization from lots of different aspects further. For example, cognitive psychology, information science, computer linguistics, sociology etc. and create a number of new directions for automatic text summarization. So far, the methods of machine learning, neural networks, artifical intelligence have been introduced into the automatic text summarization, which makes the research of automatic text summarization into a unprecedented prosperous period.However, there are a lot of problems need to be solved for automatic text summarization. For example, the representation of global word sense information, the selection and fusion of text features, understanding of text discouse structure and automatic evaluation of automatic text summarization system are challenges which automatic text summarization faces.This thesis aims at applying the quantization model of word sense information, introducing Kohonen self-organization algorithm to select and integrate text features, applying latent semantic indexing method with sigular value decomposition technique to discover the latent relations between word sense information, and extracting summaries in the manner of word sense clustering, which improves the performance of the existing automatic text summarization system called Insunabs.This thesis extracts some summaries using the algorithm of Kohonen self-organization from the aspect of word sense clustering. But this thesis doesn't analyze the discouse strucure of text, so the summaries are not improved and perfected with other methods of rethorical structure analysis like cohesion and coherence etc.Generally speaking, the thesis includes the following parts:...