A Hybrid Approach of Text Summarization Using Latent Semantic Analysis and Deep Learning

In the current scenario of Information Technology, excessive and vast information is available on online resources but it is not always easy to find relevant and useful information. Along this issue, the paper is presented a method on extractive single document text summarization using Deep Learning method - Self-Organizing Maps (SOM) which is an unsupervised method and Artificial Neural Networks (ANN) which is a supervised method. The work involves investigating the effect of adding mapped sentences from SOM visualization, and re-training the inputs on ANN for ranking the sentences. In individual experiment of the hybrid model, a different mapping of SOM is added to the ANN network as input vector. Hybrid model uses Stochastic Gradient Descent update set of parameters in an iterative manner to minimize the cost function. In addition, using back-propagation weight is being adjusted for the input vector. The empirical results show that the hybrid model using mapping clearly provides a comprehensive result and improves the F-score on average 5% on ROUGE-1, ROUGE-2, ROUGE-L and ROUGE-SU4. This novel method has been implemented on different documents, which are publicly available on Opinosis Dataset. The ROUGE toolkit has been used to evaluate summaries which are generated from the proposed model and other existing algorithms versus human generated summary.