International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 08 Issue: 02 | Feb 2021 www.irjet.net p-ISSN: 2395-0072 © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2147 Marathi Text Summarizer Using Deep Learning Model Shruti Bhoir 1 , Tanvi Hule 2 , Deepali Kadam 3 1 BE Student, Information Technology, Datta Meghe College of Engineering, Airoli, Maharashtra 2 BE Student, Information Technology, Datta Meghe College of Engineering, Airoli, Maharashtra 3 Asst. Professor, Information Technology, Datta Meghe College of Engineering, Airoli, Maharashtra ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract – Text summarization is a technique which converts the original text into short text by selecting important sentences without changing its original meaning . It is difficult for human to convert text manually. We present technique for extractive summarization of articles for Marathi language, in which it will consists of feature extraction, selecting important sentences, paragraphs etc. from the original document and catenating them into shorter form using deep learning model. In this system, we develop system in two stages. First stage is Summarization of Domain Specific Marathi article. In Second stage we will extend our model for generic article will be tested on various Marathi inputs. Such a summarization technique is known for English articles, and doing it for Marathi news is the novel part of the work. Key Words: Marathi article, Summarization, Extractive summary, Feature extraction ,Deep learning. 1.INTRODUCTION Automatic Marathi text summarization is technique of shortening the original text into shorter form which will give exact meaning of the original text. Summarization can be classified into two groups: extractive and abstractive summarization. Maximum of summarization systems are for English and other languages. For Marathi language, automatic text summarization systems are less. There is very less work done on Marathi summarization systems. 1.1 Problem Definition The overall scope of the project is to have a deeper knowledge of the techniques in Machine Learning, Deep Learning and Data Analysis in order to generate concise summaries of long texts, which lets a user see a summary. The scope also involves understanding of why Machine Learning are successful at phrasing sentences and how they treat some input words more important than the others by assigning the appropriate weights, and have a better overview internally of how Machine Learning. We chose Marathi over all other languages because there are no projects or summarizer created on Marathi language yet. We present technique for extractive summarization of articles for Marathi language, in which it will consists of selecting important sentences, paragraphs etc. from the original document and creating a summary out of it. 2. PROPOSED SYSTEM The summary produced by summarization system allows the user to easily understand the content of original documents without having to read each the whole document. 2.1 Abstractive and Extractive: A. Abstractive: Abstractive summarization consists of understanding the source text by using linguistic method to interpret the text and expressing it in own language. B. Extractive: Extractive summaries involve extracting relevant sentences from the source text in proper order. The important sentences are selected by applying statistical and language features to the input text. 2.2 Modules The proposed Marathi text summarization method is extraction based. Fig -1: Design of Module A. Pre-processing: In pre-processing step the stop-word is removed, stemming and breaking the input document into a collection of sentences. The punctuation marks, characters like ; , ――: ()[]{} space character, tab space are removed. We will eliminate these words from text. B. Stemming: In stemming, a word is split into its stem and root word. A stemmer algorithm involves removing suffixes using a list of frequent suffixes. Pre-processing output is taken as input to stemming . It is further subdivided into two parts: Root verification, Suffix removal.