Article Preview
Top1. Introduction
In the modern era, utilization of digital content has raised dramatically at all walks of our life, for example- social media: facebook (Jain et al., 2018a), twitter (Jain et al., 2018b); newswire articles (Jain et al., 2014), web corpora: health corpus (Jain et al., 2018c), tweet corpus (Jain and Arora, 2018); web advertisements (Jain et al., 2013), question answering system (Verma et al., 2020) and many more. The dire consequence is a heavy flood of information over the internet. Thus, lays a necessity to build an automated system to summarize the text that provides a meaningful and concise summary.
Earlier, the text summarization task primarily involved statistical features (Meena and Gopalani, 2016) to work upon. Later on, morphological analysis (Almazaydeh, 2018) is also done due to availability of new technologies and labeled data. Now-a-days, the text summarization process has reached to a mature stage for English (Saggion and Poibeau, 2013; Gambhir and Gupta, 2017) and other foreign languages (Gunawan et al., 2017). However, for Indian languages the text summarization is a way behind as several language based challenges persist that are unfold in this paper. Our aim is to carry forward the Text Summarization (TS) task for the Punjabi language (Gupta and Kaur, 2016) which is still in its premature stage because of scarcity of the labeled data and other constraints.
Punjabi is the third most spoken language in the Indian subcontinent with more than 100 million native speakers around the world. It is an official language of the Punjab state which encompasses northwest India and eastern Pakistan. It is the most widely spoken language in Pakistan, second/third language used by around 30 million people in India. In addition, Punjabi is a minority language in several other countries where Punjabi people have migrated, namely- United States of America, Australia, United Kingdom, and Canada. The Punjabi language comprises of canonical word order of Subject Object Verb (SOV), also contains postpositions; distinguishes gender- masculine/feminine, number- singular/plural, case- direct/oblique. The major writing system is the ਗੁਰਮੁਖੀ ‘Gurmukhi’, a Punjabi script. However, very fewer efforts are being made in the field of computer technology towards the development of the enriched Punjabi language.
Text summarization (Al-Zahrani et al., 2015) is one of the vital Natural Language Processing (NLP) (Jain, 2019) tasks that mainly consists of two phases- pre-processing and processing phases. In the pre-processing phase, various keywords & clauses are taken care using linguistic and statistical features. In the processing phase, two techniques- extractive and abstractive summarization are used. Extractive summarization (Lins et al., 2020) is a shallow technique where an extracted text is used as a summary and is comparatively easier to implement. Abstractive summarization (Song et al., 2019) requires deep understanding and analysis of a given text which is comparatively more complex to implement. In the abstractive summary, new concepts and expressions determine the original text in a concise form to convey relevant information, henceforth, summary may not contain the identical sentences as in the original document.
In this paper, text summarization task in Punjabi using Particle Swarm Optimization (PSO) (Al-Abdallah and Al-Taani, 2017) algorithm is proposed which induces a meaningful short summary out of a large web text- Unicode encoded Punjabi text. PSO is one of the most powerful bio-inspired optimization techniques which looks into the bird-flock/fish-school concept that is precise and easier to implement. The PSO is based upon intelligence which has neither overlapping nor mutation calculation issues as in the Genetic Algorithm (GA). PSO helps to predict that among a given set of solutions which one is the best solution. The search is carried out by extremely high-speed particles. It updates particle position and velocity at the end of iteration, so that during the development of generations, the personal best solution of the particle and the global best solution are updated. Calculation within PSO is quite simple as it contains greater optimal capability which can be fulfilled with ease. PSO adapts the real number code which is determined by the solution itself. Thus, PSO based text summarization is quite suitable to extract summary sentences from an input Punjabi text document, grades the sentences, and actual summary is executed as the topmost summary.