Uncorrected Author Proof Journal of Intelligent & Fuzzy Systems xx (20xx) x–xx DOI:10.3233/JIFS-191933 IOS Press 1 Paraphrase identification using collaborative adversarial networks 1 2 Jafar A. Alzubi a,∗ , Rachna Jain b , Abhishek Kathuria b , Anjali Khandelwal b , Anmol Saxena b and Anubhav Singh b 3 4 a Faculty of Engineering, AL-Balqa Applied University, Salt – Jordan 5 b Department of Computer Science and Engineering, Bharati Vidyapeeth’s College of Engineering, New Delhi, India 6 7 Abstract. The paper presents a Collaborative Adversarial Network (CAN) model for paraphrase identification, which is a collaborative network holding generator that is pitted against an adversarial network called discriminator. There has been tremendous research work and countless examinations done on sentence similarity demonstration. Learning and identifying the constant highlights, specifically in various areas and domains is the main focus of paraphrase identification. It Involves the capture of regular highlights between two sentences and the community-oriented learning upon traditional ill-disposed and adversarial learning for common feature extraction. The model outperforms the MaLSTM model, which is the baseline model, and also proves to be comparable to many of the state-of-the-art techniques. 8 9 10 11 12 13 14 Keywords: Paraphrase identification, text classification, adversarial networks, LSTM, NLP 15 1. Introduction 16 Paraphrase identification is the undertaking of 17 deciding if the two sentences written in natural 18 language are similar in semantic meaning. If two 19 sentences have the same meaning, then they are 20 called paraphrases of each other. This method sig- 21 nifies a fundamental approach in various data mining 22 techniques. It can prove to be a vital part of applica- 23 tion in numerous fields such as plagiarism detection, 24 machine translation, and others [1]. 25 Detection of paraphrases typically follows two 26 methodologies: a supervised methodology mainly 27 focused on AI (ML) and profound learning 28 Calculations; and an unsupervised methodology 29 dependent on content like likeness calculations. ML ∗ Corresponding author. Jafar A. Alzubi, Ph.D, Associate Pro- fessor, School of Engineering, Al-Balqa Applied University, School of Engineering, 19117, Jordan. Tel.: +962 001 336 582 3417; E-mail: j.zubi@bau.edu.jo. algorithms that handle the identification task in 30 paraphrases as a standard classification issue with 31 syntactic or linguistic characteristics. 32 Most generic techniques measure sentences simi- 33 larities that are expensive and vulnerable to errors, 34 such as dependency parsing, based on practical 35 techniques and linguistic tools [2, 3]. To achieve 36 sequential hidden conditions for each word, the long- 37 term memory models (LSTM) [5] utilized by the 38 Siamese Recurrent Neuro-Network (RNN) [4] is one 39 of the well- known architecture. 40 GAN (Generative Adversarial Network) is pro- 41 found learning, unsupervised machine learning 42 technique proposed by Ian Goodfellow and barely 43 any different analysts incorporating Yoshua Ben- 44 gio in 2014. It’s an antagonistic system framework, 45 where the generative model is restricted by an 46 adversary: a discriminative model which can choose 47 whether an example originates from the model cir- 48 culation of information conveyance. Consequently, 49 the name Generative Adversarial Network. In GAN, 50 ISSN 1064-1246/20/$35.00 © 2020 – IOS Press and the authors. All rights reserved