Title: Deep learning-based detection and prediction of trending topics from streaming data

Authors: Ajeet Ram Pathak; Manjusha Pandey; Siddharth Rautaray

Addresses: School of Computer Engineering, Kalinga Institute of Industrial Technology (KIIT) University, Bhubaneswar 751024, India ' School of Computer Engineering, Kalinga Institute of Industrial Technology (KIIT) University, Bhubaneswar 751024, India ' School of Computer Engineering, Kalinga Institute of Industrial Technology (KIIT) University, Bhubaneswar 751024, India

Abstract: Detecting and predicting trending topics from steaming social data has always been the point of active research area in business and research firms to take quick decisions, change marketing strategies and set new goals. Topic modelling is one of the excellent methods to analyse the contents from large collection of documents in an unsupervised manner and it is a popular method used in natural language processing, information retrieval, text processing and many other research domains. In this paper, deep learning-based topic modelling technique has been proposed to detect and predict the trending topics from streaming data. The online version of latent semantic analysis with regularisation constraints has been designed using long short-term memory network. Specifically, a problem of detecting the topics from streaming media is handled as the minimisation of quadratic loss function constrained by ℓ1 and ℓ2 regularisation. The online learning mechanism supports scalable topic modelling. For topic prediction, sequence-to-sequence long short-term memory network has been designed. Experimentally, significant results have been achieved in terms of query retrieval performance and topic relevance metrics for topic detection on our published dataset. For topic prediction, the results obtained in terms of root mean squared error are also significant.

Keywords: deep learning; topic detection; topic prediction; social media data.

DOI: 10.1504/IJRIS.2021.114631

International Journal of Reasoning-based Intelligent Systems, 2021 Vol.13 No.2, pp.59 - 68

Received: 24 Jul 2019
Accepted: 29 Nov 2019

Published online: 30 Mar 2021 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article