Deep LDA : A new way to topic model
Muzafar Rasool Bhat, Majid Kundroo, Tanveer A. Tarray, Basant Agarwal · Journal of Information and Optimization Sciences · 2019
Probabilistic topic models like Latent Semantic Indexing (LSI), Latent Dirichlet Allocation (LDA) and Biterm Topic Model (BTM) have been successfully implemented and used in many areas like movie reviews, recommender systems and text summarization etc. These models however become computationally heavy if tested on humongous corpus. Keeping in view the vide acceptability of Deep Neural network based machine learning, this research proposes two deep neural network variants (2NN DeepLDA and 3NN DeepLDA) of existing topic modeling technique Latent Dirichlet Allocation (LDA) with specific aim to handle large corpuses with less computational efforts. Two proposed models (2NN DeepLDA and 3NN DeepLDA) are used to mimic the statistical process of Latent Dirichlet Allocation. Reuters-21578 dataset has been used in the study. Results computed from LDA are compared with the proposed models (2NN DeepLDA and 3NN DeepLDA) using Support Vector Machine (SVM) classifier. Proposed models have shown significant accuracy besides computational effectiveness in comparison to traditional LDA.