Paper Title
Joint Intent Detection and Slot filling by Integrating Word embeddings and Topic Modelling in Conditional Random Field

Abstract
Intent recognition is the principle component of every spoken language understanding module.In this paper the primary aim of our research are two significant tasks of natural language processing viz. intent recognition and slot filing. In this paper we introduced a robust hybrid framework consisting of a joint model of sentence embeddings from Word2vec and topic modelling from Latent Dirichlet allocation.The features extracted from the hybrid (Word2vec + LDA) model were used to construct an overcomplete dictionary for conditional random fields to build a predictive model for joint intent detection and slot filling in text.We also performed data integration from multiple social online forums such as facebook, twitter ,quora and some other blogging sites.We prepared a consolidated dataset of utterances encoded in Inside Outside Beginning (IOB) represents a centralised repository that could be a valuable resource for researchers.The experimental results on the benchmark datasets ATIS and SNIPS showed that our proposed model has a significant absolute gain of up to 7 points in F-score and an accuracy 98.5% when compared to the baselines. Keywords - User-intent :natural language processing: spoken language understanding: latent dirichlet allocation: Word2vec: text classification: feature learning : conditional random field.