This aims to study Key Phrase extraction using Deep Recurrent Neural Network.
I have tried to develop on the work and learnings from the research paper - Key Phrase Extraction using Deep Recurrent Neural Network on Twitter Qi Zhang, Yang Wang, Yeyun Gong, Xuanjing Huang - 2016, Association for Computational Linguistics https://github.com/fudannlp16/KeyPhrase-Extraction
The joint layer concept comes from the need of doing joint processing of 2 sub-objectives, which will be combined into the main objective function. Computing one feature and passing it to other, will cause the error to propagate into further layers. Hence, this model is a 2-layer joint recurrent neural network.
The aim of layer 1 is to detect whether the input word at instance t, is a keyword or not.
Layer 2 focuses on keyword extraction. It assigns a label to the word that tells at what part the word would occur in a key-phrase.
The neural network was fed with Word Embeddings as input. Word embedding represents document vocabulary. It captures context of a word in a document, semantic and syntactic similarity, relation with other words, etc.