neural network

1. Anirban Sen, IIT Delhi
Manjira Sinha, Conduent Labs India
Sandya Mannarswamy, Conduent
Labs India

2.  ‘billions of voices’ participate in CQA forums
regularly
 Each has its own style of expression,
vocabulary or knowledge
 Results in wide lexico-syntactic gap
 Makes it hard to identify matching pairs of
question and answer
 Required to ease the load to retrieval and
answering similar questions repeatedly

3.  I write: does drinking co.ee increase blood
pressure?
 My father writes: Has there been any
known link between caffeine intake and
hypertension?
 A way to address the challenge is incorporate
vocabulary diversity in consolidated
representation

6.  Stack Exchange Data Set released by Doris
Hoogeveen, Karin M. Verspoor, and Timothy
Baldwin. 2015. CQADupStack: A Benchmark Data
Set for Community Question-Answering
Research. In Proceedings of the 20th Australasian
Document Computing Symposium (ADCS’15)
 Training 2K, Test 8K
 Balanced/Unbalanced;
 0-verlap, 2-overlap

8.  Incorporating essence of ‘multi-sentence’
input
 Experiment with attention based models
and recurring networks
 Look into different domains of discussions
separately
 Use of background domain knowledge