Quora questions pair duplication analysis using semantic analysis

Guided by- Ms. Safa Hamdare
Group Members
.
Quora Duplicate Question Pair
Detection Using Semantic Analysis
Name Roll No.
Jai Mulye 64
Anshul Pawaskar 87
Tannmay Redij 88
Akshata Talankar 89
St. Francis Institute of Technology
Department of Computer Engineering
Quora Duplicate Question Pair Detection using Semantic Analysis
1 28/05/2021

Content
● Introduction
● Literature
● Problem Statement
● Proposed Solution
● Work Flow of the system
● Algorithm with Implementation details
● Experimental Set Up
● Data Set
● Performance Evaluation Parameters
● Validation with Test Cases
● Results & Discussion
● Conclusion
● References
28/05/2021 Quora Duplicate Question Pair Detection using Semantic Analysis 2

Introduction
• What is Quora?
28/05/2021 3

Current Scenario:
Quora uses Random Forest technique to identify duplicate
questions.
Let’s look at two hypothetical questions:
1. Is it true that time flies like an arrow?
2. Do fruit flies like a banana?
There are two common words in these questions, flies and
like.
4

Let’s consider these
5

Literature
• The paper[1] explores the Transformer based
Universal Sentence Encoder which relies on
attention mechanism.
• The paper[2] introduces Deep Averaging Network
which performs well with neural networks that model
semantic and syntactic compositionality.
6
28/05/2021 Quora Duplicate Question Pair Detection using Semantic Analysis

Literature
• The paper cited [3] explores the two variants of
Universal Sentence Encoder- the transformer and
the deep averaging network (DAN).
• The paper cited [4] analyses several neural network
designs and their variations for sentence pair
modelling and compare their performance
extensively across eight datasets, including
paraphrase identiﬁcation, semantic textual similarity,
natural language inference, and question answering
tasks.
7
28/05/2021 Quora Duplicate Question Pair Detection using Semantic Analysis

Problem Statement
• On Quora, there may be people who might ask same
questions diﬀerently from an existing question. Solving
this problem will help to reduce the redundancy on the
platform and the manual task of identifying the questions
to match the correct answer for same. The task to identify
which questions asked on Quora are duplicates of
questions that have already been asked could be useful to
instantly provide answers of existing questions.
• A model created which can predict if the questions
entered are similar in meaning based on deep learning
approach using DAN & Transformer model.
28/05/2021 8

Proposed Solution
1. Pre Processing 3. Deep Learning Approach
(DAN & Transformer)
2. Sentence to Vector
Conversion (USE)
Fig 1: Workflow of the System

Work Flow of the system
28/05/2021 10
Fig 2: Architecture Diagram

Algorithm with Implementation
Details
28/05/2021 11
Fig 3: Algorithm

Algorithm with Implementation
Details
28/05/2021 12
Fig 4: Implementation

Experimental Setup
28/05/2021 13
Fig 5: Dataset[5]

Experimental Setup
28/05/2021 14
Fig 6: Model accuracy of
Transformer
Fig 7: Model loss of
Transformer

Experimental Setup
28/05/2021 15
Fig 8: Model accuracy of DAN
Fig 9: Model loss of DAN

Validation with Test cases
28/05/2021 16

Results and Discussions
28/05/2021 17
Fig 10: Browse Questions

28/05/2021 18
Fig 11: Post Questions

28/05/2021 19
Fig 12: Results by DAN Model

28/05/2021 20
Fig 13: Results by Transformer Model

Conclusion
28/05/2021 21
Model Embedding technique
F1-score
weighted average
F1- Score macro
average
Logistic
Regression
Word2Vec, Similarity
scores
0.66 0.62
Random Forest
Word2Vec, Similarity
scores
0.70 0.69
Table 1:Accuracy of machine learning models

Conclusion
28/05/2021 22
Table 2:Accuracy of Deep learning models (DAN & Transformer)
Model
Embedding
technique
Epochs
Training
accuracy (%)
Validation
accuracy (%)
Neural
Network
Universal Sentence
Encoder (DAN)
20 88.63 86
Neural
Network
Universal Sentence
Encoder
(Transformer)
20 89.16 85

Conclusion
• Deep learning models using sentence level
embedding outperform the basic classification
model.
• DAN Model sometimes under performs with the
questions having double negation.
• Transformer based Universal Sentence Encoder can
be used.
28/05/2021 23

References
[1] Mueller J, Thyagarajan A. Siamese recurrent architectures for learning
sentence similarity. In: Proceedings of the thirtieth AAAI conference on artificial
intelligence. (2016)
[2] Eneko Agirre, Aitor Gonzalez-Agirre, Inigo Lopez-Gazpio, Montse Maritxalar,
German Rigau, and Larraitz Uria. Semeval-2016 task 2: Interpretable semantic
textual similarity. In: Proceedings of the 10th International Workshop on Semantic
Evaluation (2016).
[3] Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones,
Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.
Advances in neural information processing systems, pp. 5998-6008. 2017. (2017)
[4] Cer D, Yang Y, Kong S-Y, et al. Universal Sentence Encoder for English. In:
Proceedings of the 2018 Conference on Empirical Methods in Natural Language
Processing: System Demonstrations. doi: 10.18653/v1/d18-2029 (2018)
[5] https://www.kaggle.com/c/quora-question-pairs/data
28/05/2021 24

28/05/2021 25
Thank you

Quora questions pair duplication analysis using semantic analysis

More Related Content

What's hot (20)

Similar to Quora questions pair duplication analysis using semantic analysis (9)

Recently uploaded (20)

Quora questions pair duplication analysis using semantic analysis