Paraphrase identification using features from word and sentence level embedding

Name: Niraj Pahari

keyword: Robotics and Computational Intelligence System Thesis

ThaSH: Eigenvectors

Abstract: ix, 98 leaves : illustrations, tables Paraphrase identification is the set of techniques to identify whether two sets of text have the same meaning or not. Paraphrase identification is the classification problem where semantic similarity measures can be used for the decision. Methods like bag of words are widely used in the field of Natural Language Processing (NLP) but these methods fail to capture the semantic meaning of the words. Word embeddings or representations are used to transform the words into vectors while capturing semantic meaning of the words. Word embeddings can be used to create the sentence embeddings. The word embeddings and sentence embeddings can be trained for one task and can be used in several tasks in NLP. In this thesis, the use of feature engineering in paraphrase identification task is studied. Features are extracted using the word embeddings and the sentence embeddings of the sentence. A custom vectorizer is defined to generate the vector of the sentence using the word representations and is found to be better performing than just taking the average of all the word embeddings in the sentence. The performance of the system when an ensemble of different features extracted from different methods is used, can improve the performance significantly since different methods capture the semantic meaning from different dimensions. Through the test on the SemEval 2015: Paraphrase in Twitter dataset, the proposed method shows significant improvement over the strong baselines using different techniques. The end model of the study is also used to develop multiple web applications and can be further extended in several other application areas

King Mongkut's Institute of Technology Ladkrabang. Central Library

Address: BANGKOK

Email: Lifelong@kmitl.ac.th

Name: Somyot Kaitwanidvilai

Role: Thesis Advisor

Created: 2021

Modified: 2023-01-16

Issued: 2023-01-16

วิทยานิพนธ์/Thesis

application/pdf

eng

DegreeName: Master of Engineering

Level: Master's Degree

Descipline: Robotics and Computational Intelligence System

©copyrights King Mongkut's Institute of Technology Ladkrabang

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	Niraj Pahari.pdf	4.68 MB	13	2026-05-26 18:16:01

ใช้เวลา

0.036288 วินาที