Relation identification for reasoning
Relation extraction is a very important research area in Natural Language Processing. This thesis mainly concentrate on identifying cause-effect relation which can be used in various fields like question answering and medical science. A relation classification system is built in the thesis to ach...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
2018
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/75959 |
| _version_ | 1811685101720829952 |
|---|---|
| author | Li, Linjie |
| author2 | Mao Kezhi |
| author_facet | Mao Kezhi Li, Linjie |
| author_sort | Li, Linjie |
| collection | NTU |
| description | Relation extraction is a very important research area in Natural Language Processing.
This thesis mainly concentrate on identifying cause-effect relation which can be used
in various fields like question answering and medical science. A relation
classification system is built in the thesis to achieve the target.
The whole system consists of two parts. The first one is text representation. An
accurate text representation is key to the performance of the whole classification
system. Two methods are used in this part: traditional Bag of Words and Word
embedding. Different types of word embedding methods are also compared. The
second part is classification, results of word embedding can be further used to extract
features and do the classification based on Neural Networks. Two popular structures:
Convolutional Neural Network and Long Short Time Memory are implemented and
compared.
Experiments show that using the combination of Word embedding and Neural
Network based classification performs much better than using traditional Bag of
words to represent text and do the classification directly. The distinguished
performance of CNN in solving relation classification problems are shown by
experiments. Some methods are also taken to improve the performance of
CNN-based structure in order to achieve the best classification results. |
| first_indexed | 2024-10-01T04:39:10Z |
| format | Thesis |
| id | ntu-10356/75959 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T04:39:10Z |
| publishDate | 2018 |
| record_format | dspace |
| spelling | ntu-10356/759592023-07-04T15:56:19Z Relation identification for reasoning Li, Linjie Mao Kezhi School of Electrical and Electronic Engineering DRNTU::Engineering::Electrical and electronic engineering Relation extraction is a very important research area in Natural Language Processing. This thesis mainly concentrate on identifying cause-effect relation which can be used in various fields like question answering and medical science. A relation classification system is built in the thesis to achieve the target. The whole system consists of two parts. The first one is text representation. An accurate text representation is key to the performance of the whole classification system. Two methods are used in this part: traditional Bag of Words and Word embedding. Different types of word embedding methods are also compared. The second part is classification, results of word embedding can be further used to extract features and do the classification based on Neural Networks. Two popular structures: Convolutional Neural Network and Long Short Time Memory are implemented and compared. Experiments show that using the combination of Word embedding and Neural Network based classification performs much better than using traditional Bag of words to represent text and do the classification directly. The distinguished performance of CNN in solving relation classification problems are shown by experiments. Some methods are also taken to improve the performance of CNN-based structure in order to achieve the best classification results. Master of Science (Signal Processing) 2018-09-10T13:19:12Z 2018-09-10T13:19:12Z 2018 Thesis http://hdl.handle.net/10356/75959 en 77 p. application/pdf |
| spellingShingle | DRNTU::Engineering::Electrical and electronic engineering Li, Linjie Relation identification for reasoning |
| title | Relation identification for reasoning |
| title_full | Relation identification for reasoning |
| title_fullStr | Relation identification for reasoning |
| title_full_unstemmed | Relation identification for reasoning |
| title_short | Relation identification for reasoning |
| title_sort | relation identification for reasoning |
| topic | DRNTU::Engineering::Electrical and electronic engineering |
| url | http://hdl.handle.net/10356/75959 |
| work_keys_str_mv | AT lilinjie relationidentificationforreasoning |