Keyphrase Generation with Recurrent Neural Network and Attention Model
There is currently an information overload happening in the world of digital media, with millions of articles being published in a single day. This has raised concerns regarding how someone can not only document and catalogue these articles for preservation purposes, but as well as produce a s...
| Main Author: | Shemar, Jay Anil Singh |
|---|---|
| Format: | Final Year Project |
| Language: | English |
| Institution: | Universiti Teknologi Petronas |
| Record Id / ISBN-0: | utp-utpedia.21001 / |
| Published: |
IRC
2020
|
| Subjects: | |
| Online Access: |
http://utpedia.utp.edu.my/21001/1/19716_Jay%20Anil%20Singh%20Shemar.pdf http://utpedia.utp.edu.my/21001/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: |
There is currently an information overload happening in the world of digital media, with
millions of articles being published in a single day. This has raised concerns regarding
how someone can not only document and catalogue these articles for preservation
purposes, but as well as produce a summary for others who might want a quick
understanding of its contents, as producing a summary for every article takes a long
time. As such, in order to tackle this issue, automation is intended to streamline the
entire process.
The purpose of this study is to examine the effectiveness of a model when it uses an
attention model, versus one without. To clarify, the model will initially not feature any
attention model of any sort, be it Bahdanau or Luong. We intend to qualitatively
evaluate as to whether or not the attention model alone is capable of producing better
results. These results would be obtained using F1-Score, Precision and Recall evaluation
methods.
The research will first evaluate the original version of a model, that does not feature any
attention mechanism, and obtaining the F1-Score, Precision, and Recall, then tabulating
those results. Following that, the attention model will be incorporated into the model,
and the new modified model will be tested again, and the results will be tabulated.
For the purposes of this study, three datasets will be used, namely the Gigaword dataset,
as well as two datasets generated for the purposes of this study, namely a dataset
featuring Coronavirus News, and a dataset featuring Oil and Gas News. |
|---|