Negative emotions in literature: a computational approach to tone and mood
| Project/Area Number |
22K18154
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 90020:Library and information science, humanistic and social informatics-related
|
|---|
| Research Institution | Waseda University |
|---|
Principal Investigator |
OHMAN Emily 早稲田大学, 国際学術院, 講師(テニュアトラック) (60906543)
|
|---|
| Project Period (FY) |
2022-04-01 – 2024-03-31
|
| Project Status |
Granted (Fiscal Year 2022)
|
| Budget Amount *help |
¥1,950,000 (Direct Cost: ¥1,500,000、Indirect Cost: ¥450,000)
Fiscal Year 2023: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
Fiscal Year 2022: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000)
|
|---|
| Keywords | sentiment analysis / CLS / emotion detection / NLP / machine learning / word embeddings / affect studies |
|---|
| Outline of Research at the Start |
With the creation of period- and genre-specific language models for turn-of-the-century literary texts this project aims to deliver robust tone and mood detection methods for literary studies that will not only improve existing computational literary studies approaches, but also sentiment analysis.
|
| Outline of Annual Research Achievements |
I have developed emotion lexicons for emotion association and emotion intensity. These lexicons were used to create affective word embeddings from an initial corpus of 1000 works of literature. In an iterative process, the lexicons can then be enhanced and expanded and used to fine-tune the emotion detection model for the specific domain and to adjust for semantic shifts in language. We have been able to show that mood can quite accurately be detected computationally by focusing on the first three paragraphs of a book. A by-product of this project was the development of a Finnish chapterize package that splits these books into chapters automatically.
The findings were disseminated in two peer-reviewed papers and presented at several international conferences. Three more papers are currently in peer-review. Both the lexicons and the literature corpus have been made publicly available.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
I was able to discover a method for mood detection and publish about it already, which left time to explore other uses of the same methodology resulting in new international collaborations. The project scope has therefore been expanded slightly to include fractal sentiment arcs in literature.
Research output has also exceeded expectations with two papers already published and three more currently in peer-review. Only two papers were anticipated this fiscal year.
|
| Strategy for Future Research Activity |
We plan on exploring both phonoemotions (the emotion associations of certain sounds) in the corpus as well as fractal sentiment arcs in conjunction with the methodology developed for this project. I will also work on further improving the current emotion model by incorporating and fine-tuning large language models.
|
Report
(1 results)
Research Products
(10 results)
