A Summarize-then-Search Method for Long Video Question Answering: Prompt Samples
by Kinetograph: The Video Editing Technology PublicationMay 26th, 2024
Too Long; Didn't Read
In this paper, researchers explore zero-shot video QA using GPT-3, outperforming supervised models, leveraging narrative summaries and visual matching.
This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Jiwan Chung, MIR Lab Yonsei University (https://jiwanchung.github.io/);
(2) Youngjae Yu, MIR Lab Yonsei University (https://jiwanchung.github.io/).
Table of Links
- Abstract and Intro
- Method
- Experiments
- Related Work
- Conclusion
- Limitations and References
- A. Experiment Details
- B. Prompt Samples
B. Prompt Samples
We use the following prompts for each stage of Long Story Short. We break lines for visibility and instead denote the actual linebreaks with \n. Also, listed items within the prompts are abbreviated using ellipses (...).
Screenplay to Plot.
I am a highly intelligent storytelling bot.
If you give me a script, I will give
you the short synopsis in detail.\n\n
[Generated Screenplay]\n\n
Synopsis:
Plot Index Lookup.
Plot:\n
(1) [Plot1]\n
(2) [Plot2]\n
...\n
(N) [PlotN]\n\n
I am a highly intelligent question answering bot.
If you provide me with a question, I will give you
an index of the plot you should lookup to solve it.\n
Q: [Question]\n
Top 1 Plot Index: (
Plot:\n
(1) [Plot1]\n
(2) [Plot2]\n
...\n
(N) [PlotN]\n\n
[Generated Screenplay]\n\n
I am a highly intelligent plot question answering bot.
If you ask me a question and candidates, I will give you
the index of answer.\n
Q: [Question]\n
Candidates:\n
(1): [Answer1]\n
...\n
(5): [Answer5]\n
A: (
L O A D I N G
. . . comments & more!
. . . comments & more!
