文件列表:

Utilities (0, 2023-04-17)
Utilities\OpenAI_Token_Calculator.py (988, 2023-04-17)
assets (0, 2023-04-17)
assets\propaganda_data_flow.drawio (3122, 2023-04-17)
assets\propaganda_data_flow.drawio.png (48375, 2023-04-17)
prompt_response_testing.md (33193, 2023-04-17)
propaganda_detection.ipynb (168301, 2023-04-17)

# Detecting Propaganda in News Articles Using Large Language Models In this work, we investigate the feasibility of using Large Language Models (LLMs) to detect propaganda in news articles. To do so, we feed tailored prompts coupled with news articles from **Russia Today** (**RT**), a prominent state-controlled news network, into OpenAI's [**gpt-3.5-turbo**](https://platform.openai.com/docs/models/gpt-3-5) model. ## Procedure overview At a high level, an iteration of the procedure looks as follows:  ### Role & Content The `role` and `content` properties of the request are required. There are three possible values for the `role`: `user`, `assistant`, and `system`. The `user` role is used to instruct the model, the `assistant` role can be used to store prior responses (to establish context), and the `system` role to set model behaviour (see [here](https://platform.openai.com/docs/guides/chat/introduction) for an extended description). Currently, a single `user` message containing the final prompt is sent. ### Temperature A higher `temperature` value can be used to "[make the output more random](https://platform.openai.com/docs/guides/chat/instructing-chat-models)" whilst a lower one (e.g., `0`) will cause the model to choose words with the highest probability of occurrence. Since our goal is to receive responses that are as objective as possible, `0` is selected. ## Propaganda techniques Martino et al. [1] define a list of 18 distinct propaganda techniques, which are as follows: Technique | Definition ------ | ------ Name calling | Attack an object/subject of the propaganda with an insulting label Repetition | Repeat the same message over and over Slogans | Use a brief and memorable phrase Appeal to fear | Support an idea by instilling fear against other alternatives Doubt | Questioning the credibility of someone/something Exaggeration/minimization | Exaggerate or minimize something Flag-Waving | Appeal to patriotism or identity Loaded Language | Appeal to emotions or stereotypes Reduction ad hitlerum | Disapprove an idea suggesting it is popular with groups hated by the audience Bandwagon | Appeal to the popularity of an idea Causal oversimplification | Assume a simple cause for a complex event Obfuscation, intentional vagueness | Use deliberately unclear and obscure expressions to confuse the audience Appeal to authority | Use authority’s support as evidence Black&white fallacy | Present only two options among many Thought terminating cliches | Phrases that discourage critical thought and meaningful discussions Red herring | Introduce irrelevant material to distract Straw men | Refute argument that was not presented Whataboutism | Charging an opponent with hypocrisy ## References [1] Martino, G. D. S., Cresci, S., Barron-Cedeno, A., Yu, S., Di Pietro, R., & Nakov, P. (2020). A survey on computational propaganda detection. arXiv preprint arXiv:2007.08024

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：