Card: Policy Frames Classification

          
Prompt Description
[Briefly describe the purpose of the prompt and the context in which it is intended to be used, especially in the context of artificial annotation with generative models.]
Prompt Text
Content moderation, as described above, can be linked to various other topics, such as health, crime, or equality.
For each tweet in the sample, follow these instructions:
1. Carefully read the text of the tweet, paying close attention to details. 2. Classify the tweet into one of the topics defined below.
The topics are defined as follows:
* ECONOMY: The costs, benefits, or monetary/financial implications of the issue (to an individual, family, community, or to the economy as a whole).
* Capacity and resources: The lack of or availability of physical, geographical, spatial, human, and financial resources, or the capacity of existing systems and resources to implement or carry out policy goals.
* MORALITY: Any perspective—or policy objective or action (including proposed action)that is compelled by religious doctrine or interpretation, duty, honor, righ- teousness or any other sense of ethics or social responsibility.
* FAIRNESS AND EQUALITY: Equality or inequality with which laws, punishment, rewards, and resources are applied or distributed among individuals or groups. Also the balance between the rights or interests of one individual or group compared to another individual or group.
* CONSTITUTIONALITY AND JURISPRUDENCE: The constraints imposed on or freedoms granted to individuals, government, and corporations via the Constitution, Bill of Rights and other amendments, or judicial interpretation. This deals specifically with the authority of government to regulate, and the authority of in- dividuals/corporations to act independently of government.
* POLICY PRESCRIPTION AND EVALUATION: Particular policies proposed for addressing an identified problem, and figuring out if certain policies will work, or if existing policies are effective.
* LAW AND ORDER, CRIME AND JUSTICE: Specific policies in practice and their enforcement, incentives, and implications. Includes stories about enforcement and interpretation of laws by individuals and law enforcement, breaking laws, loopholes, fines, sentencing and punishment. Increases or reductions in crime.
* SECURITY AND DEFENSE: Security, threats to security, and protection of one’s person, family, in-group, nation, etc. Generally an action or a call to action that can be taken to protect the welfare of a person, group, nation sometimes from a not yet manifested threat.
* HEALTH AND SAFETY: Health care access and effectiveness, illness, disease, sanitation, obesity, mental health effects, prevention of or perpetuation of gun violence, infrastructure and building safety.
* QUALITY OF LIFE: The effects of a policy on individuals’ wealth, mobility, access to resources, happiness, social structures, ease of day-to-day routines, quality of community life, etc.
* CULTURAL IDENTITY: The social norms, trends, values and customs constituting culture(s), as they relate to a specific policy issue.
* PUBLIC OPINION: References to general social attitudes, polling and demographic information, as well as implied or actual consequences of diverging from or “getting ahead of” public opinion or polls.
* POLITICAL: Any political considerations surrounding an issue. Issue actions or efforts or stances that are political, such as partisan filibusters, lobbyist involvement, bipartisan efforts, deal-making and vote trading, appealing to one’s base, mentions of political maneuvering. Explicit statements that a policy issue is good or bad for a particular political party.
* EXTERNAL REGULATION AND REPUTATION: The United States’ external relations with another nation; the external relations of one state with another; or relations between groups. This includes trade agreements and outcomes, comparisons of policy outcomes or desired policy outcomes.
* OTHER: Any topic that does not fit into the above categories.
Language

Prompt Language: [Specify the language of the prompt, e.g., English]
Dataset Language: [Specify the language of the dataset to which the prompt is applied, e.g., English]

NLP Task

Task: [Specify the NLP task in more detail, e.g., sentiment analysis, named entity recognition, summarization]
Subtask: [If applicable, provide any subtask or variation related to the main NLP task, e.g., binary sentiment classification, multi-class sentiment classification]

Example Input and Output

Example 1
Input: [Provide an example input for the prompt]
Output: [Provide an example output for the given input]
Example 2
Input: [Provide another example input for the prompt]
Output: [Provide another example output for the given input]

Parameters and Constraints

Parameter 1: [Specify any parameters, such as temperature or token count]
Parameter 2: [Specify additional parameters or constraints if applicable]

Evaluation Metrics
[List the evaluation metrics used to assess the quality of the generated artificial annotations, such as accuracy, F1 score, or BLEU score.]
Use Cases
[List any specific use cases or applications for the prompt in artificial annotation, such as data annotation, semi-supervised learning, or active learning.]
Limitations and Potential Biases
[Briefly discuss any limitations or potential biases associated with the prompt, as well as any steps taken to mitigate them, in the context of artificial annotation with generative models.]
Related Research and References
[List any relevant research papers, articles, or resources that informed the creation of the prompt or are closely related to it, especially in the area of artificial annotation with generative models. Include proper citations where applicable.]
Cite


Fabrizio Gilardi, Meysam Alizadeh, Maël Kubli (2023) "ChatGPT Outperforms Crowd-Workers for Text-Annotation Tasks" [Paper]