An easy-to-use API wrapper written in Python, for Perspective API, an API that uses machine learning to identify "toxic" comments.
Download & install git from https://git-scm.com/ and run below command:
python -m pip install --upgrade git+https://github.com/Yilmaz4/perspective.py.git
Or alternatively, you can install from PyPI (Python Package Index):
pyhton -m pip install --upgrade perspective.py
You need to get an API key from Google to use Perspective API. Instructions are explained in this article.
text: str
The text to analyze
requestedAttributes: list[str]
A list of attributes to analyze the text for.
language: str
The language of text.
You can use either Attributes.TOXICITY
object or simply "TOXICITY"
as a string for requestedAttributes
argument.
You can specify the language of the text by using language
argument. language
argument accepts both language codes (such as "en" or "es") and language names (such as "English" or "Spanish"). Small spelling mistakes in language names can also be accepted (such as "Eglish" or "Spamish"). If you set language
argument to None
, language will be automatically detected.
You can find a list of all attributes and languages each attribute supports in this article.
from perspective import Client, Attributes, utils
# Creating the Client object with the API key
API_KEY = "your_api_key"
client = Client(token = API_KEY)
# Make a request to Perspective API with a text to analyze and requested attributes
response = client.analyze(text = "Hey! How are you?", requestedAttributes = [Attributes.TOXICITY, Attributes.INSULT])
# Print the response (dict)
print(response)
# Print the percent of TOXICITY attribute
print(response["TOXICITY"]
print(" ")
# Iterate over the response
for attribute, result in response.items():
print(attribute.capitalize() + ": " + "%.2f" % result + "%")
print(" ")
# Or, use utils.format_response to print a formatted text of the response
print(utils.format_response(response, align_right=True))
{'TOXICITY': 7.019685000000001, 'INSULT': 3.9963423999999996}
7.019685000000001
Toxicity: 7.02%
Insult: 4.00%
Toxicity: 7.02%
Insult: 4.00%
As you can see in the output, Client.analyze
returns a dictionary with requested attributes and their analysis results as percents. You can get percents of each attribute, or iterate over dictionary.
from perspective import Client, Attributes, utils
# Creating the Client object with the API key
API_KEY = "your_api_key"
client = Client(token = API_KEY)
# Make a request to Perspective API with a text to analyze and requested attributes
response = client.analyze(text = "Hey! How are you?", requestedAttributes = Attributes.Production) # Attributes.Production includes all production-ready attributes
# Create a graph and show it by popping up a window
utils.show_graph(response=response, title="Sample graph")
# Or alternatively, you can save the graph
utils.save_graph(response=response, filename="graph.png", title="Sample graph")