misinfoclass.py

#Misinformation Classication with GPT-4
from langchain.agents import load_tools
from langchain.agents import initialize_agent
from langchain.llms import OpenAI
from langchain.chat_models import ChatOpenAI
from langchain import SerpAPIWrapper

from langchain.document_loaders import WebBaseLoader
from langchain.tools import tool
import requests
from bs4 import BeautifulSoup
#compare lora finetune to normal
@tool
def read_webpage(url: str):
  """for reading websites when you have a valid url"""
  res = requests.get(url)
  html_page = res.content
  soup = BeautifulSoup(html_page, 'html.parser')
  ans = ""
  for para in soup.find_all("p"):
    ans += para.get_text()
    if len(ans) > 1000: break
  return ans

@tool
def search(input: str):
  """for performing google searches"""
  search = SerpAPIWrapper()
  # res = search.run(input + " site:factcheck.org OR site:snopes.com")
  res = search.run(input)
  return res




def search_classify(text):
  llm = ChatOpenAI(temperature=0, model="gpt-4")
  tools = load_tools([], llm=llm)
  # tools.append(read_webpage)
  tools.append(search)
  #Add Google Factcheck API as a tool
  agent = initialize_agent(tools, llm, agent="zero-shot-react-description", verbose=True)
  from langchain.prompts import PromptTemplate

  claim = text

  result = agent.run(f"Rate the truthfulness of the following statement: {claim}. Provide a classification from 0 to 100, 0 meaning completely false and 100 meaning completely true. Do not provide any explanations, only respond with the numerical score. After 3 unhelpful searches, return 50.")
  return result


from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain


def classify(text):
    llm = ChatOpenAI(temperature=0, model="gpt-4")
    prompt_template = "Rate the truthfulness of the following statement: '{text}'. Provide a classification from 0 to 100, 0 meaning definitely false and 100 meaning definitely true. Do not provide any explanations, only respond with the numerical score. Return -1 if you are unable to respond. "
    llm_chain = LLMChain(
      llm=llm,
      prompt=PromptTemplate.from_template(prompt_template)
    )
    return llm_chain(text)['text'];


#Add Output parser
#Site whitelist