Spaces:
Sleeping
Sleeping
File size: 3,825 Bytes
d445f2a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 |
from typing import Optional, Dict, Any, Type
from crewai.tools import BaseTool
from pydantic import Field, BaseModel
# Define the input schema as a separate class
class ContentAnalyzerArgs(BaseModel):
query: str = Field(
...,
description="The search query to compare content against"
)
content: str = Field(
...,
description="The content to analyze for relevance and factuality"
)
class ContentAnalyzerTool(BaseTool):
"""
A tool for analyzing content relevance and factuality.
This tool uses LLM to judge the relevance and factual accuracy of content
in relation to a specific query.
"""
name: str = Field(
default="Content Analyzer",
description="Name of the content analysis tool"
)
description: str = Field(
default=(
"Use this tool to analyze the relevance and factuality of content "
"in relation to a specific query. "
"It helps filter out irrelevant or potentially non-factual information."
),
description="Description of what the content analyzer does"
)
# Define args_schema as a class attribute
args_schema: Type[BaseModel] = ContentAnalyzerArgs
def _run(self, query: str, content: str) -> Dict[str, Any]:
"""
Analyze the content for relevance and factuality.
Args:
query: The original search query
content: The content to analyze
Returns:
Dict with analysis results including:
- relevance_score: A score from 0-10 indicating relevance
- factuality_score: A score from 0-10 indicating factual reliability
- filtered_content: The processed content with irrelevant parts removed
- analysis: Brief explanation of the judgment
"""
# The actual implementation will use the agent's LLM
# via CrewAI's mechanism, returning the placeholders
# for now which will be replaced during execution
prompt = f"""
You are a strict content judge evaluating web search results.
QUERY: {query}
CONTENT: {content}
Analyze the content above with these criteria:
1. Relevance to the query (score 0-10)
2. Factual accuracy and reliability (score 0-10)
3. Information quality
For content scoring below 5 on relevance, discard it entirely.
For content with factuality concerns, flag these specifically.
PROVIDE YOUR ANALYSIS IN THIS FORMAT:
{{
"relevance_score": [0-10],
"factuality_score": [0-10],
"filtered_content": "The filtered and cleaned content, removing irrelevant parts",
"analysis": "Brief explanation of your judgment"
}}
ONLY RETURN THE JSON, nothing else.
"""
# This method will be handled by CrewAI's internal mechanism
# For placeholder purposes during direct testing, we return example data.
# In a real CrewAI run, the agent's LLM would process the prompt.
return {
"relevance_score": 7, # Placeholder
"factuality_score": 8, # Placeholder
"filtered_content": content, # Placeholder
"analysis": "This is a placeholder analysis. The real analysis will be performed during execution."
}
class Config:
"""Pydantic config for the tool"""
arbitrary_types_allowed = True
def run(self, query: str, content: str) -> Dict[str, Any]:
"""Public method to run content analysis"""
return self._run(query, content) |