From Dataset Labeling to Deployment: The Power of NLP and LLMs Combined.
PromptedGraphs is a Python library that aims to seamlessly integrate traditional NLP methods with the capabilities of modern Large Language Models (LLMs) in the realm of knowledge graphs. Our library offers tools tailored for dataset labeling, model training, and smooth deployment to production environments. We leverage the strengths of spacy for core NLP tasks, snorkel for effective data labeling, and async to ensure enhanced performance. Our mission is to provide a harmonized solution to knowledge graph development when you have to merge traditional and LLM-driven approaches, squarely addressing the challenges associated with accuracy, efficiency, and affordability.
- Named Entity Recognition (NER): Customize ER labels based on your domain.
- Structured Data Extraction: Extract structured data from unstructured text.
- Entity Resolution: Deduplication and normalization
- Relationship Extraction: Either open ended labels or constrain to your domain
- Entity Linking: Link references in text to entities in a graph
- Graph Construction: Create or update knowledge graphs
- Dataset Labeling: Efficient tools for labeling datasets, powered by
haystack. - Model Training: Combine the reliability of NLP and the prowess of LLMs.
- Deployment: Streamlined processes to ensure smooth transition to production.
- Python 3.10 or newer.
To install PromptedGraphs via pip:
pip install promptedgraphs
# or
poetry add promptedgraphsfrom examples/er_reviews.ipynb
from spacy import displacy
from promptedgraphs.config import Config
from promptedgraphs.extraction.entities_from_text import entities_from_text
labels = {
"POSITIVE": "A postive review of a product or service.",
"NEGATIVE": "A negative review of a product or service.",
"NEUTRAL": "A neutral review of a product or service.",
}
text_of_reviews = """
1. "I absolutely love this product. It's been a game changer!"
2. "The service was quite poor and the staff was rude."
3. "The item is okay. Nothing special, but it gets the job done."
""".strip()
# Label Sentiment
ents = []
async for msg in entities_from_text(
name="sentiment",
description="Sentiment Analysis of Customer Reviews",
text=text_of_reviews,
labels=labels,
config=Config(), # Reads `OPENAI_API_KEY` from .env file or environment
):
ents.append(msg)
# Show Results using spacy.displacy
render_entities(
text=text_of_reviews,
entities=ents,
labels=labels,
colors = {"POSITIVE": "#7aecec", "NEGATIVE": "#f44336", "NEUTRAL": "#f4f442"}
)
Generate a list of data that fits a given data model.
from examples/er_reviews.ipynb
from pydantic import BaseModel, Field
from promptedgraphs.config import Config
from promptedgraphs.ideation import brainstorm
from promptedgraphs.vis import render_entities
class BusinessIdea(BaseModel):
"""A business idea generated using the Jobs-to-be-done framework
For example "We help [adj] [target_audience] [action] so they can [benefit or do something else]"
"""
target_audience: str = Field(title="Target Audience")
action: str = Field(title="Action")
benefit: str = Field(title="Benefit or next action")
adj: str | None = Field(
title="Adjective",
description="Optional adjective describing the target audience's condition",
)
ideas = []
async for idea in brainstorm(
text=BusinessIdea.__doc__,
output_type=list[BusinessIdea],
config=Config(),
n=10,
max_workers=2,
):
ideas.append(idea)
render_entities(
f"We help {idea.adj} {idea.target_audience} {idea.action} so they can {idea.benefit}",
idea,
)
from examples/de_chatintents.ipynb
from pydantic import BaseModel, Field
from promptedgraphs.config import Config
class UserIntent(BaseModel):
"""The UserIntent entity, representing the canonical description of what a user desires to achieve in a given conversation."""
intent_name: str = Field(
title="Intent Name",
description="Canonical name of the user's intent",
examples=[
"question",
"command",
"clarification",
"chit_chat",
"greeting",
"feedback",
"nonsensical",
"closing",
"harrassment",
"unknown",
],
)
description: str | None = Field(
title="Intent Description",
description="A detailed explanation of the user's intent",
)
msg = """It's a busy day, I need to send an email and to buy groceries"""
async for intent in data_from_text(
text=msg, output_type=UserIntent, config=Config()
):
print(intent)intent_name='task' description='User wants to complete a task'
intent_name='communication' description='User wants to send an email'
intent_name='shopping' description='User wants to buy groceries'We welcome contributions! Please DM me @seankruzel or create issues or pull requests.
This project is licensed under the terms of the MIT license.
Built using quantready using template https://github.com/closedloop-technologies/quantready-api