from pydantic import NoneStr
import os
import mimetypes
import validators
import requests
import tempfile
import gradio as gr
import openai
import re
import json
from transformers import pipeline
import matplotlib.pyplot as plt
import plotly.express as px
class SentimentAnalyzer:
def __init__(self):
self.model="facebook/bart-large-mnli"
openai.api_key=os.getenv("OPENAI_API_KEY")
def analyze_sentiment(self, text):
pipe = pipeline("zero-shot-classification", model=self.model)
label=["positive","negative","neutral"]
result = pipe(text, label)
sentiment_scores= {result['labels'][0]:result['scores'][0],result['labels'][1]:result['scores'][1],result['labels'][2]:result['scores'][2]}
sentiment_scores_str = f"Positive: {sentiment_scores['positive']:.2f}, Neutral: {sentiment_scores['neutral']:.2f}, Negative: {sentiment_scores['negative']:.2f}"
return sentiment_scores_str
def emotion_analysis(self,text):
prompt = f""" Your task is to analyze {text} and predict the emotion using scores. Emotions are categorized into the following list: Sadness, Happiness, Joy, Fear, Disgust, and Anger. You need to provide the emotion with the highest score. The scores should be in the range of 0.0 to 1.0, where 1.0 represents the highest intensity of the emotion.
Please analyze the text and provide the output in the following format: emotion: score [with one result having the highest score]."""
response = openai.Completion.create(
model="text-davinci-003",
prompt=prompt,
temperature=1,
max_tokens=60,
top_p=1,
frequency_penalty=0,
presence_penalty=0
)
message = response.choices[0].text.strip().replace("\n","")
print(message)
return message
def analyze_sentiment_for_graph(self, text):
pipe = pipeline("zero-shot-classification", model=self.model)
label=["positive", "negative", "neutral"]
result = pipe(text, label)
sentiment_scores = {
result['labels'][0]: result['scores'][0],
result['labels'][1]: result['scores'][1],
result['labels'][2]: result['scores'][2]
}
return sentiment_scores
def emotion_analysis_for_graph(self,text):
list_of_emotion=text.split(":")
label=list_of_emotion[0]
score=list_of_emotion[1]
score_dict={
label:float(score)
}
print(score_dict)
return score_dict
class Summarizer:
def __init__(self):
openai.api_key=os.getenv("OPENAI_API_KEY")
def generate_summary(self, text):
model_engine = "text-davinci-003"
prompt = f"""summarize the following conversation delimited by triple backticks.
write within 30 words.
```{text}``` """
completions = openai.Completion.create(
engine=model_engine,
prompt=prompt,
max_tokens=60,
n=1,
stop=None,
temperature=0.5,
)
message = completions.choices[0].text.strip()
return message
history_state = gr.State()
summarizer = Summarizer()
sentiment = SentimentAnalyzer()
class LangChain_Document_QA:
def __init__(self):
openai.api_key=os.getenv("OPENAI_API_KEY")
def _add_text(self,history, text):
history = history + [(text, None)]
history_state.value = history
return history,gr.update(value="", interactive=False)
def _agent_text(self,history, text):
response = text
history[-1][1] = response
history_state.value = history
return history
def _chat_history(self):
history = history_state.value
formatted_history = " "
for entry in history:
customer_text, agent_text = entry
formatted_history += f"Patient: {customer_text}\n"
if agent_text:
formatted_history += f"Psycotherapist Bot: {agent_text}\n"
return formatted_history
def _display_history(self):
formatted_history=self._chat_history()
summary=summarizer.generate_summary(formatted_history)
return summary
def _display_graph(self,sentiment_scores):
labels = sentiment_scores.keys()
scores = sentiment_scores.values()
fig = px.bar(x=scores, y=labels, orientation='h', color=labels, color_discrete_map={"Negative": "red", "Positive": "green", "Neutral": "gray"})
fig.update_traces(texttemplate='%{x:.2f}%', textposition='outside')
fig.update_layout(height=500, width=200)
return fig
def _history_of_chat(self):
history = history_state.value
formatted_history = ""
client=""
agent=""
for entry in history:
customer_text, agent_text = entry
client+=customer_text
formatted_history += f"Patient: {customer_text}\n"
if agent_text:
agent+=agent_text
formatted_history += f"Mental Healthcare Doctor Chatbot: {agent_text}\n"
return client,agent
def _suggested_answer(self,text):
try:
history = self._chat_history()
try:
file_path = "/content/patient_details.json"
with open(file_path) as file:
patient_details = json.load(file)
except:
pass
prompt = f"""As an empathic AI psychotherapist chatbot, provide effective solutions to patients' mental health concerns.
if patient say thanking tone message to end the conversation with a thanking greeting when the patient expresses gratitude.
Analyse the patient json If asked for information take it from {patient_details}
Chat History:[{history}]
Patient: [{text}]
Perform as Mental Healthcare Doctor Chatbot
"""
response = openai.Completion.create(
model="text-davinci-003",
prompt=prompt,
temperature=0,
max_tokens=500,
top_p=1,
frequency_penalty=0,
presence_penalty=0.6,
)
message = response.choices[0].text.strip()
if ":" in message:
message = re.sub(r'^.*:', '', message)
return message.strip()
except:
return "How can I help you?"
def _text_box(self,customer_emotion,agent_emotion,agent_sentiment_score,customer_sentiment_score):
agent_score = ", ".join([f"{key}: {value:.2f}" for key, value in agent_sentiment_score.items()])
customer_score = ", ".join([f"{key}: {value:.2f}" for key, value in customer_sentiment_score.items()])
return f"customer_emotion:{customer_emotion}\nagent_emotion:{agent_emotion}\nAgent_Sentiment_score:{agent_score}\nCustomer_sentiment_score:{customer_score}"
def _on_sentiment_btn_click(self):
client,agent=self._history_of_chat()
customer_emotion=sentiment.emotion_analysis(client)
customer_sentiment_score = sentiment.analyze_sentiment_for_graph(client)
scores=self._text_box(customer_emotion,agent_emotion,agent_sentiment_score,customer_sentiment_score)
customer_fig=self._display_graph(customer_sentiment_score)
customer_fig.update_layout(title="Sentiment Analysis",width=775)
customer_emotion_score = sentiment.emotion_analysis_for_graph(customer_emotion)
customer_emotion_fig=self._display_graph(customer_emotion_score)
customer_emotion_fig.update_layout(title="Emotion Analysis",width=775)
return scores,customer_fig,customer_emotion_fig
def clear_func(self):
history_state.clear()
def gradio_interface(self):
with gr.Blocks(css="style.css",theme=gr.themes.Soft()) as demo:
with gr.Row():
gr.HTML("""
""")
with gr.Row():
gr.HTML("""AI Mental Healthcare ChatBot
""")
chatbot = gr.Chatbot([], elem_id="chatbot").style(height=360)
with gr.Row():
with gr.Column(scale=0.8):
txt = gr.Textbox(
show_label=False,
placeholder="Patient",elem_classes="height"
).style(container=False)
with gr.Column(scale=0.2):
emptyBtn = gr.Button(
"๐งน Clear",elem_classes="height"
)
with gr.Row():
with gr.Column(scale=0.80):
txt3 =gr.Textbox(
show_label=False,
placeholder="AI Healthcare Suggesstion"
).style(container=False)
with gr.Column(scale=0.20, min_width=0):
button=gr.Button(value="๐send")
with gr.Row():
with gr.Column(scale=0.50):
txt4 =gr.Textbox(
show_label=False,
lines=4,
placeholder="Summary",
).style(container=False)
with gr.Column(scale=0.50):
txt5 =gr.Textbox(
show_label=False,
lines=4,
placeholder="Sentiment",
).style(container=False)
with gr.Row():
with gr.Column(scale=0.50, min_width=0):
end_btn=gr.Button(
value="End",elem_classes="height"
)
with gr.Column(scale=0.50, min_width=0):
Sentiment_btn=gr.Button(
value="๐",elem_classes="height",callback=self._on_sentiment_btn_click
)
with gr.Row():
gr.HTML("""Sentiment and Emotion Score Graph
""")
with gr.Row():
with gr.Column(scale=0.50, min_width=0):
plot =gr.Plot(label="Patient", size=(500, 600))
with gr.Column(scale=0.50, min_width=0):
plot_2 =gr.Plot(label="Psychotherapist", size=(500, 600))
with gr.Row():
with gr.Column(scale=0.50, min_width=0):
plot_3 =gr.Plot(label="Patient_Emotion", size=(500, 600))
with gr.Column(scale=0.50, min_width=0):
plot_4 =gr.Plot(label="Psychotherapist_Emotion", size=(500, 600))
txt_msg = txt.submit(self._add_text, [chatbot, txt], [chatbot, txt])
txt_msg.then(lambda: gr.update(interactive=True), None, [txt])
txt.submit(self._suggested_answer,txt,txt3)
button.click(self._agent_text, [chatbot,txt3], chatbot)
txt2.submit(self._agent_text, [chatbot, txt2], chatbot).then(
self._agent_text, [chatbot, txt2], chatbot
)
end_btn.click(self._display_history, [], txt4)
emptyBtn.click(self.clear_func,[],[])
emptyBtn.click(lambda: None, None, chatbot, queue=False)
Sentiment_btn.click(self._on_sentiment_btn_click,[],[txt5,plot,plot_2,plot_3,plot_4])
demo.title = "AI Psychotherapist ChatBot"
demo.launch()
document_qa =LangChain_Document_QA()
document_qa.gradio_interface()