app.py

from sentence_transformers import SentenceTransformer, util
from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline
import gradio as gr

# Load the SentenceTransformer model
model = SentenceTransformer('msmarco-distilbert-base-v4')

# Load Hugging Face NER model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("dbmdz/bert-large-cased-finetuned-conll03-english")
ner_model = AutoModelForTokenClassification.from_pretrained("dbmdz/bert-large-cased-finetuned-conll03-english")
ner_pipeline = pipeline("ner", model=ner_model, tokenizer=tokenizer, aggregation_strategy="simple")

# Define function to extract entities from text using the Hugging Face NER pipeline
def extract_entities(text):
    entities = {"skills": [], "experience": [], "education": []}
    ner_results = ner_pipeline(text)
    for entity in ner_results:
        label = entity['entity_group']
        if "SKILL" in label:
            entities["skills"].append(entity['word'])
        elif "EXPERIENCE" in label or "JOB" in label:
            entities["experience"].append(entity['word'])
        elif "DEGREE" in label or "EDUCATION" in label:
            entities["education"].append(entity['word'])
    return entities

def match_cv_to_job(cv_text, job_description):
    debug_info = "Debug Info:\n"
    
    # Extract entities from CV and job description
    cv_entities = extract_entities(cv_text)
    job_entities = extract_entities(job_description)
    
    # Calculate similarity score between entities
    match_score = 0
    for key in cv_entities:
        if key in job_entities:
            match_score += len(set(cv_entities[key]) & set(job_entities[key])) / len(set(job_entities[key])) if job_entities[key] else 0

    # Average score by number of categories
    ner_match_score = (match_score / 3) * 100  # Normalized score for NER entities
    debug_info += f"NER Match Score: {ner_match_score:.2f}%\n"
    
    # Calculate overall similarity score using embeddings
    cv_embedding = model.encode(cv_text, convert_to_tensor=True)
    job_embedding = model.encode(job_description, convert_to_tensor=True)
    similarity_score = util.pytorch_cos_sim(cv_embedding, job_embedding).item()
    
    # Combine scores with weights (embedding similarity + NER matching)
    combined_score = (similarity_score * 0.7) + (ner_match_score / 100) * 0.3  # Weighted combined score
    match_percentage = combined_score * 100
    debug_info += f"Overall Match Percentage: {match_percentage:.2f}%\n"
    
    return {"Match Percentage": f"{match_percentage:.2f}%"}, debug_info

# Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# CV and Job Description Matcher with Embeddings and NER Matching")
    
    cv_text = gr.Textbox(label="CV Text", placeholder="Enter the CV text here", lines=10)
    job_description = gr.Textbox(label="Job Description", placeholder="Enter the entire job description text here", lines=10)
    
    match_button = gr.Button("Calculate Match Percentage")
    output = gr.JSON(label="Match Result")
    debug_output = gr.Textbox(label="Debug Info", lines=10)
    
    match_button.click(fn=match_cv_to_job, inputs=[cv_text, job_description], outputs=[output, debug_output])

demo.launch()