app.py

import streamlit as st
import spacy
from spacy.matcher import Matcher
import subprocess

# Download the language model
subprocess.run(["python", "-m", "spacy", "download", "en_core_web_lg"])

# Load the glinter-spacy model
nlp = spacy.load("en_core_web_lg")

# Initialize the PhraseMatcher with the shared vocab
matcher = Matcher(nlp.vocab)

# Define the patterns for agg_type
patterns = {
    "COUNT": [
        [{"LOWER": "how"}, {"LOWER": "many"}],
        [{"LOWER": "how"}, {"LOWER": "frequent"}],
        [{"LOWER": "how"}, {"LOWER": "often"}],
        [{"LOWER": "count"}],
        [{"LOWER": "total"}, {"LOWER": "number"}]
    ],
    "SUM": [
        [{"LOWER": "how"}, {"LOWER": "much"}],
        [{"LOWER": "total"}, {"LOWER": "amount"}],
        [{"LOWER": "sum"}, {"LOWER": "of"}],
        [{"LOWER": "sum"}]
    ],
    "AVG": [
        [{"LOWER": "usually"}, {"LOWER": "paid"}],
        [{"LOWER": "usually"}, {"LOWER": "spend"}],
        [{"LOWER": "average"}],
        [{"LOWER": "avg"}],
        [{"LOWER": "per"}, {"LOWER": "month"}],
        [{"LOWER": "usual"}, {"LOWER": "amount"}]
    ],
    "MIN": [
        [{"LOWER": "minimum"}],
        [{"LOWER": "min"}],
        [{"LOWER": "lowest"}],
        [{"LOWER": "the"}, {"LOWER": "least"}]
    ],
    "MAX": [
        [{"LOWER": "maximum"}],
        [{"LOWER": "max"}],
        [{"LOWER": "highest"}],
        [{"LOWER": "the"}, {"LOWER": "most"}]
    ]
}

# Add the patterns to the matcher with the label "agg_type"
for label, pattern in patterns.items():
    matcher.add(label, pattern)

# Streamlit app
st.title("Entity Extraction")

options = [
    "How much do I spend on average on groceries?", 
    "Which is the most I have spent at Nike?", 
    "How much do I usually spend at McDonalds per month?", 
    "How much do I usually spend at McDonalds?", 
    "How much do I spend at McDonalds per month?", 
    "How many transactions do I have with Parkichen in December 2024?",
    "How often do I shop at Peek & Cloppenburg?",
    "How frequent do I dine at Parkitchen?"
]
selection = st.pills("Sample questions about transactions", options, selection_mode="single")

# Text input from the user
user_input = st.text_area("Enter query:", value=selection if selection else "")

if user_input:
    # Process the text
    doc = nlp(user_input)

    # Apply the matcher to the doc
    matches = matcher(doc)
    
    # Extract and display entities
    st.subheader("Extracted Entities and matched patterns")
    for ent in doc.ents:
        st.markdown(f"<span class='entity'>{ent.text} <span class='label'>{ent.label_}</span></span>", unsafe_allow_html=True)
    
    # Extract and display matched patterns
    for match_id, start, end in matches:
        span = doc[start:end]
        label = nlp.vocab.strings[match_id]  # Get the label for the matched pattern
        st.markdown(f"<span class='entity'>{span.text} <span class='label'>{label}</span></span>", unsafe_allow_html=True)

# Add custom CSS for styling
st.markdown("""
    <style>
    .entity {
        display: inline-block;
        padding: 0.25em 0.4em;
        margin: 0 0.25em 0.25em 0;
        border-radius: 0.25rem;
        background: #e2e2e2;
        border: 1px solid #cccccc;
    }
    .label {
        font-size: 0.75em;
        font-weight: bold;
        color: #333333;
        margin-left: 0.5em;
    }
    </style>
    """, unsafe_allow_html=True)