streamlit_app.py

import streamlit as st
import pdfplumber, re
from transformers import pipeline, AutoTokenizer

# ───────────────── Cached pipelines ────────────────────────────────────
@st.cache_resource(ttl=86400)
def load_pipes():
    summarizer = pipeline("summarization",         model=SUMM_MODEL)
    tokenizer  = AutoTokenizer.from_pretrained(    SUMM_MODEL)
    sentiment  = pipeline("text-classification",   model=SENT_MODEL)
    ner        = pipeline("token-classification",  model=NER_MODEL,
                         aggregation_strategy="simple")
    return summarizer, tokenizer, sentiment, ner

# ───────────────── Helper functions ────────────────────────────────────
def split_by_tokens(text, max_tokens):
    words = re.split(r"(\s+)", text)
    buf, n = "", 0
    for w in words:
        ln = len(TOK(w).input_ids)
        if n + ln <= max_tokens:
            buf, n = buf + w, n + ln
        else:
            yield buf.strip(); buf, n = w, ln
    if buf.strip(): yield buf.strip()

def summarise(text):
    parts = list(split_by_tokens(text, MAX_TOK))
    per_len = max(25, min(80, TARGET_WORDS // max(1, len(parts))))
    first = [SUMMAR(p, max_length=per_len,
                       min_length=per_len//2,
                       do_sample=False)[0]["summary_text"]
             for p in parts]
    joined = " ".join(first)
    if len(joined.split()) > TARGET_WORDS:
        joined = SUMMAR(joined, max_length=TARGET_WORDS,
                                 min_length=TARGET_WORDS//2,
                                 do_sample=False)[0]["summary_text"]
    return joined

def shorten(summary, n):
    s = summary.split(". ")
    return (". ".join(s[:n]).rstrip(".") + ".") if len(s) > n else summary

def extract_pdf(file):
    txt=""
    with pdfplumber.open(file) as pdf:
        for p in pdf.pages: txt += p.extract_text() or ""
    return txt

def tag_entities(text):
    tt = {"Organization":[], "Person":[], "Location":[], "Miscellaneous":[]}
    for e in NER(text):
        grp = {"ORG":"Organization","PER":"Person",
               "LOC":"Location"}.get(e["entity_group"],"Miscellaneous")
        tt[grp].append(e["word"])
    return {k: sorted(set(v)) for k,v in tt.items() if v}

# ───────────────── Main Part ───────────────────────────────────────
st.set_page_config(page_title="Financial News Analyzer",
                   page_icon="📰",
                   layout="wide")
st.title("📰 Financial News Analyzer")
st.markdown("##### Instantly grasp news content, sentiment, and relevant entities")

# models and other constant variables
SUMM_MODEL   = "sshleifer/distilbart-cnn-12-6"
SENT_MODEL   = "nynn/Fintuned_Sentiment"
NER_MODEL    = "Babelscape/wikineural-multilingual-ner"
SUMMAR, TOK, SENT_CLF, NER = load_pipes()

MAX_TOK      = 1024
TARGET_WORDS = 225
LABEL_MAP    = {"LABEL_0":"Negative","LABEL_1":"Positive","LABEL_2":"Neutral"}
COLOR_MAP    = {"Positive":"green","Negative":"red","Neutral":"gray"}

# ───────────────── Sidebar input ───────────────────────────────────────
with st.sidebar:
    st.header("Input News to Analyze:")
    txt_input   = st.text_area("Paste news article", height=150)
    pdf_file    = st.file_uploader("Or upload PDF", type=["pdf"])
    sent_count  = st.slider("Summary length (sentences)",
                            min_value=1, max_value=5, value=3, step=1)
    run_btn     = st.button("🔍 Analyze", use_container_width=True)

raw_text = extract_pdf(pdf_file) if pdf_file else txt_input.strip()

# ───────────────── Main pipeline ───────────────────────────────────────
if run_btn:
    if not raw_text:
        st.warning("Please provide text or a PDF first.")
        st.stop()

    with st.spinner("Analyzing"):
        full_sum = summarise(raw_text)
        summary  = shorten(full_sum, sent_count)

    cols = st.columns([2,1])
    with cols[0]:
        st.subheader("📝 Summary")
        st.write(summary)

    with cols[1]:
        res   = SENT_CLF(summary)[0]
        label = LABEL_MAP.get(res["label"], res["label"])
        colour= COLOR_MAP[label]
        st.subheader("📊 Sentiment")
        st.markdown(f"<h3 style='color:{colour};margin-bottom:0'>{label}</h3>"
                    f"{res['score']*100:.1f}% Confidence</p>",
                    unsafe_allow_html=True)

    tags = tag_entities(summary)
    st.subheader("🏷️ Relevant Tags")

    if tags:
        # CSS for the badge pills
        pill_css = """
        <style>
        .tag-pill {
          display: inline-block;
          background: #f0f2f6;
          color: #333;
          padding: 4px 10px;
          margin: 2px 4px 2px 0;
          border-radius: 12px;
          font-size: 0.9em;
        }
        .tag-cat {
          font-weight: 600;
          margin-top: 0;
          margin-bottom: 4px;
        }
        </style>
        """
        st.markdown(pill_css, unsafe_allow_html=True)

        # Render each category as a header + pills
        for category, vals in tags.items():
            st.markdown(f"<div class='tag-cat'>{category}</div>", unsafe_allow_html=True)
            pills = "".join(f"<span class='tag-pill'>{v}</span>" for v in vals)
            st.markdown(pills, unsafe_allow_html=True)
    else:
        st.info("No entities detected.")