|
|
|
|
|
""" |
|
|
Verbatify — Analyse sémantique NPS (Paste-only, NPS inféré) |
|
|
- Entrée : verbatims collés (1 ligne = 1 verbatim, score NPS optionnel après |) |
|
|
- Sorties : émotion, thématiques, occurrences, synthèse, graphiques Plotly + exports |
|
|
- IA (facultatif) : OpenAI (robuste), fallback CamemBERT si installé, puis règles |
|
|
- Branding : thème Plotly + CSS Manrope intégrés, logo inline (aucun fichier externe) |
|
|
""" |
|
|
|
|
|
import os, re, json, collections, tempfile, zipfile |
|
|
from typing import List, Dict, Optional |
|
|
import pandas as pd |
|
|
import gradio as gr |
|
|
import plotly.express as px |
|
|
import plotly.graph_objects as go |
|
|
import plotly.io as pio |
|
|
|
|
|
|
|
|
VB_CSS = r""" |
|
|
@import url('https://fonts.googleapis.com/css2?family=Manrope:wght@400;600;700;800&display=swap'); |
|
|
:root{ |
|
|
--vb-bg:#F8FAFC; --vb-card:#FFFFFF; --vb-text:#0F172A; --vb-muted:#475569; |
|
|
--vb-primary:#7C3AED; --vb-primary-2:#06B6D4; --vb-accent:#2563EB; |
|
|
--vb-positive:#10B981; --vb-negative:#EF4444; --vb-neutral:#94A3B8; |
|
|
--vb-border:#E2E8F0; --vb-radius:14px; --vb-shadow:0 8px 20px rgba(2,6,23,.06); |
|
|
} |
|
|
html,body,.gradio-container{background:var(--vb-bg)!important;color:var(--vb-text); |
|
|
font-family:Manrope,system-ui,-apple-system,Segoe UI,Roboto,Arial,sans-serif;} |
|
|
.gradio-container{max-width:1120px!important;margin:0 auto!important;} |
|
|
.vb-hero{display:flex;align-items:center;gap:14px;padding:18px 20px;margin:8px 0 18px; |
|
|
background:linear-gradient(90deg,rgba(124,58,237,.10),rgba(6,182,212,.10)); |
|
|
border:1px solid var(--vb-border);border-radius:var(--vb-radius);} |
|
|
.vb-hero .vb-title{font-size:20px;font-weight:800;letter-spacing:.2px;} |
|
|
.vb-hero .vb-sub{color:var(--vb-muted);font-size:13px;margin-top:-4px;} |
|
|
.gr-markdown,.gr-box,.gr-panel,.gr-article{background:var(--vb-card)!important;border:1px solid var(--vb-border)!important;border-radius:var(--vb-radius)!important;box-shadow:var(--vb-shadow);} |
|
|
.gr-markdown h2{font-size:18px;margin-top:10px;} |
|
|
button,.gr-button{border-radius:var(--vb-radius)!important;border:0!important;box-shadow:var(--vb-shadow);font-weight:700!important;} |
|
|
.gr-button-primary{background:linear-gradient(90deg,var(--vb-primary),var(--vb-primary-2))!important;color:#fff!important;} |
|
|
.gr-button-primary:hover{filter:brightness(1.06);transform:translateY(-1px);} |
|
|
.gr-button-secondary{background:#fff!important;border:1px solid var(--vb-border)!important;color:var(--vb-text)!important;} |
|
|
input,textarea,select,.gr-textbox,.gr-textbox textarea{border-radius:calc(var(--vb-radius) - 4px)!important;border:1px solid var(--vb-border)!important;background:#fff!important;} |
|
|
input[type="checkbox"]:checked{accent-color:var(--vb-primary)!important;} |
|
|
.js-plotly-plot .plotly .bg{fill:#fff!important;} |
|
|
.js-plotly-plot .plotly .xgrid,.js-plotly-plot .plotly .ygrid{stroke:#E2E8F0!important;opacity:1;} |
|
|
.vb-footer{color:var(--vb-muted);font-size:12px;text-align:center;margin:12px 0;} |
|
|
""" |
|
|
|
|
|
def apply_plotly_theme(): |
|
|
pio.templates["verbatify"] = go.layout.Template( |
|
|
layout=go.Layout( |
|
|
font=dict(family="Manrope, system-ui, -apple-system, Segoe UI, Roboto, Arial, sans-serif", size=13, color="#0F172A"), |
|
|
paper_bgcolor="white", plot_bgcolor="white", |
|
|
colorway=["#7C3AED","#06B6D4","#2563EB","#10B981","#EF4444","#F59E0B","#14B8A6","#F43F5E"], |
|
|
xaxis=dict(gridcolor="#E2E8F0", zerolinecolor="#E2E8F0"), |
|
|
yaxis=dict(gridcolor="#E2E8F0", zerolinecolor="#E2E8F0"), |
|
|
legend=dict(borderwidth=0, bgcolor="rgba(255,255,255,0)") |
|
|
) |
|
|
) |
|
|
pio.templates.default = "verbatify" |
|
|
|
|
|
LOGO_SVG = """<svg xmlns='http://www.w3.org/2000/svg' width='224' height='38' viewBox='0 0 224 38'> |
|
|
<defs><linearGradient id='g' x1='0%' y1='0%' x2='100%'><stop offset='0%' stop-color='#7C3AED'/><stop offset='100%' stop-color='#06B6D4'/></linearGradient></defs> |
|
|
<g fill='none' fill-rule='evenodd'> |
|
|
<rect x='0' y='7' width='38' height='24' rx='12' fill='url(#g)'/> |
|
|
<circle cx='13' cy='19' r='5' fill='#fff' opacity='0.95'/><circle cx='25' cy='19' r='5' fill='#fff' opacity='0.72'/> |
|
|
<text x='46' y='25' font-family='Manrope, system-ui, -apple-system, Segoe UI, Roboto, Arial, sans-serif' font-size='20' font-weight='800' fill='#0F172A' letter-spacing='0.2'>Verbatify</text> |
|
|
</g> |
|
|
</svg>""" |
|
|
|
|
|
|
|
|
try: |
|
|
from unidecode import unidecode |
|
|
except Exception: |
|
|
import unicodedata |
|
|
def unidecode(x): |
|
|
try: |
|
|
return unicodedata.normalize('NFKD', str(x)).encode('ascii','ignore').decode('ascii') |
|
|
except Exception: |
|
|
return str(x) |
|
|
|
|
|
|
|
|
THEMES = { |
|
|
"Remboursements santé":[r"\bremboursement[s]?\b", r"\bt[eé]l[eé]transmission\b", r"\bno[eé]mie\b", |
|
|
r"\bprise\s*en\s*charge[s]?\b", r"\btaux\s+de\s+remboursement[s]?\b", r"\b(ameli|cpam)\b", |
|
|
r"\bcompl[eé]mentaire\s+sant[eé]\b", r"\bmutuelle\b", r"\battestation[s]?\b", r"\bcarte\s+(mutuelle|tiers\s*payant)\b"], |
|
|
"Tiers payant / Réseau de soins":[r"\btiers\s*payant\b", r"\br[ée]seau[x]?\s+de\s+soins\b", |
|
|
r"\b(optique|dentaire|hospitalisation|pharmacie)\b", r"\bitelis\b", r"\bsant[eé]clair\b", r"\bkalixia\b"], |
|
|
"Sinistres / Indemnisation":[r"\bsinistre[s]?\b", r"\bindemni(sation|ser)\b", r"\bexpertis[ea]\b", |
|
|
r"\bd[eé]claration\s+de\s+sinistre\b", r"\bconstat\b", r"\bbris\s+de\s+glace\b", r"\bassistance\b", r"\bd[ée]pannage\b"], |
|
|
"Adhésion / Contrat":[r"\badh[eé]sion[s]?\b", r"\bsouscription[s]?\b", r"\baffiliation[s]?\b", r"\bcontrat[s]?\b", |
|
|
r"\bavenant[s]?\b", r"\bcarence[s]?\b", r"\brenouvellement[s]?\b", r"\br[eé]siliation[s]?\b"], |
|
|
"Garanties / Exclusions / Franchise":[r"\bgarantie[s]?\b", r"\bexclusion[s]?\b", r"\bplafond[s]?\b", |
|
|
r"\bfranchise[s]?\b", r"\bconditions\s+g[eé]n[eé]rales\b", r"\bnotice\b"], |
|
|
"Cotisations / Facturation":[r"\bcotisation[s]?\b", r"\bpr[eé]l[eè]vement[s]?\b", r"\bech[eé]ancier[s]?\b", |
|
|
r"\bfacture[s]?\b", r"\berreur[s]?\s+de\s+facturation\b", r"\bremboursement[s]?\b", r"\bRIB\b", r"\bIBAN\b"], |
|
|
"Délais & Suivi dossier":[r"\bd[eé]lai[s]?\b", r"\btraitement[s]?\b", r"\bsuivi[s]?\b", r"\brelance[s]?\b", r"\bretard[s]?\b"], |
|
|
"Espace client / App / Connexion":[r"\bespace\s+client\b", r"\bapplication\b", r"\bapp\b", r"\bsite\b", |
|
|
r"\bconnexion\b", r"\bidentifiant[s]?\b", r"\bmot\s+de\s+passe\b", r"\bpaiement\s+en\s+ligne\b", |
|
|
r"\bbogue[s]?\b", r"\bbug[s]?\b", r"\bnavigation\b", r"\binterface\b", r"\bUX\b"], |
|
|
"Support / Conseiller":[r"\bSAV\b", r"\bservice[s]?\s+client[s]?\b", r"\bconseiller[s]?\b", |
|
|
r"\b[rR][eé]ponse[s]?\b", r"\bjoignable[s]?\b", r"\brapp?el\b"], |
|
|
"Communication / Transparence":[r"\binformation[s]?\b", r"\bcommunication\b", r"\btransparence\b", |
|
|
r"\bclart[eé]\b", r"\bcourrier[s]?\b", r"\bmail[s]?\b", r"\bnotification[s]?\b"], |
|
|
"Prix":[r"\bprix\b", r"\bcher[s]?\b", r"\bco[uû]t[s]?\b", r"\btarif[s]?\b", |
|
|
r"\bcomp[eé]titif[s]?\b", r"\babusif[s]?\b", r"\bbon\s+rapport\s+qualit[eé]\s*prix\b"], |
|
|
"Offre / Gamme":[r"\boffre[s]?\b", r"\bgamme[s]?\b", r"\bdisponibilit[eé][s]?\b", r"\bdevis\b", r"\bchoix\b", r"\bcatalogue[s]?\b"], |
|
|
"Produit/Qualité":[r"\bqualit[eé]s?\b", r"\bfiable[s]?\b", r"\bconforme[s]?\b", r"\bnon\s+conforme[s]?\b", |
|
|
r"\bd[eé]fectueux?[es]?\b", r"\bperformant[e]?[s]?\b"], |
|
|
"Agence / Accueil":[r"\bagence[s]?\b", r"\bboutique[s]?\b", r"\baccueil\b", r"\bconseil[s]?\b", r"\battente\b", r"\bcaisse[s]?\b"], |
|
|
} |
|
|
|
|
|
|
|
|
POS_WORDS = {"bien":1.0,"super":1.2,"parfait":1.4,"excellent":1.5,"ravi":1.2,"satisfait":1.0, |
|
|
"rapide":0.8,"efficace":1.0,"fiable":1.0,"simple":0.8,"facile":0.8,"clair":0.8,"conforme":0.8, |
|
|
"sympa":0.8,"professionnel":1.0,"réactif":1.0,"reactif":1.0,"compétent":1.0,"competent":1.0, |
|
|
"top":1.2,"recommande":1.2,"recommandé":1.2,"bon":0.8} |
|
|
NEG_WORDS = {"mauvais":-1.2,"horrible":-1.5,"nul":-1.2,"lent":-0.8,"cher":-0.9,"arnaque":-1.5, |
|
|
"déçu":-1.2,"decu":-1.2,"incompétent":-1.3,"bug":-0.9,"bogue":-0.9,"problème":-1.0, |
|
|
"probleme":-1.0,"attente":-0.6,"retard":-0.9,"erreur":-1.0,"compliqué":-0.8,"complique":-0.8, |
|
|
"défectueux":-1.3,"defectueux":-1.3,"non conforme":-1.2,"impossible":-1.0,"difficile":-0.7} |
|
|
NEGATIONS = [r"\bpas\b", r"\bjamais\b", r"\bplus\b", r"\baucun[e]?\b", r"\brien\b", r"\bni\b", r"\bgu[eè]re\b"] |
|
|
INTENSIFIERS = [r"\btr[eè]s\b", r"\bvraiment\b", r"\bextr[eê]mement\b", r"\bhyper\b"] |
|
|
DIMINISHERS = [r"\bun[e]?\s+peu\b", r"\bassez\b", r"\bplut[oô]t\b", r"\bl[eé]g[eè]rement\b"] |
|
|
INTENSIFIER_W, DIMINISHER_W = 1.5, 0.7 |
|
|
|
|
|
|
|
|
OPENAI_AVAILABLE = False |
|
|
try: |
|
|
from openai import OpenAI |
|
|
if os.getenv("OPENAI_API_KEY"): |
|
|
_client = OpenAI() |
|
|
OPENAI_AVAILABLE = True |
|
|
except Exception: |
|
|
OPENAI_AVAILABLE = False |
|
|
|
|
|
|
|
|
def normalize(t:str)->str: |
|
|
if not isinstance(t,str): return "" |
|
|
return re.sub(r"\s+"," ",t.strip()) |
|
|
|
|
|
def to_analyzable(t:str)->str: |
|
|
return unidecode(normalize(t.lower())) |
|
|
|
|
|
def window_has(patterns:List[str], toks:List[str], i:int, w:int=3)->bool: |
|
|
s=max(0,i-w); e=min(len(toks),i+w+1); win=" ".join(toks[s:e]) |
|
|
return any(re.search(p,win) for p in patterns) |
|
|
|
|
|
def lexical_sentiment_score(text:str)->float: |
|
|
toks = to_analyzable(text).split(); score=0.0 |
|
|
for i,t in enumerate(toks): |
|
|
base = POS_WORDS.get(t,0.0) or NEG_WORDS.get(t,0.0) |
|
|
if not base and i<len(toks)-1: |
|
|
bi=f"{t} {toks[i+1]}"; base = NEG_WORDS.get(bi,0.0) |
|
|
if base: |
|
|
w=1.0 |
|
|
if window_has(INTENSIFIERS,toks,i): w*=INTENSIFIER_W |
|
|
if window_has(DIMINISHERS,toks,i): w*=DIMINISHER_W |
|
|
if window_has(NEGATIONS,toks,i): base*=-1 |
|
|
score+=base*w |
|
|
return max(min(score,4.0),-4.0) |
|
|
|
|
|
def lexical_sentiment_label(s:float)->str: |
|
|
return "positive" if s>=0.3 else ("negatif" if s<=-0.3 else "neutre") |
|
|
|
|
|
def detect_themes_regex(text:str): |
|
|
t=to_analyzable(text); counts={} |
|
|
for th,pats in THEMES.items(): |
|
|
c=sum(len(re.findall(p,t)) for p in pats) |
|
|
if c>0: counts[th]=c |
|
|
return list(counts.keys()), counts |
|
|
|
|
|
def nps_bucket(s): |
|
|
try: |
|
|
v=int(s) |
|
|
except: |
|
|
return "inconnu" |
|
|
return "promoter" if v>=9 else ("passive" if v>=7 else ("detractor" if v>=0 else "inconnu")) |
|
|
|
|
|
def compute_nps(series): |
|
|
vals=[] |
|
|
for x in series.dropna().tolist(): |
|
|
try: |
|
|
v=int(x) |
|
|
if 0<=v<=10: vals.append(v) |
|
|
except: pass |
|
|
if not vals: return None |
|
|
tot=len(vals); pro=sum(1 for v in vals if v>=9); det=sum(1 for v in vals if v<=6) |
|
|
return 100.0*(pro/tot - det/tot) |
|
|
|
|
|
def anonymize(t:str)->str: |
|
|
if not isinstance(t,str): return "" |
|
|
t=re.sub(r"[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}","[email]",t) |
|
|
t=re.sub(r"\b(?:\+?\d[\s.-]?){7,}\b","[tel]",t) |
|
|
return t |
|
|
|
|
|
|
|
|
def df_from_pasted(text:str, sep="|", has_score=False) -> pd.DataFrame: |
|
|
lines = [l.strip() for l in (text or "").splitlines() if l.strip()] |
|
|
rows = [] |
|
|
for i, line in enumerate(lines, 1): |
|
|
if has_score and sep in line: |
|
|
verb, score = line.split(sep, 1) |
|
|
rows.append({"id": i, "comment": verb.strip(), "nps_score": pd.to_numeric(score.strip(), errors="coerce")}) |
|
|
else: |
|
|
rows.append({"id": i, "comment": line.strip(), "nps_score": None}) |
|
|
return pd.DataFrame(rows) |
|
|
|
|
|
|
|
|
def openai_json(model:str, system:str, user:str, temperature:float=0.0) -> Optional[dict]: |
|
|
if not OPENAI_AVAILABLE: return None |
|
|
try: |
|
|
resp = _client.chat.completions.create( |
|
|
model=model, temperature=temperature, |
|
|
messages=[{"role":"system","content":system},{"role":"user","content":user}], |
|
|
) |
|
|
txt = resp.choices[0].message.content.strip() |
|
|
m = re.search(r"\{.*\}", txt, re.S) |
|
|
return json.loads(m.group(0) if m else txt) |
|
|
except Exception: |
|
|
return None |
|
|
|
|
|
def oa_sentiment(comment:str, model:str, temperature:float=0.0) -> Optional[dict]: |
|
|
system = "Tu es un classifieur FR. Réponds strictement en JSON." |
|
|
user = f'Texte: {comment}\nDonne "label" parmi ["positive","neutre","negatif"] et "score" entre -4 et 4. JSON.' |
|
|
return openai_json(model, system, user, temperature) |
|
|
|
|
|
def oa_themes(comment:str, model:str, temperature:float=0.0) -> Optional[dict]: |
|
|
system = "Tu maps le texte client vers un thésaurus assurance. Réponds strictement en JSON." |
|
|
user = f"Texte: {comment}\nThésaurus: {json.dumps(list(THEMES.keys()), ensure_ascii=False)}\nRetourne {{'themes': [...], 'counts': {{...}}}}" |
|
|
return openai_json(model, system, user, temperature) |
|
|
|
|
|
def oa_summary(nps:Optional[float], dist:Dict[str,int], themes_df:pd.DataFrame, model:str, temperature:float=0.2) -> Optional[str]: |
|
|
system = "Tu es un analyste CX FR. Donne une synthèse courte et actionnable en Markdown." |
|
|
top = [] if themes_df is None else themes_df.head(6).to_dict(orient="records") |
|
|
user = f"Données: NPS={None if nps is None else round(nps,1)}, Répartition={dist}, Thèmes={json.dumps(top, ensure_ascii=False)}" |
|
|
j = openai_json(model, system, user, temperature) |
|
|
if isinstance(j, dict) and "text" in j: return j["text"] |
|
|
if isinstance(j, dict): return ' '.join(str(v) for v in j.values()) |
|
|
return None |
|
|
|
|
|
|
|
|
def make_hf_pipe(): |
|
|
try: |
|
|
from transformers import pipeline |
|
|
return pipeline("text-classification", |
|
|
model="cmarkea/distilcamembert-base-sentiment", |
|
|
tokenizer="cmarkea/distilcamembert-base-sentiment") |
|
|
except Exception: |
|
|
return None |
|
|
|
|
|
|
|
|
def infer_nps_from_sentiment(label: str, score: float) -> int: |
|
|
scaled = int(round((float(score) + 4.0) * 1.25)) |
|
|
scaled = max(0, min(10, scaled)) |
|
|
if label == "positive": |
|
|
return max(9, scaled) |
|
|
if label == "negatif": |
|
|
return min(6, scaled) |
|
|
return 8 if score >= 0 else 7 |
|
|
|
|
|
|
|
|
def fig_nps_gauge(nps: Optional[float]) -> go.Figure: |
|
|
v = 0.0 if nps is None else float(nps) |
|
|
return go.Figure(go.Indicator(mode="gauge+number", value=v, |
|
|
gauge={"axis":{"range":[-100,100]}, "bar":{"thickness":0.3}}, |
|
|
title={"text":"NPS (−100 à +100)"})) |
|
|
|
|
|
def fig_sentiment_bar(dist: Dict[str,int]) -> go.Figure: |
|
|
order = ["negatif","neutre","positive"] |
|
|
x = [o for o in order if o in dist]; y = [dist.get(o,0) for o in x] |
|
|
return px.bar(x=x, y=y, labels={"x":"Sentiment","y":"Nombre"}, title="Répartition des émotions") |
|
|
|
|
|
def fig_top_themes(themes_df: pd.DataFrame, k: int) -> go.Figure: |
|
|
if themes_df is None or themes_df.empty: return go.Figure() |
|
|
d = themes_df.head(k); fig = px.bar(d, x="theme", y="total_mentions", title=f"Top {k} thèmes — occurrences") |
|
|
fig.update_layout(xaxis_tickangle=-30); return fig |
|
|
|
|
|
def fig_theme_balance(themes_df: pd.DataFrame, k: int) -> go.Figure: |
|
|
if themes_df is None or themes_df.empty: return go.Figure() |
|
|
d = themes_df.head(k) |
|
|
d2 = d.melt(id_vars=["theme"], value_vars=["verbatims_pos","verbatims_neg"], var_name="type", value_name="count") |
|
|
d2["type"] = d2["type"].map({"verbatims_pos":"Positifs","verbatims_neg":"Négatifs"}) |
|
|
fig = px.bar(d2, x="theme", y="count", color="type", barmode="stack", title=f"Top {k} thèmes — balance Pos/Neg") |
|
|
fig.update_layout(xaxis_tickangle=-30); return fig |
|
|
|
|
|
|
|
|
def analyze_text(pasted_txt, has_sc, sep_chr, |
|
|
do_anonymize, use_oa_sent, use_oa_themes, use_oa_summary, |
|
|
oa_model, oa_temp, top_k): |
|
|
|
|
|
df = df_from_pasted(pasted_txt or "", sep=sep_chr or "|", has_score=bool(has_sc)) |
|
|
if df.empty: |
|
|
raise gr.Error("Colle au moins un verbatim (une ligne).") |
|
|
|
|
|
if do_anonymize: |
|
|
df["comment"]=df["comment"].apply(anonymize) |
|
|
|
|
|
|
|
|
if (use_oa_sent or use_oa_themes or use_oa_summary) and not OPENAI_AVAILABLE: |
|
|
use_oa_sent = use_oa_themes = use_oa_summary = False |
|
|
|
|
|
hf_pipe = make_hf_pipe() |
|
|
def hf_sent(text:str): |
|
|
if hf_pipe is None or not text.strip(): return None |
|
|
try: |
|
|
res=hf_pipe(text); lab=str(res[0]["label"]).lower(); p=float(res[0].get("score",0.5)) |
|
|
if "1" in lab or "2" in lab: return {"label":"negatif","score":-4*p} |
|
|
if "3" in lab: return {"label":"neutre","score":0.0} |
|
|
return {"label":"positive","score":4*p} |
|
|
except Exception: |
|
|
return None |
|
|
|
|
|
rows=[]; theme_agg=collections.defaultdict(lambda:{"mentions":0,"pos":0,"neg":0}) |
|
|
used_hf=False; used_oa=False; any_inferred=False |
|
|
|
|
|
for idx, r in df.iterrows(): |
|
|
cid=r.get("id", idx+1); comment=normalize(str(r["comment"])) |
|
|
|
|
|
|
|
|
sent=None |
|
|
if use_oa_sent: |
|
|
sent=oa_sentiment(comment, oa_model, float(oa_temp or 0.0)); used_oa = used_oa or bool(sent) |
|
|
if not sent: |
|
|
hf=hf_sent(comment) |
|
|
if hf: sent=hf; used_hf=True |
|
|
if not sent: |
|
|
s=float(lexical_sentiment_score(comment)) |
|
|
sent={"label":lexical_sentiment_label(s),"score":s} |
|
|
|
|
|
|
|
|
themes, counts = detect_themes_regex(comment) |
|
|
if use_oa_themes: |
|
|
tjson=oa_themes(comment, oa_model, float(oa_temp or 0.0)) |
|
|
if isinstance(tjson, dict): |
|
|
used_oa=True |
|
|
for th, c in (tjson.get("counts",{}) or {}).items(): |
|
|
if th in THEMES and int(c) > 0: |
|
|
counts[th] = max(counts.get(th, 0), int(c)) |
|
|
themes = [th for th, c in counts.items() if c > 0] |
|
|
|
|
|
|
|
|
given = r.get("nps_score", None) |
|
|
try: |
|
|
given = int(given) if given is not None and str(given).strip() != "" else None |
|
|
except Exception: |
|
|
given = None |
|
|
|
|
|
if given is None: |
|
|
inferred = infer_nps_from_sentiment(sent["label"], float(sent["score"])) |
|
|
nps_final, nps_source, any_inferred = inferred, "inferred", True |
|
|
else: |
|
|
nps_final, nps_source = given, "given" |
|
|
|
|
|
bucket = nps_bucket(nps_final) |
|
|
|
|
|
for th, c in counts.items(): |
|
|
theme_agg[th]["mentions"] += c |
|
|
if sent["label"] == "positive": theme_agg[th]["pos"] += 1 |
|
|
elif sent["label"] == "negatif": theme_agg[th]["neg"] += 1 |
|
|
|
|
|
rows.append({ |
|
|
"id": cid, "comment": comment, |
|
|
"nps_score_given": given, "nps_score_inferred": nps_final if given is None else None, |
|
|
"nps_score_final": nps_final, "nps_source": nps_source, "nps_bucket": bucket, |
|
|
"sentiment_score": round(float(sent["score"]), 3), "sentiment_label": sent["label"], |
|
|
"sentiment_source": "openai" if (use_oa_sent and used_oa) else ("huggingface" if used_hf else "rules"), |
|
|
"themes": ", ".join(themes) if themes else "", "theme_counts_json": json.dumps(counts, ensure_ascii=False) |
|
|
}) |
|
|
|
|
|
out_df=pd.DataFrame(rows) |
|
|
nps=compute_nps(out_df["nps_score_final"]) |
|
|
dist=out_df["sentiment_label"].value_counts().to_dict() |
|
|
|
|
|
|
|
|
trs=[] |
|
|
for th, d in theme_agg.items(): |
|
|
trs.append({"theme":th,"total_mentions":int(d["mentions"]), |
|
|
"verbatims_pos":int(d["pos"]),"verbatims_neg":int(d["neg"]), |
|
|
"net_sentiment":int(d["pos"]-d["neg"])}) |
|
|
themes_df=pd.DataFrame(trs).sort_values(["total_mentions","net_sentiment"],ascending=[False,False]) |
|
|
|
|
|
|
|
|
method = "OpenAI + HF + règles" if (use_oa_sent and used_hf) else ("OpenAI + règles" if use_oa_sent else ("HF + règles" if used_hf else "Règles")) |
|
|
nps_label = "NPS global (inféré)" if any_inferred else "NPS global" |
|
|
lines=[ "# Synthèse NPS & ressentis clients", |
|
|
f"- **Méthode** : {method}", |
|
|
f"- **{nps_label}** : {nps:.1f}" if nps is not None else f"- **{nps_label}** : n/a" ] |
|
|
if dist: |
|
|
tot=sum(dist.values()); pos=dist.get("positive",0); neg=dist.get("negatif",0); neu=dist.get("neutre",0) |
|
|
lines.append(f"- **Répartition émotions** : positive {pos}/{tot}, neutre {neu}/{tot}, négative {neg}/{tot}") |
|
|
if not themes_df.empty: |
|
|
lines.append("\n## Thèmes les plus cités") |
|
|
for th,m in themes_df.head(5)[["theme","total_mentions"]].values.tolist(): |
|
|
lines.append(f"- **{th}** : {m} occurrence(s)") |
|
|
summary_md="\n".join(lines) |
|
|
|
|
|
if use_oa_summary: |
|
|
md = oa_summary(nps, dist, themes_df, oa_model, float(oa_temp or 0.0)) |
|
|
if md: summary_md = md + "\n\n---\n" + summary_md |
|
|
|
|
|
|
|
|
tmpdir=tempfile.mkdtemp(prefix="nps_gradio_") |
|
|
enriched=os.path.join(tmpdir,"enriched_comments.csv") |
|
|
themes=os.path.join(tmpdir,"themes_stats.csv") |
|
|
summ=os.path.join(tmpdir,"summary.md") |
|
|
out_df.to_csv(enriched,index=False,encoding="utf-8-sig") |
|
|
themes_df.to_csv(themes,index=False,encoding="utf-8-sig") |
|
|
with open(summ,"w",encoding="utf-8") as f: f.write(summary_md) |
|
|
zip_path=os.path.join(tmpdir,"nps_outputs.zip") |
|
|
with zipfile.ZipFile(zip_path,"w",zipfile.ZIP_DEFLATED) as z: |
|
|
z.write(enriched,arcname="enriched_comments.csv") |
|
|
z.write(themes,arcname="themes_stats.csv") |
|
|
z.write(summ,arcname="summary.md") |
|
|
|
|
|
|
|
|
def make_panels(dfT: pd.DataFrame): |
|
|
if dfT is None or dfT.empty: return "—","—","—" |
|
|
pos_top = dfT.sort_values(["verbatims_pos","total_mentions"], ascending=[False,False]).head(4) |
|
|
neg_top = dfT.sort_values(["verbatims_neg","total_mentions"], ascending=[False,False]).head(4) |
|
|
def bullets(df, col, label): |
|
|
lines=[f"**{label}**"] |
|
|
for _, r in df.iterrows(): lines.append(f"- **{r['theme']}** — {int(r[col])} verbatims") |
|
|
return "\n".join(lines) |
|
|
ench_md = bullets(pos_top, "verbatims_pos", "Points d’enchantement") |
|
|
irr_md = bullets(neg_top, "verbatims_neg", "Irritants") |
|
|
RECO_RULES = { |
|
|
"Délais & Suivi dossier": "Réduire les délais (SLA), suivi proactif.", |
|
|
"Cotisations / Facturation": "Clarifier factures, alerter anomalies.", |
|
|
"Espace client / App / Connexion": "Corriger login/MDP, QA navigateurs.", |
|
|
"Support / Conseiller": "Améliorer joignabilité, scripts, rappel auto.", |
|
|
"Communication / Transparence": "Notifications étapes clés, messages clairs.", |
|
|
"Sinistres / Indemnisation": "Transparence délais + suivi dossier.", |
|
|
} |
|
|
rec_lines=["**Recommandations**"] |
|
|
for _, r in neg_top.iterrows(): |
|
|
rec_lines.append(f"- **{r['theme']}** — {RECO_RULES.get(r['theme'],'Plan d’action dédié')}") |
|
|
return ench_md, irr_md, "\n".join(rec_lines) |
|
|
|
|
|
ench_md, irr_md, reco_md = make_panels(themes_df) |
|
|
fig_gauge = fig_nps_gauge(nps) |
|
|
fig_emots = fig_sentiment_bar(dist) |
|
|
k = max(1, int(top_k or 10)) |
|
|
fig_top = fig_top_themes(themes_df, k) |
|
|
fig_bal = fig_theme_balance(themes_df, k) |
|
|
|
|
|
return (summary_md, themes_df.head(100), out_df.head(200), [enriched, themes, summ, zip_path], |
|
|
ench_md, irr_md, reco_md, fig_gauge, fig_emots, fig_top, fig_bal) |
|
|
|
|
|
|
|
|
apply_plotly_theme() |
|
|
with gr.Blocks(title="Verbatify — Analyse NPS", css=VB_CSS) as demo: |
|
|
gr.HTML( |
|
|
"<div class='vb-hero'>" |
|
|
f"{LOGO_SVG}" |
|
|
"<div><div class='vb-title'>Verbatify — Analyse NPS</div>" |
|
|
"<div class='vb-sub'>Émotions • Thématiques • Occurrences • Synthèse</div></div>" |
|
|
"</div>" |
|
|
) |
|
|
|
|
|
with gr.Column(): |
|
|
pasted = gr.Textbox(label="Verbatims (un par ligne)", lines=10, |
|
|
placeholder="Exemple :\nRemboursement rapide, télétransmission OK | 10\nConnexion impossible à l’app | 3\nDélais corrects | 7") |
|
|
with gr.Row(): |
|
|
has_score = gr.Checkbox(label="J’ai un score NPS par ligne", value=False) |
|
|
sep = gr.Textbox(label="Séparateur score", value="|", scale=1) |
|
|
|
|
|
with gr.Row(): |
|
|
anon=gr.Checkbox(label="Anonymiser emails / téléphones", value=True) |
|
|
use_oa_sent=gr.Checkbox(label="OpenAI pour le sentiment", value=False) |
|
|
use_oa_themes=gr.Checkbox(label="OpenAI pour les thèmes", value=False) |
|
|
use_oa_summary=gr.Checkbox(label="OpenAI pour la synthèse", value=False) |
|
|
|
|
|
with gr.Row(): |
|
|
oa_model=gr.Textbox(label="Modèle OpenAI", value="gpt-4o-mini") |
|
|
oa_temp=gr.Slider(label="Température", minimum=0.0, maximum=1.0, value=0.1, step=0.1) |
|
|
top_k=gr.Slider(label="Top thèmes (K) pour les graphes", minimum=5, maximum=20, value=10, step=1) |
|
|
run=gr.Button("Lancer l'analyse", variant="primary") |
|
|
|
|
|
with gr.Row(): |
|
|
ench_panel=gr.Markdown(); irr_panel=gr.Markdown(); reco_panel=gr.Markdown() |
|
|
|
|
|
summary=gr.Markdown(label="Synthèse") |
|
|
themes_table=gr.Dataframe(label="Thèmes — statistiques") |
|
|
enriched_table=gr.Dataframe(label="Verbatims enrichis (aperçu)") |
|
|
files_out=gr.Files(label="Téléchargements (CSV & ZIP)") |
|
|
|
|
|
with gr.Row(): |
|
|
plot_nps = gr.Plot(label="NPS — Jauge") |
|
|
plot_sent= gr.Plot(label="Répartition des émotions") |
|
|
with gr.Row(): |
|
|
plot_top = gr.Plot(label="Top thèmes — occurrences") |
|
|
plot_bal = gr.Plot(label="Top thèmes — balance Pos/Neg") |
|
|
|
|
|
run.click( |
|
|
analyze_text, |
|
|
inputs=[pasted, has_score, sep, anon, use_oa_sent, use_oa_themes, use_oa_summary, oa_model, oa_temp, top_k], |
|
|
outputs=[summary, themes_table, enriched_table, files_out, ench_panel, irr_panel, reco_panel, plot_nps, plot_sent, plot_top, plot_bal] |
|
|
) |
|
|
|
|
|
|
|
|
apply_plotly_theme() |
|
|
with gr.Blocks(title="Verbatify — Analyse NPS", css=VB_CSS) as demo: |
|
|
gr.HTML( |
|
|
"<div class='vb-hero'>" |
|
|
f"{LOGO_SVG}" |
|
|
"<div><div class='vb-title'>Verbatify — Analyse NPS</div>" |
|
|
"<div class='vb-sub'>Émotions • Thématiques • Occurrences • Synthèse</div></div>" |
|
|
"</div>" |
|
|
) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
gr.HTML( |
|
|
'<div class="vb-footer">© Verbatify.com — Construit par ' |
|
|
'<a href="https://jeremy-lagache.fr/" target="_blank" rel="dofollow">Jérémy-lagache.fr</a></div>' |
|
|
) |
|
|
|
|
|
if __name__ == "__main__": |
|
|
demo.launch(share=False, show_api=False) |
|
|
|
|
|
|
|
|
if __name__ == "__main__": |
|
|
demo.launch(share=False, show_api=False) |
|
|
|
