ReviewSignal: reading reviews in two languages

A dual-notebook system that turns raw customer reviews into decision-ready intelligence reports, in English and Spanish.

How it’s sold: fixed scope, fixed price, 5–7 day delivery, delivered as a shareable HTML report. A productized service, not an open-ended engagement. Core mission: Turn bulk product/service reviews into sentiment breakdowns, recurring complaint/praise themes, and concrete recommendations, without a human reading every review by hand.

Pipeline overview

Both language versions follow the same 7-step pipeline:

Load & normalize: ingest reviews (Amazon, Shopify, Trustpilot, MercadoLibre, a Kaggle dataset or a custom CSV), map columns to a platform-agnostic schema.
Sentiment labeling: star ratings as ground truth; sentiment model scores as a continuous signal.
Sentiment by product: surfaces which products/SKUs carry the most negativity.
Sentiment over time: tracks whether sentiment is trending up, down, or spiking.
Topic modeling: NMF extracts the top recurring complaint and praise themes.
Recommendations: quantifies each complaint theme and ties it to affected SKUs. Themes are ranked by frequency × sentiment severity, not raw counts.
Export: Markdown summary + supporting CSVs + a standalone HTML report.

English vs. Spanish

Aspect	English	Spanish
Sentiment model	VADER (lexicon-based)	nlptown/bert-base-multilingual-uncased-sentiment (neural)
Why	Fast, no model download	Native Spanish grammar/accent handling, context-aware
Stop words	English-optimized	Spanish-optimized

Logic highlight: topic extraction

vectorizer = TfidfVectorizer(stop_words=STOP_WORDS, ngram_range=(1, 2), max_df=0.95)
X = vectorizer.fit_transform(review_texts)

nmf = NMF(n_components=N_TOPICS, random_state=42, max_iter=500)
W = nmf.fit_transform(X)
H = nmf.components_

for topic_idx, topic in enumerate(H):
    top_words = [feature_names[i] for i in topic.argsort()[-8:][::-1]]
    # -> recurring theme, ranked by share of negative/positive reviews

Outputs

reviewsignal_report.md: sentiment mix, top 5 complaint themes, top 5 praised features, ranked recommendations
reviewsignal_report.html: standalone report with embedded charts
top_complaints.csv / top_praise.csv / recommendations.csv: structured tables per theme

Tech stack

NLP: VADER, Hugging Face Transformers (multilingual BERT), scikit-learn (TF-IDF, NMF)
Data: Pandas, NumPy, kagglehub for dataset ingestion
Reporting: Matplotlib, Seaborn, Jupyter/nbformat

Real-world note

Topic modeling needs volume to find genuine recurring patterns. The pipeline is built for bulk datasets (hundreds to tens of thousands of reviews), the scale a DTC/e-commerce brand generates on Amazon, Shopify, or Trustpilot. For a business with a handful of reviews, reading them directly beats statistical topic modeling. The tool is built for the scale where that stops being true.