ABSA for Reviews: Detailed Aspect Sentiment Analysis with LLMs

How to Extract Actionable Insights from Reviews Using Aspect-Based Sentiment Analysis

You run a hotel chain and receive thousands of reviews monthly. Overall sentiment is 4.2 stars, but you don't know what to improve. Guests write: "Room is clean, but Wi-Fi is awful and breakfast is mediocre." Three aspects, three opinions — yet you only have an average score. ABSA (Aspect-Based Sentiment Analysis as defined on Wikipedia) extracts these details and turns them into actionable insights.

Our team has deployed ABSA for 15 clients in e-commerce, HoReCa, and logistics — over 2 million reviews processed, average project time is 3 weeks. The approach is straightforward: LLM + structured output + aggregation. In this article, we break down how this works in production and why generic solutions fall short.

Why ABSA Beats General Sentiment

Without ABSA, you lose context. For example, the phrase "staff is friendly but queues are horrible": overall positive, but the aspect 'service speed' is negative. Only aspect-based analysis sends a signal to the manager: "increase cashiers during peak hours." According to our data, 68% of negative reviews contain at least one positive aspect — without ABSA these signals are lost.

How LLM Solves the ABSA Task

Classic approaches (CRF, BiLSTM, BERT fine-tuning) require labeled corpora. For a new domain, that means weeks of annotation. LLM with structured output solves it faster: one prompt yields JSON with aspects and sentiments. Our approach is 3x faster to deploy than custom ML models.

Here's the Pydantic model we use in production:

class AspectSentiment(BaseModel):
    aspect: str
    sentiment: Literal["positive", "negative", "neutral", "mixed"]
    opinion_phrase: str
    score: float  # -1.0 to 1.0

class ReviewAnalysis(BaseModel):
    aspects: list[AspectSentiment]
    overall_sentiment: Literal["positive", "negative", "neutral", "mixed"]
    key_issues: list[str]
    key_positives: list[str]

The GPT-4o model outputs this JSON with ~92% accuracy on Russian-language reviews (based on our tests on 5,000 examples). For sensitive data we use LLaMA 3 70B, hosted via vLLM — latency p99 under 800 ms.

Step-by-step process:

1. Define aspects relevant to your domain (e.g., cleanliness, staff, price).
2. Craft few-shot prompts with 10–15 examples per aspect.
3. Use chain-of-thought reasoning to improve accuracy.
4. Parse LLM output into a validated JSON schema.
5. Aggregate results across reviews and visualize trends.

Why Aggregation Matters More Than Single Analysis

One review is noise. Thousands are signal. We build a pipeline that collects results across all reviews for a period and aggregates them:

Trends reveal that Wi-Fi is deteriorating — a signal to replace routers. The rising breakfast sentiment shows the new chef is doing well.

We also build time-based aspect heatmaps: you can see how the 'cleanliness' aspect dropped after a cleaner change. Automatic alerts: if the sentiment for 'delivery speed' falls below a threshold, a Slack notification goes to the logistics manager.

How We Ensure Analysis Accuracy

For high accuracy we use few-shot prompts with 10–15 examples per domain and chain-of-thought reasoning. Each result undergoes schema validation. For critical industries, we calibrate confidence — the model returns a score from -1 to 1, and we discard predictions with low confidence (<0.3). This yields precision up to 96% on key aspects.

Comparison: Open Source vs GPT-4o

The choice depends on your privacy and budget requirements. We help select the optimal option. GPT-4o is 2x more accurate than open-source models for rare aspects.

What Our Work Includes (Deliverables)

Our deliverable package includes:

• Consultation and audit of your current review flow: channels (Google Maps, Yandex, Otzovik, CRM), format (CSV, API, JSON)
• Stack selection: OpenAI / open-source model, vector DB (Qdrant or pgvector) for embedding storage
• Prompt development with few-shot and chain-of-thought for high accuracy
• Integration with your dashboard (Metabase, PowerBI, Tableau) or our default draft on Plotly
• Documentation and team training: how to interpret the aspect radar and trends
• 3-month support guarantee: bug fixes, aspect adjustments for new seasons
• Access to dashboards and alerting system

How to Start?

Describe your task — we'll propose an architecture suited to your budget and timeline. A full implementation costs between $8,000 and $12,000, with a typical ROI of 3 months. We evaluate the project in 2 business days. Result: JSON schema, prototype on 50 reviews, and roadmap to production. Contact us — we'll show how ABSA solves your specific pain points. Get a consultation on ABSA architecture for your business.

NLP Development: Text Classification, NER, Embeddings, and Information Extraction

We often receive a task: process 50,000 support tickets — currently all manual. Dataset — 3,000 labeled examples, 12 categories, imbalance: one category occupies 40% of the sample, three at 1-2% each. Baseline accuracy — 78%. Sounds decent until you look at recall for rare classes: 0.31, 0.44, 0.28. These classes — complaints and churn threats — are most important to the business.

This is a typical NLP development project. The problem is not the algorithm but that accuracy is the wrong metric. Our experience across 30+ projects shows: we start by analyzing business metrics and only then choose the model.

Why accuracy is not the right metric for rare classes?

Accuracy ignores imbalance. If the "churn" class appears in 2% of cases, the model can predict "all good" and get 98% accuracy — but the business loses clients. Solution: F1 macro (averaged over all classes) or weighted F1. For NER — strict entity F1 (exact matches only). We guarantee: after choosing the correct metric, model quality becomes measurable and predictable.

Text Classification: From BERT to Distillation

BERT-like models are the standard for classification. ruBERT-base or ruBERT-large from DeepPavlov for Russian. multilingual-e5-large — for multiple languages in one pipeline. XLM-RoBERTa-large — a strong multilingual backbone.

Fine-tuning for classification: add a classification head on top of the [CLS] token, train for 3-5 epochs with lr=2e-5, weight decay=0.01. For imbalance — weighted CrossEntropyLoss or focal loss with gamma=2.0. Contact us — we will show a code snippet.

Imbalance case study. Dataset — 3,000 examples, imbalance 1:20. Solution: class_weight via sklearn + CrossEntropyLoss. Additionally — augmentation of rare classes via backtranslation (ru→en→ru through MarianMT). Recall for rare classes rose from 0.31 to 0.67 with a slight drop in accuracy (76%→74%). Full NLP development end-to-end took 3 weeks.

Distillation for production. BERT-large gives F1 0.89, but inference on CPU — 180ms. Distillation into DistilBERT or ruBERT-tiny2 reduces latency to 25ms with F1 0.84. Export to ONNX Runtime provides an additional 1.5-2x speedup. DistilBERT achieves 7x lower latency than BERT-large with only a 5% drop in macro F1 – a typical production trade-off.

How to choose between BERT and LLM for your task?

For most classification and extraction tasks, BERT-sized models offer the best trade-off between cost and performance. Shift to LLMs only when the task demands generation, complex reasoning, or zero-shot generalization.

NER: Named Entity Recognition

NER — extracting persons, organizations, locations, dates, amounts, document numbers. For general categories (PER, ORG, LOC), pre-trained models work well. For specialized ones (medical terms, legal concepts) — fine-tuning is needed.

Data annotation. The main cost of an NER project. For a quality model — 500-2,000 labeled sentences per entity type. Tools: Label Studio (open source) or Prodigy (by spaCy creators). IOB2 format — standard.

Architecture. Token classification on top of BERT: each token gets a label (B-PER, I-PER, O). spaCy 3.x with transformer pipeline — a convenient production choice.

Nested entities. Standard IOB models cannot handle nested entities (organization inside an address). For such tasks — span-based NER: SpanBERT or SpERT. More complex but correct.

Post-processing is mandatory. The model predicts tokens — normalized entities are needed. Date — dateparser. Amounts — regex + validation. Names — deduplication via rapidfuzz. Included in our standard delivery.

Sentiment Analysis and Opinion Mining

Binary classification positive/negative works out of the box with BERT. Complexity — aspect-based sentiment analysis (ABSA): "the restaurant has good food but terrible service." For ABSA: aspect extraction (NER) + sentiment per aspect. Joint models BERT-for-ABSA — quality on Russian data is lower due to dataset scarcity. RuSentiment, SentiRuEval — main resources.

For production with simple positive/negative/neutral: distil models are enough. Three classes, balanced dataset, 2,000+ examples — F1 macro 0.82-0.87 in 1-2 days.

Text Summarization

Extractive summarization (select sentences) — TextRank or BM25 without training. Fast, no hallucinations. Good for long documents.

Abstractive (generates new text) — seq2seq: mT5, mBART, FRED-T5, ruT5-large. For production via LLM API (GPT-4, Claude) — often the best cost/quality/speed trade-off.

Embeddings: Vector Representations of Text

Embeddings are the foundation of semantic search, deduplication, clustering, RAG. Quality critically affects downstream tasks.

Models. E5-large-v2, BGE-M3, multilingual-e5-large — strong multilingual embedders. sentence-transformers/paraphrase-multilingual-mpnet-base-v2 — fast option. For Russian: ru-en-RoSBERTa (Skoltech) performs well on semantic textual similarity.

Embedding quality evaluation uses the MTEB benchmark as standard. But top results on MTEB don't guarantee success on a domain dataset — we build domain-specific eval.

Fine-tuning embeddings. If standard models don't give the required Recall@k — contrastive learning on domain pairs with MultipleNegativesRankingLoss. How to perform this for domain data:

1. Collect 500–2,000 semantically similar pairs from your domain.
2. Apply MultipleNegativesRankingLoss with a batch size of 32–64.
3. Train for 1–3 epochs using AdamW (lr=2e-5).
4. Evaluate Recall@k on a held-out domain test set.

This approach yields a 5–15% improvement in Recall@k in practice.

Dimensionality and storage. E5-large: 1024 dim, float32 — 4KB per vector. For 10M documents — 40GB. Quantization int8 reduces to 10GB. FAISS IVF_PQ — more compact but with losses. Included in our deployment recommendations.

Information Extraction

Structured extraction is a frequent task. Examples: key contract terms, technical characteristics, dates and amounts from invoices.

1. Regex + rule-based. For INN, OGRN, amounts, dates — more reliable than neural networks. No data required.
2. NER + post-processing. For variable formats.
3. LLM with structured output. GPT‑4 / Claude with JSON schema — for complex documents. Cost: minimal per document. For 10k+ documents/day — we calculate the economics.

We guarantee a hybrid: regex/NER for typical fields + LLM for edge cases. Our guarantee is backed by years of production experience and more than 30 projects.

Work Stages

Prototype on existing data — 1-3 weeks. Production system with CI/CD — 1.5–2.5 months. Cost is calculated individually — get a consultation for a project estimate.

What's Included

• Model and pipeline architecture documentation
• Access to the model via REST API (FastAPI + ONNX)
• Client team training (2-hour webinar + Q&A)
• Accuracy guarantee on the agreed test set
• Months of post-delivery support (bug fixes, adaptation to new data)

Our Experience

Years of NLP projects from classification to RAG systems. The team includes ML engineers experienced with Hugging Face, spaCy, LangChain, MLOps. We use vLLM, Kubeflow, Weights & Biases — a production stack, not toys. Contact us to evaluate your NLP project within two days — request a free consultation on your text processing pipeline.