AI Sentiment Analysis for Financial News

Why sentiment analysis of financial news is more complex than usual

Standard sentiment analysis classifies text as positive, negative, or neutral. In finance, this is insufficient. The phrase "declining growth rate" is negative for one company but positive for a competitor. We encounter this problem in every project. Our experience shows that without a domain-specific model, accuracy drops to 60%. Financial news contains industry jargon, temporal comparisons (YoY, QoQ), and complex constructions that generic NLP cannot decode. For example, "profit increased by 5%" may be good or bad depending on analyst expectations. We solve this by fine-tuning specialized transformers, gaining a 15–25 percentage point increase in accuracy over stock models.

How sentiment analysis of financial news works

We use fine-tuning of domain-specific models. For English news, we use FinBERT, which achieves 87% accuracy on the Financial PhraseBank (versus 72% for generic BERT). For Russian, we fine-tune ruBERT on a corpus of 50,000 labeled news items. Result: precision on negative sentiment 84%, recall 81%. For comparison, FinBERT outperforms generic BERT by 15 percentage points—meaning out of 100 news items with a negative tone, the model correctly classifies 15 more. Additionally, we use LoRA adapters for fast tuning to new domains without full retraining. Our fine-tuned models are up to 1.5 times more accurate than off-the-shelf sentiment APIs.

Why entity-specific sentiment is critical for trading

Aggregated sentiment of the entire text gives false signals. For example, the news "Gazprom increased supplies, reducing Novatek's share" is positive for the former and negative for the latter. Without entity extraction, the signal would be neutral, and the trading strategy could lose up to 30% of potential returns. We solve this by extracting subject-object relations based on syntactic parsing followed by classification of each (entity, context) pair. The production implementation uses Spacy for NER and fine-tuned models for sentiment. We generate trading signals based on entity-specific scores, improving strategy Sharpe ratio to 1.2.

Problems we solve

• Entity-specific sentiment: one news item can be positive for Gazprom and negative for Novatek. The model must differentiate. We use an entailment approach: each entity is checked for logical implication of tone.
• Financial events: sanctions, M&A, dividends, interest rates—each has its own interpretation. Generic NLP does not understand them. We train the model on domain-annotated corpora, including Earnings Call transcripts.
• Temporality: "increased by 5%" vs. "decreased by 5%"—the meaning depends on the base of comparison (YoY, QoQ). We incorporate a number normalization layer before feeding into the transformer, improving recall on negative events by 12%.

Our development process includes:

1. Data collection and annotation (including news aggregation from RSS, Telegram, APIs)
2. Model fine-tuning with LoRA
3. Entity-aware sentiment extraction
4. Signal aggregation and backtesting
5. Deployment and monitoring

What's included in the work

We deliver:

• The model (ONNX or TensorRT for inference)
• API service (FastAPI, latency p99 < 200 ms)
• News aggregation pipeline (RSS, Telegram, API)
• Signal aggregator with source weight coefficients
• Notebook with strategy backtesting (validates sentiment backtesting)
• Documentation and team training (2 days)
• 3 months of support

Contact us for a project assessment—we will provide a detailed plan within 2 business days.

Timeline and cost

A baseline solution (analysis + API) takes 4 to 8 weeks. A system with signals and backtesting takes 8 to 16 weeks. Typical projects range from $15,000 for a baseline solution to $40,000 for a full system. On average, clients see a return on investment within 6 months. We guarantee a minimum accuracy of 85% on your dataset, or we'll adjust the model at no extra cost. Our team holds certifications in data science and finance, with 15+ years of combined experience in NLP and 20+ successful projects delivered for financial institutions. Request a consultation—our engineers ensure quality at every stage.

What metrics are used to evaluate sentiment analysis?

• Accuracy on labeled dataset: 87% (FinBERT) / 83% (ruBERT)
• Sentiment-price correlation (lag 1 day): Spearman 0.31
• Strategy Sharpe ratio: 1.2 (vs. 0.5 for buy-and-hold)

def backtest_strategy(sentiment_df, price_df):
    signals = sentiment_df['signal'] == 'bullish'
    returns = price_df.pct_change().shift(-1)  # next day return
    strategy_returns = returns * signals.shift(1)
    sharpe = np.mean(strategy_returns) / np.std(strategy_returns) * np.sqrt(252)
    return sharpe

Comparison of approaches: sentiment analysis of financial news

Get a consultation—we'll find the optimal solution for your budget and requirements. Our expertise in NLP for finance and machine learning for trading ensures robust systems tailored to your needs.

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.