Automatic Fact Extraction: LLM and NLP Pipelines

Every day, companies process thousands of documents: contracts, invoices, reports, news. Manual entity search — parties, amounts, dates — takes hours and yields 70–80% accuracy. A typical error: missing a date or misattributing a counterparty leads to fines and missed deadlines. Information extraction (IE) solves this: it extracts entities, relations, and events with up to 95% accuracy. Fact extraction from text is no longer a chore.

We automate the full cycle: from parsing to normalization. We use LLMs (GPT-4o, Claude, LLaMA 3) and classic NLP pipelines. We choose the approach based on the task — or combine them. A hybrid architecture achieves F1 up to 95% with latency under 200 ms. Average savings on manual processing reach 60% of costs — or 500 person-hours per month.

What Typical Errors Occur in Data Extraction?

When extracting from contracts, company names are often confused due to typos or abbreviations — "LLC Romashka" and "Romashka LLC" are treated as different. Normalization via edit distance (Levenshtein) and synonym dictionaries solves this. Entity duplicates — the same information appears in different documents with different attributes. We use entity linking through a knowledge graph. Finally, context ambiguity: the word "account" can be a bank account or an invoice. Disambiguation via LLM with few-shot examples.

Why LLM Outperforms Classic Pipelines?

LLMs with structured output (Pydantic, OpenAI function calling) handle extraction without fine-tuning. Example:

from pydantic import BaseModel
from openai import OpenAI

class CompanyInfo(BaseModel):
    name: str
    revenue: float | None
    revenue_year: int | None
    ceo: str | None
    headquarters: str | None
    employees_count: int | None

client = OpenAI()
response = client.beta.chat.completions.parse(
    model="gpt-4o-mini",
    messages=[{
        "role": "user",
        "content": f"Extract company information from the text:\n{text}"
    }],
    response_format=CompanyInfo,
)
result = response.choices[0].message.parsed

This code works for any text — from financial reports to news. No annotation, no pipeline. But for >1000 doc/hour and latency <100 ms, classic pipelines are cheaper and faster. On a marketplace project, we extracted 500k+ product cards from 1 million PDFs. We used a combination of OCR (Tesseract) + LayoutLM for tables + GPT-4o for attribute extraction. F1 on the test set: 93%.

Comparison of Approaches

How Fact Extraction from Text Solves Document Processing

Fact extraction from text is a key step in building knowledge bases. In a typical project, we go through five stages:

Timeline: from 2 weeks to 2 months depending on complexity. Cost is calculated individually.

What's Included in the Work?

• Analytical report with extracted entities and relations
• Trained model or pipeline (LLM or classic)
• REST API with documentation (OpenAPI)
• Integration with CRM (Bitrix24, 1C, etc.)
• Employee training (up to 5 days)
• Technical support for 3 months

Step-by-Step Implementation Process

1. Data audit — collect document samples, define target entities and quality metrics
2. Architecture selection — test LLM, BERT, spaCy on your data, establish baseline
3. Pipeline development — write code, configure inference, optimize latency
4. Testing — calculate Precision/Recall/F1, A/B tests, adjustments
5. Deployment — containerization, deploy on your infrastructure or cloud
6. Support — monitor metrics, retrain when data changes

Extraction Quality Assessment

Precision/Recall/F1 metrics are calculated for each entity type. For relations — relation-level F1, for slots — slot filling accuracy. Typical results: 90–95% F1 on reports, 75–85% on news.

Detailed metrics: F1 is the harmonic mean of Precision and Recall. For each entity type we compute separately. If recall is critical (e.g., finding all mentions), we tune the pipeline for higher recall (up to 98%) at the expense of precision.

We guarantee: fixed quality metrics in the contract, certified engineers with 10+ projects of experience (NLP, Computer Vision, LLM), post-deployment support. If you want to automate data extraction from your documents, contact our engineers for a consultation. Order a pilot project without prepayment — we'll assess your case in 2 days.

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.