Relation Extraction Development for Semantic Link Extraction

Parsing corporate news often reveals relationships like 'Sberbank appointed German Gref as chairman.' Named Entity Recognition (NER) finds entities, but relation extraction (RE) identifies the 'appointment' connection, converting unstructured text into triples (subject, relation, object). In a project for a legal department, we processed 100,000 court rulings: manual labeling would have taken six months and cost $50,000, but fine-tuning BERT with Distant Supervision achieved F1 78% in 3 weeks for $15,000. Our engineers have delivered over 15 RE projects for financial and legal corpora, including building knowledge graphs. With more than 5 years of NLP experience, we guarantee F1 no lower than 75% on standard benchmarks such as TACRED, DocRED, and SemEval. RE systems already save companies up to 70% of time on contract and judgment analysis. The RE approach is detailed on Wikipedia.

How Relation Extraction Extracts Semantic Links

NER only identifies entities, but does not determine the type of connection between them. Relationship extraction (RE) extends NER: the output is semantic triples like (Sberbank, appoints, German Gref). Such triples are the foundation for building knowledge graphs and automatic text analysis. We use open benchmarks TACRED and DocRED for validation — on these, fine-tuned BERT consistently yields F1 75–80%.

How Fine-Tuning BERT Outperforms Prompt-Based LLMs

Prompt-based approaches with LLMs (e.g., GPT-4) can get RE running in a day, but on large data volumes inference cost is high — latency p99 can reach 1 second per triple. Fine-tuned BERT with entity-marker tokens ([E1]entity[/E1]) achieves F1 10–15% higher with latency 10–50 ms. Comparison of approaches:

Fine-tuned BERT is 10 times cheaper at inference and 10–15% more accurate — the optimal choice for high-load systems. Metrics are strict: an answer is considered correct only if entities, direction, and relation type match exactly.

How Distant Supervision Reduces Labeling Costs

Labeling data for RE is an expensive step. Distant Supervision automatically creates a training set by aligning texts with a knowledge base (e.g., Wikidata). This yields 10 times more examples for the same cost as manual labeling, with some loss in accuracy (5–8% F1). To compensate for noise, we use weighted loss and confidence filtering.

Comparison of labeling methods:

RE Implementation Process

1. Corpus audit and relation schema definition — determine the list of relations and verify NER quality.
2. Approach selection — based on data volume and latency requirements, choose prompt-based, fine-tuning, or REBEL.
3. Labeling — if labeled data is scarce, apply Distant Supervision.
4. Training and validation — tune hyperparameters, monitor F1 on held-out set.
5. Deployment — package model into Docker, API on FastAPI with metrics and logging.
6. Support — train your team, provide 3 months of maintenance.

What's Included

• Trained RE model with target metrics
• API service on FastAPI in a Docker container
• Documentation: architecture description, fine-tuning guide, API specification
• Source code and configs under version control
• Training for your team (2–3 sessions)
• Technical support for 3 months after deployment

Common Mistakes and How to Avoid Them

• Missing long-tail relations: rare types introduce noise in distant supervision. Solution — balance the dataset and use weighted loss.
• Entity errors: if NER is inaccurate, RE inherits the errors. We recommend a cascade with intermediate validation.
• Ignoring context: the same word may indicate different relations in different domains. Fine-tuning on the target corpus solves the problem.

To find out which approach fits your corpus, contact us for a free audit. We are ready to discuss details. Get a consultation on choosing the approach for your corpus — it's free.

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.