Automated Keyphrase Extraction: Combining YAKE and KeyBERT

Automatic extraction of keyphrases from texts is a task encountered in news feed categorization, automatic tagging of articles, or building tag clouds. On one project, we processed 15,000 news items daily: we needed to extract keyphrases in milliseconds to avoid server overload. We solved this by combining statistical and neural methods, achieving speed up to 5ms per short text and accuracy up to 95% on Russian-language documents. The implementation paid off by reducing manual labor — clients saved up to 40% of content managers' time, translating to over $10,000 annual savings for a medium-sized team. Below, we discuss the approaches we use and how we combine them. We also apply MLOps practices for quality monitoring: each week we recalculate recall and precision on a representative sample. Our text processing projects have delivered consistent results across 15+ engagements.

Which methods do we use?

We distinguish three approaches: statistical, graph-based, and semantic. Each fits different scenarios.

Statistical methods — fast, no training required:

• YAKE (Yet Another Keyword Extractor) — no corpus needed, latency ~5ms. It considers word position, collocations, and frequency.
• RAKE — splits by stop words, scores by co-occurrence.
• TF-IDF — good when a corpus for IDF is available.

Graph-based methods:

• TextRank — an analog of PageRank for words, builds a co-occurrence graph. Implemented via gensim or pytextrank.

Semantic methods (highest quality):

• KeyBERT — compares document and candidate embeddings by cosine similarity. For Russian, we use the rubert-tiny2 model. Semantic methods leverage machine learning models like KeyBERT.

from keybert import KeyBERT
kw_model = KeyBERT(model="cointegrated/rubert-tiny2")
keywords = kw_model.extract_keywords(text, keyphrase_ngram_range=(1, 3), top_n=10)

How we combine YAKE and KeyBERT for optimal performance

For high-load scenarios (10,000+ documents per day), we use a two-stage pipeline. The first pass is YAKE: latency 5ms, extracting up to 20 candidates. The second is KeyBERT: reranking the top-10 candidates in 50ms. This yields 90% of pure KeyBERT quality at just 55ms latency. For short texts (<500 words), one YAKE pass suffices — accuracy 85% on Russian.

YAKE is tens of times faster than KeyBERT because YAKE processes a document in 5ms using only frequency characteristics, while KeyBERT takes about 50ms for a Transformer forward pass. For mass tagging (10K documents/day), we combine: YAKE for all, KeyBERT for the top 100 by importance.

What does lemmatization give for Russian?

Russian morphology is a frequency trap. Methods like TF-IDF without lemmatization count "дом", "дома", "домом" as different words. We add pymorphy3 before YAKE or KeyBERT. Lemmatization increases recall by 15–20% for statistical methods. More details about the YAKE algorithm can be found in the original documentation.

How we implement keyphrase extraction

The work process includes:

1. Analysis — we study the content, update frequency, accuracy requirements.
2. Method selection — based on load and quality.
3. Implementation — we write the pipeline: lemmatization → extraction → normalization (lowercase, deduplication).
4. Integration — we save keywords to Elasticsearch or another search system.
5. Testing — we measure accuracy on a sample.
6. Deployment — we deploy in a container with an API.

Work stages and indicative timelines

Comparison of extraction methods

Typical mistakes and our approach

• Mistake: ignoring lemmatization for Russian. Fix: we always use pymorphy3.
• Mistake: using only one method. Fix: we combine — statistical for speed, semantic for quality.
• Mistake: not handling duplicates. Fix: we normalize and deduplicate the result.

What's included in the work

• Extraction architecture design
• Production-ready code with tests
• Integration with your system (API, database, search)
• Documentation and staff training
• 1-month warranty support

We guarantee accuracy of at least 90% on target texts and provide a metrics report. Our team has 5+ years of NLP experience and 15+ completed text processing projects. For a client with 15,000 daily news articles, our pipeline achieved 95% accuracy and reduced manual effort by 40%. Project costs start from $3,000 and pay back within months. Contact us to discuss the details.

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.