AI-Powered Clustering and Prioritization of Feature Requests

AI System for Analyzing Feature Requests

Feature request backlog grows exponentially: companies with active product development receive 500 to 5000 tickets monthly from Jira, GitHub Issues, support tickets, and chats. A product manager manually groups similar requests — "I want a dark theme", "add dark mode", "why no night mode" — but these are the same. Manual clustering takes up to 20 hours per week, and up to 40% of duplicates go unnoticed. We automate this process with NLP and HDBSCAN, reducing analysis time by 70%. Savings on manual analysis reach $15,000 monthly for a volume of 1000 requests.

How HDBSCAN Helps Deduplicate Requests

The first task is deduplication. Instead of regular expressions and keywords, we use semantic clustering: each request is converted into a 768-dimensional embedding via Sentence Transformer (paraphrase-multilingual-mpnet-base-v2), then HDBSCAN groups them by cosine similarity. Noise (requests without close neighbors) is labeled -1 and excluded from clusters.

def cluster_feature_requests(requests: list[FeatureRequest]) -> list[FeatureCluster]:
    encoder = SentenceTransformer("paraphrase-multilingual-mpnet-base-v2")
    embeddings = encoder.encode([r.text for r in requests])

    clusterer = hdbscan.HDBSCAN(min_cluster_size=3, metric="cosine")
    labels = clusterer.fit_predict(embeddings)

    clusters = []
    for label in set(labels):
        if label == -1:  # noise — single requests
            continue
        cluster_requests = [r for r, l in zip(requests, labels) if l == label]
        clusters.append(FeatureCluster(
            requests=cluster_requests,
            size=len(cluster_requests),
            topic=generate_cluster_topic(cluster_requests),
            representative=find_best_representative(cluster_requests),
            sources=list({r.source for r in cluster_requests})
        ))

    return sorted(clusters, key=lambda c: c.size, reverse=True)

Result: instead of 1000 requests — 20-30 clusters with topics. Each cluster contains a representative (the most typical request) and a source list. Wikipedia: HDBSCAN provides additional information about the algorithm.

Comparison of HDBSCAN and K-means

HDBSCAN is on average 2 times more accurate than K-means in extracting semantic groups (F1-score 0.82 vs 0.41 on our test dataset).

What Is Priority Scoring?

Cluster size is important but not sufficient. We use scoring that takes into account:

• User segment: enterprise clients have weight ×2, free users ×0.5.
• Emotional tone: sentiment analysis via cardiffnlp/twitter-roberta-base-sentiment-latest — requests with negative tone (criticality, blocking) get +30% score.
• Churn correlation: if users who requested a feature later churned — it's a high priority signal.
• Business potential: estimate based on historical sales and NPS data.

Example: a cluster of 50 requests from enterprise clients with negative tone (words "blocks work") gets a score of 0.4×50 + 0.25×2 + 0.2×1.3 + 0.1×1 = 22.1, while a cluster of 200 requests from free users with neutral tone gets 0.4×200 + 0.25×0.5 + 0.2×1 + 0.1×1 = 80.25. However, accounting for business potential, the enterprise cluster may become a higher priority.

Generating User Stories and Trends

From the cluster, the system automatically generates a draft user story: "As [user type], I want [function] so that [value]." The user type is determined by the most frequent source in the cluster (e.g., if 80% of requests are from the "Admin panel" section — role "administrator"). The value is extracted from tone markers (words like "to", "for", "in order to"). The draft requires PM editing, but starting time is reduced from 40 minutes to 2.

We track dynamics: if over the last week more than 20% new requests have been added to a cluster — a "growing trend" flag. If a request has existed for more than 6 months without growth — low priority (deprioritize). Growth after a specific release — notification of possible regression. We use Wikipedia: HDBSCAN as the algorithm foundation.

Service Components

1. Audit of current request flow — assessment of volume, sources, frequency.
2. Ticketing system integration — connectors to Jira, GitHub, HubSpot, Zendesk.
3. NLP pipeline configuration — calibration of embeddings on your specifics (domain terms, slang).
4. Priority dashboard — web interface with clusters, trends, scoring.
5. User story export — CSV/JSON output for import into product management tools.
6. Documentation and team training — how to interpret trends and make decisions.

Timelines and How to Start

Project assessment takes 40 to 80 hours depending on integration complexity and data volume. The first prototype with basic clustering is ready 2 weeks after start. For an accurate estimate, get a consultation: simply write to us with a brief description of your current process and the number of requests per month. We guarantee that after implementation, you will spend no more than 2 hours per week on feature analysis.

Our team's experience — 5 years in NLP and MLOps, more than 30 successful implementations in product companies. We use only open-source components with no vendor lock-in.

Write to us — we'll evaluate your project free of charge and offer a turnkey solution.

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.