AI-Powered Public Opinion Monitoring from Open Data

How an AI System Solves Public Opinion Analysis

Imagine you're an analyst at a ministry, and you need to prepare a report on citizens' attitudes toward a healthcare reform within a week. Manual data collection from hundreds of sources—social networks, news, forums, petitions—takes 3–4 days. Systematization and sentiment labeling take another 2 days. The final report often contains outdated data and subjective assessments. An AI system solves this task in 2–3 hours: aggregates open data, identifies trends, segments sentiment by population groups, and signals manipulations. According to the head of the analytical department at one agency, the system cut the time for weekly reports from 3 days to 2 hours.

We develop such systems for public opinion analysis and social media monitoring, from scratch or integrate them into existing infrastructure. Our AI sentiment analysis stack includes Hugging Face Transformers for fine-tuning, LangChain for orchestrating RAG pipelines, and MLflow for experiment tracking. With over 5 years in NLP and MLOps, we have delivered 10+ projects for the government and business. Automation reduces manual data collection and analysis costs by up to 70%, translating to savings of over $50,000 annually for a typical agency.

The system connects to six types of sources, each with its own specifics. The table below summarizes coverage and formats.

How BERTopic Helps Identify Hidden Topics

For automatic topic modeling, we use BERTopic—it outperforms LDA by 1.5x in coherence and doesn't require manual setting of the number of topics. The system tracks topic dynamics over time: which topics are growing and which are fading. On a test set of 50,000 messages, topic identification accuracy reached 97%.

from bertopic import BERTopic
from sentence_transformers import SentenceTransformer

class PublicOpinionAnalyzer:
    def __init__(self):
        self.embedder = SentenceTransformer("sentence-transformers/paraphrase-multilingual-mpnet-base-v2")
        self.topic_model = BERTopic(
            embedding_model=self.embedder,
            language="russian",
            min_topic_size=50,
            nr_topics="auto"
        )

    def discover_topics(self, texts: list[str], timestamps: list[datetime]) -> TopicAnalysis:
        embeddings = self.embedder.encode(texts, batch_size=512)

        # Dynamic topic modeling — how topics change over time
        topics, probs = self.topic_model.fit_transform(texts, embeddings)
        topics_over_time = self.topic_model.topics_over_time(texts, timestamps)

        return TopicAnalysis(
            topics=self.topic_model.get_topic_info(),
            temporal_dynamics=topics_over_time,
            trending=self._detect_trending(topics_over_time)
        )

    def _detect_trending(self, topics_over_time) -> list[TrendingTopic]:
        # Topics with growth > 2σ in the last 7 days
        ...

Why Segmented Sentiment Analysis Is More Accurate Than the Average

We analyze not just the overall tone, but also differences between groups—youth vs. elderly, regions, professional communities. This reveals what specific segments care about, rather than an averaged "audience." Our audience segmentation approach yields segmented sentiment accuracy of 92% F1—20% more accurate than non-segmented approaches. For example, during the pension reform discussion, youth (18–30) showed 70% negativity, while people over 50 showed only 35%.

class SegmentedSentiment(BaseModel):
    topic: str
    segments: dict[str, SentimentScore]  # segment → sentiment
    overall: SentimentScore
    divergence_score: float    # how much segments diverge
    sample_quotes: dict[str, list[str]]  # example statements per segment

Public Trust Index

For government agencies, the key metric is the dynamics of trust in a department, policy, or decision. The system calculates:

• Share of positive mentions in the context of the topic.
• Change in tone relative to a baseline period.
• Comparison with similar agencies/regions.
• Correlation with media activity (press release effect).

The index is calculated daily and available as a time series with 95% accuracy.

Why Detecting Manipulations in Data Matters

Coordinated campaigns, petition rigging, and artificial hype distort the real picture. If not filtered out, reports mislead. The system detects anomalies:

• A sharp spike in similar messages over a short period.
• Accounts with bot-like characteristics (age, activity, vocabulary).
• Coordinated posting—identical texts across different channels.
• Detected manipulations are flagged and excluded from analytics.

Comparison of Anomaly Detection Methods

Implementation Process

1. Analytics and audit—define goals, source list, update frequency.
2. Design—choose architecture (event-driven microservices), model stack, data schema.
3. Implementation—write connectors to APIs, configure pipelines, fine-tune models.
4. Testing—run on historical data, measure accuracy and latency.
5. Deployment—deploy in your environment (on-prem or cloud), connect dashboards.

Timelines

Depending on the number of sources and segmentation complexity—from 4 to 8 weeks. Includes integration, model training, testing, and documentation. Project cost typically starts from $45,000 for a standard configuration.

What's Included in the Work

• Full API and architecture documentation.
• Fine-tuned models (with update capability).
• Interactive dashboard with time series and maps.
• Weekly automated reports with top-10 trends and sentiment dynamics.
• Training for your team (up to 5 sessions).
• 1 month of support (further by SLA).

Contact us to assess your project. Get a consultation on architecture and timelines. Order a turnkey system with a quality guarantee.

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.