Sentiment Forecasting Engineer
When to Use
- Build aggregate sentiment indices from high-volume text streams (social, news, reviews, surveys)
- Design temporal rollups — hourly, daily, weekly aggregation with consistent weighting rules
- Forecast opinion trajectories — point forecasts, prediction intervals, and scenario bands
- Model leading/lagging relationships between sentiment and sales, traffic, volatility, or brand KPIs
- Select and implement time-series and sequence models — ARIMA, Prophet, state-space, TFT, etc.
- Run nowcasts and choose forecast horizons aligned to decision cadence
- Engineer features from volume, velocity, topic mix, and engagement quality
- Backtest with walk-forward validation and report calibration of uncertainty
- Handle spikes, bot noise, sample bias, and regime shifts in language or product mix
- Integrate outputs with BI dashboards, brand monitoring, or research workflows (methodology only)
When NOT to Use
- Per-document or per-span polarity labeling, annotation, or classifier training →
sentiment-analysis-engineer
- Generic demand, inventory, or logistics forecasting without sentiment inputs →
predictive-logistics-developer, data-scientist
- Investment advice, trade recommendations, or actionable trading signals → provide forecasting methodology and uncertainty only
- Marketing copy, campaigns, or brand voice →
content-creator, brand-voice-enforcement
- Broad macro econometrics or financial modeling without text-derived sentiment →
financial-analyst (partial overlap only)
- Exploratory NLP or single-shot sentiment scores on a static corpus →
sentiment-analysis-engineer
- LLM product features, agents, or RAG (unless sentiment forecasting is one pipeline component) →
ai-engineer
Related skills
| Need |
Skill |
| Document-level polarity, ABSA, annotation, classifier eval |
sentiment-analysis-engineer |
| General ML, experimentation, non-time-series predictive modeling |
data-scientist |
| Warehouse metrics, dbt, analytics pipelines (if present in repo) |
analytics-engineer |
| Demand/inventory forecasting without opinion indices |
predictive-logistics-developer |
| Campaign ROI and channel performance (if present in repo) |
marketing-analyst |
| Ratios, valuation, macro series without text sentiment (if present) |
financial-analyst |
| LLM apps, feature stores for agent products |
ai-engineer |
Core Workflows
1. Scope and index design
Clarify population (brand, product, geo), text sources, aggregation grain, target horizon, and downstream KPIs.
See references/sentiment_forecasting_engineer_scope.md.
2. Indices, aggregation, and features
Define index formulas, rollups, topic/strata splits, and covariates (volume, velocity, mix).
See references/indices_aggregation_and_features.md.
3. Time-series and forecast models
Choose baselines and advanced models; align seasonality, holidays, and exogenous drivers.
See references/time_series_and_forecast_models.md.
4. Backtesting, validation, and metrics
Walk-forward evaluation, interval calibration, and spike-event holdouts.
See references/backtesting_validation_and_metrics.md.
5. Data quality, bias, and events
Bot filtering, sample bias, language drift, and shock labeling for scenario analysis.
See references/references_data_quality_bias_and_events.md.
6. Production monitoring and stakeholders
Serving cadence, drift monitors, dashboard contracts, and stakeholder-ready narratives.
See references/production_monitoring_and_stakeholders.md.
Outputs
- Index specification — formula, universe, weights, strata, and revision policy
- Feature catalog — engineered signals with definitions and lag structure
- Forecast spec — horizon, frequency, model family, and exogenous inputs
- Backtest report — walk-forward metrics, interval coverage, and failure slices
- Nowcast playbook — latency budget, refresh rules, and stale-data handling
- Monitoring plan — drift, spike alerts, and human review triggers
- Stakeholder brief — trajectory narrative with explicit uncertainty (no trade advice)
Principles
- Forecast aggregates, not individual opinions — index stability and definitional clarity come first
- Treat index construction as part of the model — changing weights invalidates historical comparability
- Prefer walk-forward evaluation over single holdout splits for time-ordered data
- Report intervals and scenarios, not point estimates alone; disclose coverage on backtests
- Separate methodology from decisions — do not present forecasts as buy/sell or guaranteed outcomes
- Document known biases (platform mix, bot share, demographic skew) beside every published index
1---2name: sentiment-forecasting-engineer3description: This skill should be used when the user asks to forecast aggregate sentiment and opinion dynamics over time—sentiment indices from text streams; temporal rollups; leading/lagging KPI links; time-series and sequence models (ARIMA, Prophet, state-space, ML); nowcasting; spikes, bots, and bias; walk-forward backtests; intervals and scenarios; volume/velocity/topic features; BI or brand dashboards. Triggers: sentiment forecasting, forecast sentiment, sentiment index, opinion trend forecast, social sentiment time series, brand sentiment trajectory, nowcast sentiment, sentiment leading indicator, aggregate polarity forecast, sentiment backtest, walk-forward sentiment, sentiment spike prediction. Not for per-text labeling (sentiment-analysis-engineer), demand forecasting without sentiment (predictive-logistics-developer, data-scientist), trade advice (methodology only), marketing copy (content-creator), macro without text sentiment (financial-analyst partial).4---56# Sentiment Forecasting Engineer78## When to Use910- Build **aggregate sentiment indices** from high-volume text streams (social, news, reviews, surveys)11- Design **temporal rollups** — hourly, daily, weekly aggregation with consistent weighting rules12- Forecast **opinion trajectories** — point forecasts, prediction intervals, and scenario bands13- Model **leading/lagging relationships** between sentiment and sales, traffic, volatility, or brand KPIs14- Select and implement **time-series and sequence models** — ARIMA, Prophet, state-space, TFT, etc.15- Run **nowcasts** and choose forecast horizons aligned to decision cadence16- Engineer features from **volume, velocity, topic mix**, and engagement quality17- **Backtest** with walk-forward validation and report calibration of uncertainty18- Handle **spikes, bot noise, sample bias**, and regime shifts in language or product mix19- Integrate outputs with **BI dashboards**, brand monitoring, or research workflows (methodology only)2021## When NOT to Use2223- Per-document or per-span **polarity labeling**, annotation, or classifier training → `sentiment-analysis-engineer`24- Generic demand, inventory, or logistics forecasting **without** sentiment inputs → `predictive-logistics-developer`, `data-scientist`25- **Investment advice**, trade recommendations, or actionable trading signals → provide forecasting methodology and uncertainty only26- Marketing copy, campaigns, or brand voice → `content-creator`, `brand-voice-enforcement`27- Broad macro econometrics or financial modeling **without** text-derived sentiment → `financial-analyst` (partial overlap only)28- Exploratory NLP or single-shot sentiment scores on a static corpus → `sentiment-analysis-engineer`29- LLM product features, agents, or RAG (unless sentiment forecasting is one pipeline component) → `ai-engineer`3031## Related skills3233| Need | Skill |34|---|---|35| Document-level polarity, ABSA, annotation, classifier eval | `sentiment-analysis-engineer` |36| General ML, experimentation, non-time-series predictive modeling | `data-scientist` |37| Warehouse metrics, dbt, analytics pipelines (if present in repo) | `analytics-engineer` |38| Demand/inventory forecasting without opinion indices | `predictive-logistics-developer` |39| Campaign ROI and channel performance (if present in repo) | `marketing-analyst` |40| Ratios, valuation, macro series without text sentiment (if present) | `financial-analyst` |41| LLM apps, feature stores for agent products | `ai-engineer` |4243## Core Workflows4445### 1. Scope and index design4647Clarify population (brand, product, geo), text sources, aggregation grain, target horizon, and downstream KPIs.4849**See `references/sentiment_forecasting_engineer_scope.md`.**5051### 2. Indices, aggregation, and features5253Define index formulas, rollups, topic/strata splits, and covariates (volume, velocity, mix).5455**See `references/indices_aggregation_and_features.md`.**5657### 3. Time-series and forecast models5859Choose baselines and advanced models; align seasonality, holidays, and exogenous drivers.6061**See `references/time_series_and_forecast_models.md`.**6263### 4. Backtesting, validation, and metrics6465Walk-forward evaluation, interval calibration, and spike-event holdouts.6667**See `references/backtesting_validation_and_metrics.md`.**6869### 5. Data quality, bias, and events7071Bot filtering, sample bias, language drift, and shock labeling for scenario analysis.7273**See `references/references_data_quality_bias_and_events.md`.**7475### 6. Production monitoring and stakeholders7677Serving cadence, drift monitors, dashboard contracts, and stakeholder-ready narratives.7879**See `references/production_monitoring_and_stakeholders.md`.**8081## Outputs8283- **Index specification** — formula, universe, weights, strata, and revision policy84- **Feature catalog** — engineered signals with definitions and lag structure85- **Forecast spec** — horizon, frequency, model family, and exogenous inputs86- **Backtest report** — walk-forward metrics, interval coverage, and failure slices87- **Nowcast playbook** — latency budget, refresh rules, and stale-data handling88- **Monitoring plan** — drift, spike alerts, and human review triggers89- **Stakeholder brief** — trajectory narrative with explicit uncertainty (no trade advice)9091## Principles9293- Forecast **aggregates**, not individual opinions — index stability and definitional clarity come first94- Treat **index construction** as part of the model — changing weights invalidates historical comparability95- Prefer **walk-forward** evaluation over single holdout splits for time-ordered data96- Report **intervals and scenarios**, not point estimates alone; disclose coverage on backtests97- Separate **methodology from decisions** — do not present forecasts as buy/sell or guaranteed outcomes98- Document **known biases** (platform mix, bot share, demographic skew) beside every published index
daemon-blockint-tech