Metrics Documentation
Understanding AudioAlpha's sentiment analysis
AudioAlpha analyzes crypto podcast discourse to generate actionable sentiment metrics. This guide explains each metric, its scale, and how it's visualized in our charts.
Circle Diagram Metrics
The concentric ring chart displays 6 key metrics. Each ring fills from 0-100% based on the metric's normalized value. The center shows the primary α-sentiment score.
α-sentiment
The primary market sentiment score derived from podcast discourse analysis. Measures overall bullish or bearish sentiment expressed by crypto podcast hosts and guests.
α_sentiment = AVG(episode_sentiment_scores)
where episodes are from that calendar day (UTC)Sentiment Z-Score
How the current sentiment compares to historical norms. A z-score of +2 means sentiment is 2 standard deviations above average - unusually bullish. Negative values indicate below-average sentiment.
z = (α_sentiment_today - μ) / σ
where μ = AVG(α_sentiment) over past 90 days
and σ = STDDEV_SAMP(α_sentiment) over past 90 days
(excludes current day)1-Day Delta (Δ1d)
Change in sentiment over the past calendar day (UTC). Positive values indicate improving sentiment, negative values indicate declining sentiment.
Δ1d = α_sentiment_today - α_sentiment_yesterday7-Day Delta (Δ7d)
Change in sentiment over the past 7 calendar days. Shows the weekly trend direction and magnitude.
Δ7d = α_sentiment_today - α_sentiment_7_days_agoConsensus Score
Measures agreement among podcast hosts. High consensus (80%+) means most voices agree on direction. Low consensus indicates mixed or conflicting opinions.
consensus = exp(-σ / 0.8)
where σ = STDDEV_SAMP(episode_sentiments)
Lower std deviation = higher agreement = higher score
Default: 0.5 when n ≤ 1Episode Volume Score
Sigmoid-normalized z-score of episode count by day of week. Shows whether there's unusually high or low podcast activity compared to typical patterns for that weekday.
score = 1 / (1 + exp(-z))
where z = (episode_count - μ_weekday) / σ_weekday
μ and σ calculated from same weekday over past 180 daysSpider Chart Metrics
The spider (radar) chart shows 5 dimensions for each asset. All metrics are normalized to 0-1 scale. The filled area represents the asset's overall profile - larger area generally indicates stronger signals.
α-sentiment (Spider)
Asset-specific sentiment score on the spider chart. Normalized to 0-1 scale for visualization alongside other metrics.
normalized = raw_sentiment / 10Consensus
How much podcast hosts agree on their assessment of this specific asset. High consensus strengthens the signal reliability.
consensus = EXP(-σ / 0.8)
where σ = STDDEV_SAMP(sentiment_scores)
Default: 0.5 when episode_count ≤ 1Z-score
Normalized sentiment deviation from the mean. Shows how far the current sentiment is from typical levels, scaled to 0-1.
z_score_norm = 0.5 + 0.5 × TANH(sentiment_zscore / 2)
where sentiment_zscore = (sentiment - mean) / stddevNarrative Intensity
The strength and conviction of narratives being expressed about the asset. High intensity suggests strong opinions and active debate.
narrative_intensity = attention_share × (1 + ABS(bull_ratio - bear_ratio)) × √episode_count
Normalized: narrative_intensity_norm = MIN(1.0, intensity / p95_intensity)Daily Market Metrics
These metrics are available in the daily snapshot data (daily_snapshot table) and can be exported via the dashboard. They provide market-wide statistical context and are calculated across all episodes for each day. See also Circle Diagram Metrics above for visualization-specific metrics.
α-Sentiment Median
The median sentiment score across all episodes for the day. Less affected by outliers than the mean.
PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY episode_sentiment)α-Sentiment Std Dev
Standard deviation of episode sentiments. Higher values indicate more disagreement/variance in opinions.
σ = STDDEV_SAMP(episode_sentiments)α-Sentiment Variance
Variance of episode sentiments. The squared standard deviation, measuring spread of sentiment values.
VAR_SAMP(episode_sentiments)α-Sentiment OHLC
Open, High, Low, Close sentiment values for the day - similar to price candlesticks. Open = first episode, Close = last episode.
Open = first episode sentiment (by publish time)
High = MAX(sentiment)
Low = MIN(sentiment)
Close = last episode sentiment (by publish time)Sentiment Skewness
Measures asymmetry in the distribution. Positive skew = tail toward bullish outliers. Negative skew = tail toward bearish outliers.
skew = (n / ((n-1)(n-2))) × Σ((x - μ)³) / σ³
(Fisher-Pearson standardized moment coefficient)Sentiment Kurtosis
Measures 'tailedness' of distribution. High kurtosis = more outliers/extreme opinions. Low kurtosis = opinions clustered around mean.
kurt = ((n(n+1)) / ((n-1)(n-2)(n-3))) × Σ((x - μ)⁴) / σ⁴
- (3(n-1)² / ((n-2)(n-3)))
(excess kurtosis, normal distribution = 0)Episode Count
Total number of podcast episodes analyzed for that day.
COUNT(*) WHERE snapshot_date = target_dateSentiment Strength Index
Measures the conviction/strength of the overall market sentiment signal. Combines sentiment deviation from neutral with sample size confidence.
SSI = (((mean_sentiment - 5) / 5) × MIN(1.0, √n / 3.0)) / 2.0 + 0.5
where n = episode count for the daySentiment Strength Index Raw
Measures the conviction/strength of the overall market sentiment signal. With stronger sensitivity to rarely mentioned assets and centered around 0 for modeling/alpha research.
SSI_raw = ((mean_sentiment - 5.5) / 4.5) × MIN(1.0, √n / 10.0)
where n = episode count for the dayMarket Regime
Concise market regime label.
Classified based on sentiment levels, trends, and volatility patternsAsset-Level Metrics
These metrics are available in the daily asset snapshot data (daily_asset_snapshot table) and can be exported for your favorite assets. Each metric is calculated per-asset per-day, allowing you to track individual asset sentiment over time. See also Spider Chart Metrics above for visualization-specific metrics.
Composite Scores
α-Index measures narrative strength (how robust and reliable the signal is), while α-Pulse measures narrative momentum (how rapidly sentiment is changing).
α-Index
Composite narrative strength score combining sentiment level, historical sentiment deviation (z-score), host consensus, and coverage reach. Requires 2+ episode mentions to produce a meaningful score. Higher values indicate stronger, more reliable narrative signals.
α-Index = (S·0.22 + Z·0.28 + C·0.22 + V·0.28) × (1 + η·0.15)
Where:
S = (sentiment_mean - 1) / 9
Z = 0.5 + 0.5 × tanh(sentiment_zscore / 2)
C = consensus_score × min(episode_count / 5, 1)
V = ½ × (max(min(attention_share, 1), 0.15) + max(min(intensity_norm, 1), 0.15))
η = novelty_scoreα-Pulse
Composite narrative momentum score measuring how rapidly an asset's podcast narrative is accelerating. Combines sentiment z-score, 1-day sentiment delta, attention share, and surprise score. Dampened for assets with low episode counts to reduce noise.
α-Pulse = ((B × (1 + η·0.15) + 0.70) / 1.40) × d + 0.40 × (1 - d)
Where:
B = tanh(Zs/2)·0.40 + tanh(δ1d/1.5)·0.30 + surprise·0.15 + attention·0.15
Zs = sentiment_zscore
δ1d = sentiment_delta_1d
d = min(episode_count / 3, 1) (episode dampener)
η = novelty_scoreEpisode Count (Asset)
Number of episodes mentioning this specific asset on the given day.
COUNT(*) WHERE ticker = asset AND snapshot_date = target_dateDays Since Last Mention
Number of days since this asset was last mentioned in a podcast.
snapshot_date - last_mention_dateα-Sentiment Mean (Asset)
Average sentiment score for this specific asset across all mentions.
AVG(sentiment_score) WHERE ticker = assetα-Sentiment Median (Asset)
Median sentiment score for this asset. Less affected by outliers.
PERCENTILE_CONT(0.5) WHERE ticker = assetα-Sentiment Std Dev (Asset)
Standard deviation of sentiment for this asset.
STDDEV_SAMP(sentiment_score) WHERE ticker = assetα-Sentiment Variance (Asset)
Variance of sentiment scores for this asset.
VAR_SAMP(sentiment_score) WHERE ticker = assetα-Sentiment OHLC (Asset)
Open, High, Low, Close sentiment for this asset on the given day.
Open = first mention sentiment
High = MAX
Low = MIN
Close = last mention sentiment1-Day Delta (Asset)
Change in this asset's sentiment compared to yesterday.
sentiment_today - sentiment_yesterday7-Day Delta (Asset)
Change in this asset's sentiment compared to 7 days ago.
sentiment_today - sentiment_7d_agoBull Ratio
Proportion of mentions with bullish sentiment for this asset.
COUNT(bullish_mentions) / COUNT(total_mentions)Bear Ratio
Proportion of mentions with bearish sentiment for this asset.
COUNT(bearish_mentions) / COUNT(total_mentions)Neutral Ratio
Proportion of mentions with neutral sentiment for this asset.
COUNT(neutral_mentions) / COUNT(total_mentions)Consensus Score (Asset)
How much hosts agree on their assessment of this specific asset. Higher values indicate stronger agreement.
consensus = EXP(-σ / 0.8)
where σ = STDDEV_SAMP(sentiment_scores) for this asset
Default: 0.5 when episode_count ≤ 1Surprise Score
How unexpected or surprising the current attention is compared to historical patterns. Combines recency of mentions with attention deviation.
surprise = MIN(
0.5 × (1.0 - days_mentioned / 7.0) +
0.5 × ABS(attention_share - avg_attention) / NULLIF(avg_attention, 0),
1.0
)
where avg_attention = historical average attention shareα-Sentiment Momentum
Combines sentiment change direction with conviction (consensus) and discussion volume. Positive values indicate bullish momentum, negative values bearish.
momentum = sentiment_delta_1d × consensus_score × LN(episode_count + 1)
Returns 0 when any component is NULLNarrative Intensity
The strength and conviction of narratives being expressed about the asset. Combines attention share, polarization (bull vs bear divergence), and discussion volume.
narrative_intensity = attention_share × (1 + ABS(bull_ratio - bear_ratio)) × √episode_countNarrative Intensity (Normalized)
Normalized version of narrative intensity for cross-asset comparison. Uses the 95th percentile of all assets on that day as the reference point.
narrative_intensity_norm = MIN(1.0, intensity / p95_intensity)
where p95_intensity = PERCENTILE_CONT(0.95) WITHIN GROUP
(ORDER BY narrative_intensity)
for all assets on the snapshot dateColor Coding
Sentiment Color Scale
Spider Chart Scoring
Have questions about our metrics? Check our FAQ.