TidyTuesday: Attention Analysis - SOFI vs IONQ

Static panel visualization of Reddit attention metrics (z-score, Share of Attention) with price overlay and bottom candidate detection

Overview

Building on last week’s raw comment count analysis, this week we implement the three-metric attention framework:

1. Raw Count – daily Reddit comment volume
2. Share of Attention – ticker’s fraction of total discussion
3. Abnormal Attention (z-score) – deviation from the 20-day rolling mean

When z-score drops below -1.5 while the stock has declined over 20 days, we flag a bottom candidate – a period where “attention has dried up after a selloff.”

• Data Source: Reddit public API + yfinance (via prepare_data.py)
• Period: 60 days
• Visualization: ggplot2 + patchwork (static panels)

Data

library(tidyverse)
library(scales)
library(glue)
library(patchwork)

data_dir <- file.path(getwd(), "data")

metrics <- read_csv(file.path(data_dir, "attention_metrics.csv"),
                    show_col_types = FALSE) |>
  mutate(date = as.Date(date))

price <- read_csv(file.path(data_dir, "price_data.csv"),
                  show_col_types = FALSE) |>
  mutate(date = as.Date(date))

bc <- metrics |> filter(bottom_candidate == TRUE)

cat(glue("Date range: {min(metrics$date)} ~ {max(metrics$date)}\n",
         "Bottom candidates: {nrow(bc)}"))

Date range: 2025-12-24 ~ 2026-02-21
Bottom candidates: 0

Visualizations

colors <- c("SOFI" = "#6366f1", "IONQ" = "#f59e0b")

theme_attention <- theme_minimal(base_size = 12) +
  theme(
    legend.position = "top",
    plot.title = element_text(face = "bold", size = 13),
    plot.subtitle = element_text(color = "gray50", size = 10),
    panel.grid.minor = element_blank(),
    strip.text = element_text(face = "bold", size = 12)
  )

1. Price + Bottom Candidates

# Use only rows with price data
price_metrics <- metrics |> filter(!is.na(close))

p_price <- ggplot(price_metrics, aes(x = date, y = close, color = symbol)) +
  geom_line(linewidth = 0.9) +
  geom_point(
    data = price_metrics |> filter(bottom_candidate == TRUE),
    aes(x = date, y = close),
    shape = 24, size = 3, fill = "#ef4444", color = "#ef4444"
  ) +
  scale_color_manual(values = colors) +
  scale_x_date(date_labels = "%m/%d", date_breaks = "1 week") +
  scale_y_continuous(labels = dollar) +
  facet_wrap(~symbol, scales = "free_y") +
  labs(
    title = "Close Price with Bottom Candidate Signals",
    subtitle = "Red triangles = z < -1.5 AND 20-day return < 0",
    x = NULL, y = "Price ($)", color = NULL
  ) +
  theme_attention

p_price

2. z-score (Abnormal Attention)

p_zscore <- ggplot(metrics, aes(x = date, y = z_score, color = symbol)) +
  geom_line(linewidth = 0.7) +
  geom_point(size = 1, alpha = 0.6) +
  geom_hline(yintercept = -1.5, linetype = "dashed", color = "#ef4444", linewidth = 0.5) +
  geom_hline(yintercept = 0, linetype = "dotted", color = "gray60") +
  annotate("text", x = min(metrics$date) + 2, y = -1.5, label = "z = -1.5",
           vjust = -0.5, color = "#ef4444", size = 3) +
  scale_color_manual(values = colors) +
  scale_x_date(date_labels = "%m/%d", date_breaks = "1 week") +
  facet_wrap(~symbol, scales = "free_y") +
  labs(
    title = "Abnormal Attention (z-score)",
    subtitle = "Rolling 20-day z-score | Below -1.5 = attention dried up",
    x = NULL, y = "z-score", color = NULL
  ) +
  theme_attention

p_zscore

3. Share of Attention

p_share <- ggplot(metrics, aes(x = date, y = share, fill = symbol)) +
  geom_area(alpha = 0.6, position = "identity") +
  scale_fill_manual(values = colors) +
  scale_x_date(date_labels = "%m/%d", date_breaks = "1 week") +
  scale_y_continuous(labels = percent) +
  labs(
    title = "Share of Attention",
    subtitle = "Each ticker's fraction of total Reddit discussion",
    x = NULL, y = "Share", fill = NULL
  ) +
  theme_attention

p_share

4. Raw Comment Count

p_raw <- ggplot(metrics, aes(x = date, y = raw_count, fill = symbol)) +
  geom_col(position = position_dodge(width = 0.8), width = 0.7, alpha = 0.7) +
  scale_fill_manual(values = colors) +
  scale_x_date(date_labels = "%m/%d", date_breaks = "1 week") +
  scale_y_continuous(labels = comma) +
  labs(
    title = "Raw Comment Count",
    subtitle = "Daily Reddit mentions across investment subreddits",
    x = NULL, y = "Comments", fill = NULL
  ) +
  theme_attention

p_raw

5. Combined Panel

(p_price / p_zscore / p_share / p_raw) +
  plot_annotation(
    title = "Reddit Attention Analysis: SOFI vs IONQ (60 days)",
    subtitle = "Three-metric framework: Raw Count | Share of Attention | z-score + Bottom Candidates",
    theme = theme(
      plot.title = element_text(size = 17, face = "bold"),
      plot.subtitle = element_text(size = 12, color = "gray50")
    )
  )

Key Findings

1. z-score normalization reveals attention “droughts” that raw counts miss – a day with zero comments might be normal for IONQ but abnormal for SOFI.
2. Share of Attention tracks which ticker is capturing the conversation – useful for identifying relative momentum shifts.
3. Bottom candidates (z < -1.5 with negative 20-day returns) are rare signals. When they cluster, they suggest “capitulation” – the stock has fallen and nobody is talking about it anymore.
4. The combined panel format (price / z-score / share / raw count) provides a complete attention dashboard for each ticker.

---

This post is part of the TidyTuesday weekly data visualization project.

CautionDisclaimer

This analysis is for educational and practice purposes only. Reddit comment counts and attention metrics are based on publicly available data and may not represent complete or current information. Bottom candidate signals are experimental and do not constitute investment advice.