如何基于3个情感因子水平构建0-100区间的负面累积指数

Let's walk through this step by step—this approach will give you a clear, meaningful index that aligns exactly with your requirements:

Step 1: Clean Sentiment Data & Map to Scores #

First, we need to handle that empty factor level in your data and map each valid sentiment to its corresponding score (3 for negative, 2 for neutral, 1 for positive). We'll turn empty strings into NA since they aren't part of your defined sentiment levels.

library(dplyr)

# Load your data (assuming it's already in a data frame called `data`)
# Map sentiment levels to scores, handling empty strings as NA
score_mapping <- c("negative" = 3, "neutral" = 2, "positive" = 1)
data$sentiment_score <- score_mapping[data$sentiment]

# Optional: Filter out rows with NA sentiment if you don't want to include them
# data <- data %>% filter(!is.na(sentiment_score))

Step 2: Calculate Daily Cumulative Scores #

Next, we'll group your data by day and compute two key values for each day:

• The total sum of sentiment scores (your "cumulative result" for the day)
• The number of valid sentiment observations (to use for normalization)

daily_summary <- data %>%
  group_by(date) %>% # Replace `date` with your actual day column name
  summarise(
    total_daily_score = sum(sentiment_score, na.rm = TRUE),
    valid_obs = sum(!is.na(sentiment_score)) # Count of non-NA sentiment entries
  )

Step 3: Normalize to 0-100 Range #

This is the critical part—we'll normalize each day's score relative to its theoretical minimum and maximum possible values. This ensures 100 means all observations that day are negative, 0 means all are positive, and neutral sits at 50. We'll also handle days with no valid observations to avoid division by zero.

daily_summary <- daily_summary %>%
  mutate(
    # Theoretical min (all positive) and max (all negative) scores for the day
    min_possible = valid_obs * 1,
    max_possible = valid_obs * 3,
    # Calculate normalized index
    normalized_negative_index = case_when(
      valid_obs == 0 ~ NA_real_, # Mark days with no data as NA
      TRUE ~ ((total_daily_score - min_possible) / (max_possible - min_possible)) * 100
    )
  )

Why This Works: #

• Per-day normalization: Unlike global normalization, this accounts for varying numbers of observations per day. A day with 5 all-negative entries will hit 100, just like a day with 100 all-negative entries—fairly comparing sentiment intensity across days.
• Clear interpretation: The index directly reflects how far the day's sentiment leans toward negative (100) or positive (0). Neutral days will land at 50, which makes intuitive sense.

Example Output #

If you run head(daily_summary), you'll see something like this:

In this example:

• Jan 1 has all negative sentiment (100)
• Jan 2 has 3 positive + 2 neutral (20, low negative)
• Jan3 has all neutral (50, midpoint)

Optional: Global Normalization (If Needed) #

If you prefer to normalize across all days (instead of per day), use this code instead of Step3. Note this is less ideal for comparing intensity, but it's an option:

# Get global min and max scores across all days
global_min <- min(daily_summary$total_daily_score, na.rm = TRUE)
global_max <- max(daily_summary$total_daily_score, na.rm = TRUE)

daily_summary <- daily_summary %>%
  mutate(
    global_normalized_index = ((total_daily_score - global_min) / (global_max - global_min)) * 100
  )

内容的提问来源于stack exchange，提问作者Kash