更新qdap情感分析词典：组合词极性修正需求

Great question—this is a common gotcha with rule-based sentiment analysis tools like qdap::polarity, where multi-word phrases get split into individual terms and their sentiments cancel out. Here's how you can add custom multi-word negative phrases to the dictionary to get accurate results:

Step 1: Access and Modify the Default Polarity Dictionary #

First, we'll make a copy of qdap's built-in polarity dictionary, then add your custom phrases as single entries marked as negative.

# Load required packages
library(qdap)
library(qdapDictionaries)

# Create a copy of the default polarity dictionary
custom_polarity <- polarity_df

# Add your multi-word negative phrase (use lowercase to match qdap's default text normalization)
custom_polarity <- rbind(
  custom_polarity,
  data.frame(
    term = "pretty bad",
    polarity = -1,
    stringsAsFactors = FALSE
  )
)

# Add more phrases if needed—just repeat the data.frame row for each one
# Example:
# custom_polarity <- rbind(
#   custom_polarity,
#   data.frame(term = "terribly wrong", polarity = -1, stringsAsFactors = FALSE),
#   data.frame(term = "super disappointing", polarity = -1, stringsAsFactors = FALSE)
# )

Step 2: Run Polarity Analysis with the Custom Dictionary #

When you call the polarity function, pass your modified dictionary using the custom.dict parameter. This tells qdap to prioritize your custom multi-word phrases over individual term matches.

# Test with your target phrase
test_text <- "That restaurant experience was pretty bad"

# Run polarity analysis with the custom dictionary
sentiment_result <- polarity(test_text, custom.dict = custom_polarity)

# View the results—you'll see the correct negative polarity score
print(sentiment_result)

Key Notes #

• Case Consistency: qdap automatically converts text to lowercase during analysis, so make sure your custom phrases are in lowercase too (e.g., "pretty bad" not "Pretty Bad").
• Phrase Priority: The polarity function will match multi-word phrases first before splitting into individual terms, so your custom entries will override conflicting single-word sentiment scores.
• Batch Additions: For multiple phrases, just add more rows to the data.frame when binding to the default dictionary.

This approach ensures phrases like "Pretty Bad" are treated as a single negative unit, so their sentiment doesn't get canceled out by conflicting individual terms.

内容的提问来源于stack exchange，提问作者Rana Usman