如何在R语言中不使用Google API获取情感分数与量级（极性）

Hey there! Sorry to hear you hit that frustrating billing glitch with the Google API—let’s walk through some robust, local R alternatives that deliver comparable sentiment scores and magnitude (polarity intensity) for your multi-column consumer survey data.

1. Using sentimentr (Lightweight, Context-Aware) #

The sentimentr package is built specifically to calculate both sentiment scores and their associated intensity (which maps directly to your "magnitude" need). It handles contextual nuances better than basic dictionary tools, making it perfect for open-ended survey responses.

Step-by-Step Implementation: #

• First, install and load required packages:

  install.packages("sentimentr")
  library(sentimentr)
  library(dplyr)
  library(purrr)
  

• Create a reusable function to process each text column, then apply it across all your survey response columns. Example assumes your data frame is named survey_data and text columns start with response_:

  # Function to extract sentiment score and magnitude for a single column
  get_sentiment_metrics <- function(text_col) {
    # Calculate sentence-level sentiment
    sentiment_results <- sentiment(text_col)
    # Aggregate to row-level metrics
    sentiment_results %>%
      group_by(element_id) %>%
      summarize(
        sentiment_score = mean(sentiment, na.rm = TRUE),
        magnitude = mean(abs(sentiment), na.rm = TRUE) # Magnitude = average absolute intensity
      ) %>%
      select(sentiment_score, magnitude)
  }
  
  # Identify all text columns and process them
  text_columns <- grep("^response_", names(survey_data), value = TRUE)
  sentiment_output <- map_dfc(text_columns, function(col) {
    results <- get_sentiment_metrics(survey_data[[col]])
    colnames(results) <- paste0(col, "_", c("sentiment", "magnitude"))
    results
  })
  
  # Combine metrics with original data
  final_data <- bind_cols(survey_data, sentiment_output)
  

2. Using transformers (State-of-the-Art BERT Models, Near-Google Precision) #

If you want accuracy on par with commercial APIs, the transformers package lets you load pre-trained BERT models fine-tuned for sentiment analysis. These models capture subtle language nuances better than rule-based methods, ideal for complex survey feedback.

Step-by-Step Implementation: #

• Install and load dependencies:

  install.packages("transformers")
  library(transformers)
  library(dplyr)
  

• Load a fast, accurate pre-trained model (DistilBERT is great for balance of speed and performance):

  tokenizer <- tokenizer$from_pretrained("distilbert-base-uncased-finetuned-sst-2-english")
  model <- model$from_pretrained("distilbert-base-uncased-finetuned-sst-2-english")
  

• Create a function to get sentiment score and magnitude (confidence):

  bert_sentiment <- function(text) {
    # Handle missing values
    if (is.na(text)) {
      return(tibble(sentiment_score = NA, magnitude = NA))
    }
    # Tokenize input text
    inputs <- tokenizer(
      text,
      return_tensors = "pt",
      padding = TRUE,
      truncation = TRUE
    )
    # Get model predictions
    outputs <- model(inputs$input_ids, attention_mask = inputs$attention_mask)
    # Convert logits to probabilities
    probs <- torch::softmax(outputs$logits, dim = 2)$detach()$numpy()
    # Calculate metrics: score = positive prob - negative prob; magnitude = highest confidence
    tibble(
      sentiment_score = probs[1,2] - probs[1,1],
      magnitude = max(probs[1,1], probs[1,2])
    )
  }
  
  # Apply to a single column (repeat for others or use purrr for bulk processing)
  survey_data <- survey_data %>%
    mutate(
      response_1_sentiment = map_dbl(response_1, ~bert_sentiment(.)$sentiment_score),
      response_1_magnitude = map_dbl(response_1, ~bert_sentiment(.)$magnitude)
    )
  

3. Using tidytext with Enhanced Dictionaries (Customizable, Simple Use Cases) #

If you prefer a dictionary-based approach, tidytext pairs well with the afinn dictionary (which provides numeric sentiment scores). You can calculate magnitude as the total absolute intensity of sentiment-laden words.

Quick Example: #

install.packages("tidytext")
library(tidytext)
library(dplyr)

# Process a single response column
survey_data %>%
  mutate(row_id = row_number()) %>%
  unnest_tokens(word, response_1) %>%
  inner_join(get_sentiments("afinn")) %>%
  group_by(row_id) %>%
  summarize(
    sentiment_score = sum(value),
    magnitude = sum(abs(value)) # Total intensity of sentiment words
  )

All these methods work locally (no external API calls) and offer strong accuracy: transformers will get you closest to Google API performance, while sentimentr balances speed and precision for most survey scenarios.

内容的提问来源于stack exchange，提问作者Madhuri Patil