Evaluate a Random Forest Model on Test Data

Description

Evaluate a Random Forest Model on Test Data

Usage

evaluate_rf_model(
  rf_model,
  test_texts,
  test_labels,
  tfidf_model,
  vectorizer,
  stopwords,
  verbose = TRUE
)

Arguments

Value

A list with test accuracy, test predictions, and aligned test data.

Handle Negation in Token List

Description

This function processes a character vector of tokens and handles negations by combining the word "not" with the immediately following word (e.g., "not happy" becomes "not_happy"). This technique helps to better preserve sentiment polarity during text analysis.

Usage

handle_negation(tokens)

Arguments

Details

The negation handling procedure follows these steps:

• Iterate through each token.

• If a token is "not" and followed by another token, merge them into a single token separated by an underscore (e.g., "not_happy").

• Skip the next token after merging to avoid duplication.

• Otherwise, keep the token unchanged.

This method is especially useful in sentiment analysis tasks where the presence of negations can invert the sentiment polarity of words.

Value

A character vector of tokens with negations handled by combining "not" with the next word.

Examples

handle_negation(c("i", "am", "not", "happy"))
# Returns: c("i", "am", "not_happy")

handle_negation(c("this", "is", "not", "good", "but", "not", "terrible"))
# Returns: c("this", "is", "not_good", "but", "not_terrible")

handle_negation(c("nothing", "to", "worry", "about"))
# Returns: c("nothing", "to", "worry", "about")

Predict Emotion with Emoji Representation

Description

This function takes input text, preprocesses it, extracts TF-IDF features using a pre-trained model, predicts the emotion using a trained classifier, and returns the result with optional emoji representation.

Usage

predict_emotion_with_emoji(text, output_type = "textemoji")

Arguments

Value

Depending on output_type:

• For "emotion": character string of predicted emotion

• For "emoji": character string of corresponding emoji

• For "textemoji": original text with appended emoji

The function also prints the result to console.

Examples

## Not run: 
# This example is not run because it requires manually downloading
# a large external model (~30MB), which cannot be retrieved automatically
# and may fail in offline environments.
predict_emotion_with_emoji("I'm so happy today!")
predict_emotion_with_emoji("This makes me angry", "emoji")
predict_emotion_with_emoji("I feel scared", "emotion")

## End(Not run)

Preprocess Text with Slang Handling

Description

This function performs multi-stage text preprocessing, including lowercasing, HTML cleaning, punctuation normalization, contraction expansion, internet slang replacement, emoticon replacement, and final standardization.

Usage

preprocess_text(text, use_textclean = TRUE, custom_slang = NULL)

Arguments

Details

The preprocessing pipeline includes:

• Lowercasing the text.

• Replacing HTML entities and non-ASCII characters.

• Expanding common English contractions (e.g., "I'm" -> "I am").

• Replacing internet slang and emoticons if use_textclean is TRUE.

• Handling additional slang defined by the user.

• Normalizing repeated punctuations and whitespace.

Value

A character vector of cleaned and normalized text.

Examples

preprocess_text("I'm feeling lit rn!!!")
preprocess_text("I can't believe it... lol :)", use_textclean = TRUE)

Train a full model pipeline including text preprocessing, TF-IDF vectorization, random forest tuning, and training.

Description

Train a full model pipeline including text preprocessing, TF-IDF vectorization, random forest tuning, and training.

Arguments

Value

A list containing the trained TF-IDF model, vectorizer, random forest model, and test accuracy.

Train a Random Forest Model with TF-IDF Features

Description

Train a Random Forest Model with TF-IDF Features

Usage

train_rf_model(
  train_matrix,
  train_labels,
  ntree = 300,
  mtry = NULL,
  seed = 123,
  verbose = TRUE,
  train_df_cache_path = "train_df_cached.rds"
)

Arguments

Value

A trained 'ranger' model object.

Train a TF-IDF Model (for Training Phase)

Description

Train a TF-IDF model with customizable tokenization and vocabulary pruning.

Usage

train_tfidf_model(
  preprocessed_text,
  max_features = 10000,
  min_df = 2,
  max_df = 0.8
)

Arguments

Details

This function performs the following steps:

1. Tokenizes the preprocessed text into words and removes stopwords. 2. Defines custom stopwords and retains important emotional function words. 3. Creates a vocabulary based on unigrams and trigrams, pruning terms based on document frequency and term count. 4. Builds the TF-IDF sparse matrix for the input text.

Value

A list with the following components:

tfidf_model

The trained TF-IDF model object.

vectorizer

The vocabulary vectorizer used in training.

tfidf_matrix

The TF-IDF sparse matrix representing the text data.

Examples

preprocessed_text <- c("I'm feeling so happy today!", "I feel really excited and hopeful!")
result <- train_tfidf_model(preprocessed_text)
result$tfidf_model  # Access the trained TF-IDF model

Tune Random Forest Model Hyperparameters

Description

This function performs hyperparameter tuning for a Random Forest model using grid search. It searches over the grid of 'mtry' (number of variables to consider at each split) and 'ntree' (number of trees in the forest) to find the best model based on training accuracy.

Usage

tune_rf_model(
  train_matrix,
  train_labels,
  mtry_grid = c(5, 10, 20),
  ntree_grid = c(100, 200, 300),
  seed = 123,
  verbose = TRUE
)

Arguments

Details

The function trains multiple Random Forest models using different combinations of 'mtry' and 'ntree' values, and evaluates their performance based on training accuracy. The hyperparameters that give the highest accuracy are returned as the best parameters. The process uses the 'ranger' package for training the Random Forest model.

Value

A list containing the best hyperparameters ('mtry', 'ntree', and 'accuracy'):

• 'mtry': The best number of variables to consider at each split.

• 'ntree': The best number of trees in the forest.

• 'accuracy': The accuracy achieved by the model with the best hyperparameters.