Tweet Emotion Recognition

This repository contains the files to build a model to evaluate the emotions from tweets.

Setup

• Ensure you have the right versions of the dependencies. Run:

  pip install -r requirements.txt

• Create a .env file using the .env.sample:

  cp .env.sample .env

• Set the variables at the .env file

Data Fetch

This module is responsible for the tweets gathering process.

get_tweets

python -m data_fetch.get_tweets [QUERY] -s <SAVE_DIR> -m <MAX_REQUESTS> -c <COUNT> -n

Call the Twitter's API and search for tweets w/ the given QUERY.

Arguments

• query: String. Search parameter.
• -s | --save_dir: String. Path to the directory where the data will be saved.
• -m | --max_requests: Int. Max number of requests to the API. Default: 10.
• -c | --count: Int. Max number of tweets for each request. Default: 100.
• -n | -newer: Boolean. Collect the newer tweets. Default: True.

Returns

A DataFrame w/ the collected data when a SAVE_DIR is provided. Otherwise, None.

concat_datasets

python -m data_fetch.concat_datasets [QUERY] -s <SAVE_DIR> -m <MAX_REQUESTS> -c <COUNT> -n

Create a new file w/ the datasets containing the given QUERY.

Arguments

• query: String. Search parameter.
• dataset_dir: String. Path to the directory where the files are saved.

kmeans_data_filter

python -m data_fetch.kmeans_data_filter [FILE] -o <OUTPUT_NAME> -od <OUTPUT_DIR_NAME>
                                        -mf <MAX_FEATURES> -n <N_CLUSTERS>
                                        -ni <N_INIT> -mi <MAX_ITER>
                                        -r <RANDOM_STATE>

Filter the data from a file, based on the result of a clustering process using the K-means algorithm.

Arguments

• file: String. Path to the file that will be processed.
• -o | --output_name: String. Name of the output file, including the extension.
• -od | --output_dir_name: String. Name of the directory where the output file will be placed. Default: datasets/kmeans.
• -mf | --max_features: Int. Max number of features to use for the TF-IDF scoring. Default: 8000.
• -n | --n_clusters: Int. Number of clusters to create. Default: 4.
• -ni | --n_init: Int. Number of time the k-means algorithm will be run with different centroid seeds. Default: 10.
• -mi | --max_iter: Int. Maximum number of iterations of the k-means algorithm for a single run. Default: 300.
• -r | --random_state: Int. Determines random number generation for centroid initialization. Default: 20.

Sentiment Analysis

This module is responsible for filtering the tweets, based the emotion associated w/ the query and the predicted sentiment.

train

python -m sentiment_analysis.train [MODEL_TYPE] [DATASET_PATH] [TOKENIZER_PATH]
                                   [SAVE_DIR] -l <LABEL_COL> -t <TEXT_COL>
                                   -v <VALIDATION_SPLIT> -ed <EMBEDDING_DIM>
                                   -lr <LEARNING_RATE> -e <EPOCHS> -b <BATCH_SIZE>

Train a sentiment analysis model, using the given MODEL_TYPE and save the weights.

Arguments

• model_type: String. Model to use. Choices: [lstm, lstm_conv]
• dataset_path: String. Path to the dataset.
• tokenizer_path: String. Path to the tokenizer.
• save_dir: Stirng. Path to the directory where the weights will be saved.
• -l | --label_col: String. Name of the label column. Default: label.
• -t | --text_col: String. Name of the text column. Default: text.
• -v | --validation_split: Float. Fraction of the dataset to use as validation data. Default: 0.3.
• -ed | --embedding_dim: Int. Output dimension of the embedding layer. Default: 100.
• -lr | --learning_rate: Float. Initial learning rate. Default: 1e-3.
• -e | --epochs: Int. Total number of epochs. Default: 10.
• -b | --batch_size: Int. Number of batches per epoch. Default: 32.

train_nb

python -m sentiment_analysis.train_nb [DATA_FILE] [MODEL_FILE] [SAVE_DIR] -t <TEXT_COL> -p

Train a naive bayes model and save it.

Arguments

• filename: String. Path to the dataset.
• -l | label_col: String: Name of the label column. Default: label.
• -t | text_col: String: Name of the text column. Default: text.
• -t | validation_split: Float. Fraction of the dataset to use as validation data. Default: 0.3.

predict_from_directory

python -m sentiment_analysis.predict_from_directory [FILES_DIR] [MODEL_FILE] [SAVE_PATH] -t <TEXT_COL>

Create an annotated dataset, based on the search query and the Naive Bayes' sentiment analysis.

Arguments

• files_dir: String. Path to the directory containing the datasets.
• model_file: String. Path to the naive bayes .pickle file.
• save_path: String. Path where the resulting dataset will be saved.
• -t | --text_col: String. Name of the text column. Default: text.

predict_nb

python -m sentiment_analysis.predict_nb [DATA_FILE] [MODEL_FILE] [SAVE_DIR] -t <TEXT_COL> -p

Predict the tweet polarity and save the purge the oddly fetched examples.

Arguments

• data_file: String. Path to the data file.
• model_file: String. Path to the model file.
• save_dir: String. Path to the directory where the result will be saved.
• -t | --text_col: String. Name of the text column. Default: text.
• -p | --positive: Boolean. If the emotion can be classified as positive. Default: False.