to_delete

# --------------------------
#  1. Recommendation system - collaborative filtering
# --------------------------

# Load necessary libraries
library(dplyr)
library(tidyr)
library(recommenderlab)
library(tidyverse)

## Loading out data
df_t <- read_csv("original_df.csv")
unique_tweet_df <-read_csv("unique_tweets_data.csv")

# Ensure label_numeric is binary (1 for YES, 0 for NO)
df_t$label_numeric <- ifelse(df_t$label_task1_1 == "YES", 1, 0)

# Pivot the data from long to wide format: annotator_id vs tweet_id
rating_matrix <- df_t %>%
  select(annotator_id, id_EXIST, label_numeric) %>%
  pivot_wider(names_from = id_EXIST, values_from = label_numeric, values_fill = list(label_numeric = NA))

head(rating_matrix)

# Convert to matrix format 
mat <- as.matrix(rating_matrix[,-1])  # remove annotator_id column
rownames(mat) <- rating_matrix$annotator_id

# Convert to realRatingMatrix
bin_mat <- as(mat, "realRatingMatrix")

# print the bin_mat
as(bin_mat, "matrix")[1:6, 1:5]

# train-test split (80% of users for training)
set.seed(123)
evaluation_scheme <- evaluationScheme(bin_mat, method = "split", train = 0.8, given = 2)

# train a user-based collaborative filtering (UBCF)
recommender_model <- Recommender(getData(evaluation_scheme, "train"), method = "UBCF")

# predict labels for the test set (ratings type)
predictions <- predict(recommender_model, getData(evaluation_scheme, "known"), type = "ratings")

# evaluate model performance
error <- calcPredictionAccuracy(predictions, getData(evaluation_scheme, "unknown"))

# print RMSE, MSE, MAE 
print(error)

# print predictions
as(predictions, "matrix")[1:10, 1:5]

summary(rating_matrix)

# create the matrix of the prediction, in order to create a feature nemd cf_bias with the bias for each anotator
cf_matrix <- as(predictions, "matrix")
cf_matrix[1:10, 1:5]
# crete the column cf_bias with the mean for the bias for each anotator
annotator_bias <- rowMeans(cf_matrix, na.rm = TRUE)
bias_df <- data.frame(annotator_id = names(annotator_bias), cf_bias = annotator_bias)

print(bias_df)

### Evaluating some of the results given in the cf_bias columns for the anotator.

### CASE 1 - cf_bias =  -0.010904255,  Annotator_411

inital_df <- read_csv("original_df.csv")
inital_df <- inital_df[,-1] 

inital_df[inital_df$annotator_id == "Annotator_411", c("id_EXIST", "label_task1_1", "annotator_id")]

inital_df %>%
  filter(annotator_id == "Annotator_411") %>%
  count( annotator_id, label_task1_1)

# --------------------------
# A tibble: 2 × 3
# annotator_id	label_task1_1	n
# <chr>	<chr>	<int>
# Annotator_411	NO	35
# Annotator_411	YES	13
# --------------------------

# In this case the anotator labels more tweets as NO.

### CASE 2 - cf_bias =  1.006666667,  Annotator_725
inital_df[inital_df$annotator_id == "Annotator_725", c("id_EXIST", "label_task1_1",  "annotator_id")]

inital_df %>%
  filter(annotator_id == "Annotator_725") %>%
  count( annotator_id, label_task1_1)

# --------------------------
# A tibble: 2 × 3
# annotator_id	label_task1_1	n
# <chr>	<chr>	<int>
# Annotator_725	NO	12
# Annotator_725	YES	38
# --------------------------

# In this case the anotator labels more tweets as Yes.

### CASE 3 - cf_bias =  0.502752504,  Annotator_683
inital_df[inital_df$annotator_id == "Annotator_683", c("id_EXIST", "label_task1_1",  "annotator_id")]

inital_df %>%
  filter(annotator_id == "Annotator_683") %>%
  count( annotator_id, label_task1_1)

# --------------------------
# A tibble: 2 × 3
# annotator_id	label_task1_1	n
# <chr>	<chr>	<int>
# Annotator_683	NO	26
# Annotator_683	YES	24
# --------------------------

# In this case the anotator labels the same amount of NO and YES.