In canfielder/CausalityExtraction: Hypothesis Extraction and Analysis for Scholarly Papers in Social Sciences

knitr::opts_chunk$set(echo = TRUE)

Purpose

The purpose of this notebook is to test how to implement the Entity Extraction model

Import

Libraries

  if (!require(pacman)) {install.packages('pacman')}
  p_load(
    dplyr,
    readxl,
    reticulate,
    stringr,
    tokenizers,
    tidyr
  )

Source Files

The following imports functions defined in the sourced R scripts.

# Import All Scripts
script_path <- "../R/"
file_paths <- list.files(recursive = TRUE, 
                         path = script_path, pattern = ".R", 
                         full.names = TRUE)

for (file in file_paths){
  source(file)
}

file_paths

fastText Model

## Load Model
path_ft_model <- "./../models/fasttext_model.bin"
ft_model <- fastTextR::ft_load(path_ft_model)

Data

# PDF Input
folder_path <- "./../data/sample_papers/"
pdf_paths <- list.files(recursive = FALSE, 
                       path = folder_path, 
                       pattern = ".pdf", 
                       full.names = TRUE)
pdf_path <-pdf_paths[3]
print(pdf_path)

# Error Word Splits
df_word_split_error <- read.csv(file ="./../data/processing_next_line_split_error.csv", 
                                stringsAsFactors = FALSE)
word_split_error <- df_word_split_error %>% select(word) %>% pull

Process Data

Preprocess Text Steps

# Process Text
text_processed <- process_text(pdf_path)

input_text <- text_processed

Processing - Hypothesis Exrtaction

  # Concatenate All Vector Elements, Separated By Line Split
  processing_text <- str_c(input_text, collapse = " ")
  processing_text <- tokenize_sentences(processing_text,
                                        strip_punct = FALSE) %>% unlist()

  # Replace Double Spaces
  processing_text <- str_replace_all(string = processing_text,
                            pattern = "  ",
                            replacement = " ")

  # Normalize Text ------------------------------------------------------------
  processing_text <- tolower(processing_text)

# Regex - Between Hypo and Colon
regex_hypo_marker <- "<split>hypo (.*?):"

# Hypothesis Extraction -----------------------------------------------------
  # Identify Lines with Hypothesis Pattern
  h_match <- processing_text %>% str_match(regex_hypo_marker)
  # Extract Hypotheses Number
  h_match_num <- h_match[,2]

  # Identify Unique Hypothesis Numbers
  h_match_num_unq <- unique(h_match_num)

  # Drop NA
  h_match_num_unq <- h_match_num_unq[!is.na(h_match_num_unq)]

  # Determine Vector Index of Initial Hypothesis Statements
  h_initial <- c()
  for (i in h_match_num_unq){
    intial_idx <- tapply(seq_along(h_match_num),
                         h_match_num,
                         min)[i]
    h_initial <- c(h_initial, intial_idx)
  }

  # Reduce Text to Only Initial Hypothesis Instances
  h_statements <- processing_text[h_initial]
  h_statements

  # Split Statements On Indicator (Defined in Processing) ---------------------
  ## Define
  split_indicator <- "<split>"

  ## Split on Indicator
  h_statements <- str_split(string = h_statements,
                                     pattern = split_indicator) %>%
    unlist()

  ## Detect Statements Which Contain "Hypo"
  logical_hypothesis_2 <- str_detect(h_statements, "hypo")

  ## Drop Statements that Do Not Include "Hypo"
  h_statements <- h_statements[logical_hypothesis_2]

  h_statements

h_statements

detect_string <- ".*: "

h_statements %>% 
  str_extract("hypo (.*?):") %>% 
  str_remove_all("hypo ") %>% 
  str_remove_all(":") %>% 
  as.integer()



h_number <- h_statements %>% 
  str_extract(detect_string) %>% 
  str_remove_all("hypo ") %>% 
  str_remove_all(": ") %>% 
  as.integer()

h_number

output <- vector(mode = "logical", length = length(h_number))
tracker <- vector(mode = "integer", length = length(h_number))

for (i in seq_along(h_number)) {
  num <- h_number[i]

  if (is.na(num)){

    output[i] = FALSE
    tracker[i] <- -1

  } else if (num %in% tracker) {

    output[i] = FALSE
    tracker[i] <- -1

  } else {

    output[i] = TRUE
    tracker[i] <- num

  }
}

h_statements <- h_statements[output]

# 
#   for (i in seq_along(word_split_error)) {
#     word_split <- word_split_error[i]
#     word_fix <- str_replace_all(string = word_split, 
#                                   pattern = " ",
#                                   replacement = "")
#     
#     h_statements <- h_statements %>% 
#       str_replace_all(pattern = word_split, replacement = word_fix)
#   }
# 
# h_statements

  # Maintain Data 
  hypothesis_causality <- h_statements

  # Drop ~Hypo #:~
  hypothesis_entity <- gsub(".*: ","",h_statements)
  hypothesis_entity

hypothesis_pred <- fastTextR::ft_predict(
    model   = ft_model,
    newdata = hypothesis_entity,
    rval    = "dense"
  ) %>%
    as.data.frame()

  ## Assign prediction column names
  col_names <- names(hypothesis_pred)

  col_names[1]

  ## Drop statements which were predicted as non-hypothesis class
  if (!("__label__0" %in% col_names)) {
    response <- vector(
      mode   = "logical",
      length = length(hypothesis_entity)
    )

    for (i in seq_along(hypothesis_entity)){
      response[i] <- TRUE

    }

  } else if (!("__label__1" %in% col_names)) {
    response <- vector(
      mode   = "logical",
      length = length(hypothesis_entity))

    for (i in seq_along(hypothesis_entity)){
      response[i] <- FALSE

    }
  } else {
    response <- hypothesis_pred %>%
      dplyr::mutate(
        Response = dplyr::if_else(
          condition = col_names[1] <= col_names[2],
          true      =  TRUE,
          false     = FALSE
        )
      ) %>%
      dplyr::pull(Response)

  }
hypothesis_entity <- hypothesis_entity[response]
hypothesis_entity

hypothesis_pred 

fasttext_tag = FALSE
# fasttext_tag = TRUE

if (fasttext_tag) {
  output_hypothesis <- apply_fasttext(hypothesis_entity, hypothesis_causality)

  hypothesis_causality = output_hypothesis[[1]]
  hypothesis_entity = output_hypothesis[[2]]
}

hypothesis_entity
hypothesis_causality

# Convert Text
hypothesis_entity <- str_to_sentence(hypothesis_entity, locale = "en")
hypothesis_entity

  # Create Dataframe with Hypothesis Number and Hypothesis
df_hypothesis <- data.frame(hypothesis_entity,
                            hypothesis_causality, 
                            stringsAsFactors = FALSE)

df_hypothesis

canfielder/CausalityExtraction documentation built on Jan. 5, 2022, 10:55 a.m.