cuisine_algorithm.R

# Generating Adjectives used in describing food recipes - American and Mexican

library(tidyverse) # general utility & workflow functions
library(tidytext) # tidy implimentation of NLP methods
library(topicmodels) # for LDA topic modelling 
library(tm) # general text mining functions, making document term matrixes
library(SnowballC) # for stemming
library(stringr)
library(sqldf)
library(udpipe)
udmodel = udpipe_download_model(language = "english")   #load language for POS model

library("rjson")
library(jsonlite)



#load datset-------------------------------------------------------------------------------------
json_file <- "American.json" #Mexican.json for Mexican
cuisineJSON <- fromJSON(paste(readLines(json_file), collapse=""), flatten=TRUE)



#remove duplicates
df_cuisine = cuisineJSON %>% distinct(name, .keep_all = T)

#reduce variables/columns to name, ingredients & course
df_cuisine = df_cuisine %>% select(name, ingredients, course)



#---LOAD DICTIONARY of COMMON CULINARY TERMS AND UNITS
dic_unit = read.csv("dict/cooking_units.txt",encoding = "UTF-8")
names(dic_unit)="unit"

dic_terms = read.csv("dict/cooking_terms.txt",encoding = "UTF-8")
names(dic_terms)="term"

#REMOVE MEASURES
stop_measures = c("1/2","1/3","1/4","1/8","1/16","2/3","2/4","3/4","3/8","about","half","one","quarter", "and", "or", "of", "if", "to", "with", "each", "nbsp")


# CREATE FILTER FUNCTIONS ---------------------------------------
stopw_filter <- function(string, stopwords=c()){
  # Create something like:  "\\b( the|Jura)\\b"
  new_regex <- paste0("\\b( ", paste0(stopwords, collapse="|"), ")\\b")
  x = gsub(new_regex, " ", string)
}

all_but_hyphen_filter_comma <- function(string){
  x = gsub("[^-a-zA-Z[:space:]]+"," ",string, perl = TRUE)
}
no_numbers_filter <- function(string){
  x = gsub("[0-9]"," ",string)
}
no_newline_filter <- function(string){
  x = gsub("\n"," ",string)
}

units_filter <- function(string, dic_unit=c()){
  
  units_regex <- paste0("\\b( ", paste0(dic_unit$unit, collapse="|"), ")\\b")
  y = gsub(units_regex, " ", string)
}

common_filter <- function(string, dic_terms=c()){
  
  common_regex <- paste0("\\b( ", paste0(dic_terms$term, collapse="|"), ")\\b")
  gsub(common_regex, " ", string)  
}


#TOPIC MODELLING FUNCTION-----------------------
top_terms_by_topic_LDA <- function(input_text,  plot = T,  number_of_topics = 5, plottitle)
{
  Corpus <- Corpus(VectorSource(input_text)) 
  DTM <- DocumentTermMatrix(Corpus)
  
  unique_indexes <- unique(DTM$i)
  DTM <- DTM[unique_indexes,]
  
  lda <- LDA(DTM, k = number_of_topics, control = list(seed = 1234))
  topics <- tidy(lda, matrix = "beta")
  top_terms <- topics  %>%
    group_by(topic) %>% 
    top_n(5, beta) %>%
    ungroup() %>%
    arrange(topic, -beta)
  
  
  if(plot == T){
    # plot the top ten terms for each topic in order
    top_terms %>% # take the top terms
      mutate(term = reorder(term, beta)) %>% # sort terms by beta value 
      ggplot(aes(term, beta, fill = factor(topic))) + # plot beta by theme
      ggtitle(plottitle)+
      geom_col(show.legend = FALSE) + # as a bar plot
      facet_wrap(~ topic, scales = "free") + # which each topic in a seperate plot
      labs(x = NULL, y = "Beta") + # no x label, change y label 
      coord_flip() # turn bars sideways
  }else{ 
    # if the user does not request a plot
    # return a list of sorted terms instead
    return(top_terms)
  }
  
}


#SELECT THE COURSE TO ANALYSE -------------------------------------------------
#list all courses to know which Course to input into the     
df_cuisine %>% 
  select(course) %>%
  distinct()




#extract the course
df_course_cuisine = df_cuisine %>%
  filter(str_detect(course, "Salad") ) # replace with any course (eg. "Breakfast and Brunch") from previous query 
  


#LOAD DATASET AS CORPUS-----------------------------
df_DTM <- function(course_dataframe) {
  recipesCorpus <- Corpus(VectorSource(df_course_cuisine$ingredients))   
  recipesDTM <- DocumentTermMatrix(recipesCorpus)           #CONVERT CORPUS TO DOC-TERM-MATRIX
  recipesDTM_tidy <- tidy(recipesDTM)                       #TIDY DTM
  recipesDTM_tidy_cleaned <- recipesDTM_tidy 
  
  #FILTER FRACTIONS --- DICTIONARY WORDS (TERMS AND UNIS)
  recipesDTM_tidy_cleaned=recipesDTM_tidy[!grepl(paste(stop_measures, collapse="|"), recipesDTM_tidy$term),]
  recipesDTM_tidy_cleaned=recipesDTM_tidy_cleaned[!grepl(paste(dic_unit$unit, collapse="|"),  recipesDTM_tidy_cleaned$term),]
  recipesDTM_tidy_cleaned=recipesDTM_tidy_cleaned[!grepl(paste(dic_terms$term, collapse="|"), recipesDTM_tidy_cleaned$term),]
  
  #remove punctuations
  clean_recipesDTM_tidy_cleaned = recipesDTM_tidy_cleaned %>%
    mutate_all(funs(gsub("[[:punct:]]", "", .))) 
  
  clean_recipesDTM_tidy_cleaned = sqldf("SELECT * FROM clean_recipesDTM_tidy_cleaned WHERE count>1 ") #select count greater than 1
  clean_recipesDTM_tidy_cleaned = sqldf("SELECT * FROM clean_recipesDTM_tidy_cleaned WHERE LENGTH(term)>2 ") #select character greater than 2

}

#run DTM function
df_course_DTM = df_DTM(course_dataframe = df_course_cuisine$ingredients)



#create dataframe grouped (tidy)
cc_cleaned_documents <- df_course_DTM %>%
  group_by(document) %>% 
  mutate(terms = toString(rep(term, count))) %>%
  select(document, terms) %>%
  unique()

#FILTER
cc_cleaned_documents = stopw_filter(cc_cleaned_documents)
cc_cleaned_documents = no_numbers_filter(cc_cleaned_documents)
cc_cleaned_documents = no_newline_filter(cc_cleaned_documents)
cc_cleaned_documents = all_but_hyphen_filter_comma(x2)
cc_cleaned_documents = common_filter(cc_cleaned_documents)
cc_cleaned_documents = units_filter(cc_cleaned_documents)


#generate parts of speech
c_lean_POS = udpipe(cc_cleaned_documents[2],object=udmodel)

#INTERCHANGEABLE QUERIES (q1,q2,q3)-------------------------------BEGINS----------HERE-------
#--------for general topics (q1)
#c_l_ean_POS = sqldf("SELECT * FROM c_lean_POS WHERE upos NOT IN ('ADJ','ADV','PART','VERB','PROPN','INTJ','') ") #REMOVE Some Parts of Speech

#--------for adjective topics (q2)
c_l_ean_POS = sqldf("SELECT * FROM c_lean_POS WHERE upos='ADJ' ") #FILTER ONLY ADJECTIVES

#--------for noun topics (q3)
#c_l_ean_POS = sqldf("SELECT * FROM c_lean_POS WHERE upos='NOUN' ") #FILTER ONLY NOUNS

#INTERCHANGEABLE QUERIES (q1,q2,q3)-------------------------------ENDS-----------HERE-------

#plot after filtering PARTS OF SPEECH - LEMMA -------------------------------topics lemmas
top_terms_by_topic_LDA(c_l_ean_POS$lemma, number_of_topics = 5, plottitle = "Top Adjectives associated with Salads in American Cuisines")    #lemma topics plot