skincare-recommend.R

# auxiliary function to compute cosine similarity between two vectors
cossim <- function(x,y){
  return (sum(x*y)/sqrt(sum(x*x))/sqrt(sum(y*y)))
}

# "recommend" takes in a query(qr), number of products to return (number) 
# and a dataframe of precalculated tf-idf for skincare products(dt)

recommend <- function(qr,number,dt){
  # product_words that contains words in the query
  product_words_sub <- dt%>%
    filter(word %in% qr$word)%>%
    select(word,Product,tf_idf)
  
  # spread dataframe into column for Product
  product_tf_idf <- spread(product_words_sub , key = Product, value = tf_idf)
  
  # Assign zero to NA to perform cosine similarity operation
  product_tf_idf[is.na(product_tf_idf)] <- 0
  product_tf_idf <- product_tf_idf[order(product_tf_idf$word),]
  
  # label a third row for the cossine similarity values
  product_tf_idf[dim(product_tf_idf )[1]+1,1] = "cossim"
  
  # perform cosine similarity with query
  for(i in 2:dim(product_tf_idf)[2]){
    product_tf_idf[dim(product_tf_idf)[1],i] = 
      cossim(product_tf_idf[1:dim(product_tf_idf)[1]-1,i],qr$tf_idf)
  }
  
  # manipulate data to be returned as data frame
  result <- t(product_tf_idf[,-1])
  colnames(result) <- product_tf_idf$word
  result<-as.data.frame(result)
  result$Product <- rownames(result)
  
  #arrange and return top_n(number) results
  result <- result%>%
    select(Product, cossim)%>%
    arrange(desc(cossim))%>%
    top_n(number)
  
  # restrict the amount of returned values to "number" in case of ties
  if(dim(result)[1] > number){
    return (result[1:number,])
  }else {
    return (result) 
  }
}