library(lattice)

library(gdata)

library(RColorBrewer)

#################################################################################

# configuration flags

#################################################################################

# Set to 1(0) to load(compute) the partner similarity matrices

loadSim = 0

# Set to 1(0) to write(not-write) the JSON file

writeJSON = 1

#################################################################################

# load the data

#################################################################################

data = read.csv("Expedia/expedia_casestudy_20170127.csv", header = TRUE, sep = ",", dec = ".")

# list of partners

affiliate_id_list = unique(data$affiliate_id)

# total number of partners

num_of_affiliates = length(affiliate_id_list)

# list of markets

mkt_list = unique(data$mkt)

# total number of markets

num_of_mkts = length(mkt_list)

#################################################################################

# Compute total booking value per partner and per market

#################################################################################

# total booking value per partner and per market

totbkgs_per_id_per_mkt = aggregate(data$bkgs, by=list(affiliate_id=data$affiliate_id,mkt=data$mkt),FUN=sum)

# number of markets per partner

num_of_mkt_per_partner = aggregate(totbkgs_per_id_per_mkt$mkt, by=list(affiliate_id=totbkgs_per_id_per_mkt$affiliate_id), FUN=length)

#################################################################################

# Search for similar partners

#################################################################################

if (loadSim==1) {

# load the derived data frames

load("Expedia/affiliates_similarity_matrix.Rda")

load("Expedia/affiliates_shared_markets_matrix.Rda")

}else{

# A similarity score is assigned to each pair of partners.

# The score is defined as the number of common markets with similar booking value distributions

# divided by the total number of unique markters among the pair of partners.

# The similarity between booking value distributions is defined based on the

# p-value from the Mann-Whitney-Wilcoxon test evaluated at 95% CL

# define a dataframe to store the similary score

affiliates_sim <- data.frame("affiliate_id1" = integer(), "affiliate_id2" = integer(), "score" = numeric())

affiliates_shamkt <- data.frame("affiliate_id1" = integer(), "affiliate_id2" = integer(), "score" = numeric())

# split data per partern

data_split = split(data, f = data$affiliate_id)

# loop over partners

cat("Computing similarity matrix ...\n")

for (i in 1:num_of_affiliates) {

cat("Analysing data for partner ",i," ...\n")

data_split_i = data_split[[i]]

data_i = data_split_i[,2:3]

mkt_i = split(data_i, f = data_i$mkt, drop = TRUE)

# loop over other partners

for (j in i:num_of_affiliates) {

data_split_j = data_split[[j]]

data_j = data_split_j[,2:3]

mkt_j = split(data_j, f = data_j$mkt, drop = TRUE)

# define the score for current partner

score = 0

# compute the list of common markets among the two partners

mkts = intersect(data_i[,1], data_j[,1])

if( length(mkts) > 0 ) {

for(m in 1:length(mkts)) {

bkg_split_i = mkt_i[[mkts[m]]]

bkg_split_j = mkt_j[[mkts[m]]]

bkg_i = bkg_split_i[,2:2]

bkg_j = bkg_split_j[,2:2]

# neglects markets with less than 5 bookings

if (length(bkg_i) > 5 & length(bkg_j) > 5) {

Wtest = wilcox.test(bkg_i,bkg_j)

if (Wtest$p.value > 0.05) {

score = score + 1

}

}

} # end of loop over markets

} # if common markets exist

# store score in the data frame

affiliates_sim[nrow(affiliates_sim)+1,] <- c(i, j, score)

affiliates_shamkt[nrow(affiliates_shamkt)+1,] <- c(i, j, length((mkts)))

} # end of loop over other partners

} # end of loop over partners

save(affiliates_sim,file="Expedia/affiliates_similarity_matrix.Rda")

save(affiliates_shamkt,file="Expedia/affiliates_shared_markets_matrix.Rda")

}

#################################################################################

# build matrices for partner vs partner comparisons

#################################################################################

# fraction of shared markets

shaMktMatrix = matrix(data=0, nrow = num_of_affiliates, ncol=num_of_affiliates)

# fraction of markets with similar booking value distributions among pair of partners

simShaMatrix = matrix(data=0, nrow = num_of_affiliates, ncol=num_of_affiliates)

# fraction of markets with similar booking value distributions among pair of partners

# this is the quantity used to define when two partners are similar.

simMatrix = matrix(data=0, nrow = num_of_affiliates, ncol=num_of_affiliates)

for( i in 1:num_of_affiliates) {

num_of_mkt_i = num_of_mkt_per_partner[num_of_mkt_per_partner$affiliate_id==affiliate_id_list[i],2:2]

for( j in i:num_of_affiliates) {

num_of_mkt_j = num_of_mkt_per_partner[num_of_mkt_per_partner$affiliate_id==affiliate_id_list[j],2:2]

num_of_shared_mkt_i_j = affiliates_shamkt[affiliates_shamkt$affiliate_id1==i & affiliates_shamkt$affiliate_id2==j,3:3]

num_of_similar_mkt_i_j = affiliates_sim[affiliates_sim$affiliate_id1==i & affiliates_sim$affiliate_id2==j,3:3]

num_of_unique_mkt_i_j = num_of_mkt_i + num_of_mkt_j - num_of_shared_mkt_i_j

shaMktMatrix[i,j] = num_of_shared_mkt_i_j / num_of_unique_mkt_i_j

simShaMatrix[i,j] = num_of_similar_mkt_i_j / num_of_shared_mkt_i_j

simMatrix[i,j] = num_of_similar_mkt_i_j / num_of_unique_mkt_i_j

}

}

# symmetrise the matrices

lowerTriangle(shaMktMatrix) <- upperTriangle(shaMktMatrix, byrow=TRUE)

lowerTriangle(simShaMatrix) <- upperTriangle(simShaMatrix, byrow=TRUE)

lowerTriangle(simMatrix) <- upperTriangle(simMatrix, byrow=TRUE)

# data frame for final similarity score

affiliates_simfinal <- data.frame("affiliate_id1" = integer(), "affiliate_id2" = integer(), "score" = numeric())

for( i in 1:num_of_affiliates) {

for( j in 1:num_of_affiliates) {

affiliates_simfinal[nrow(affiliates_simfinal)+1,] <- c(i, j, simMatrix[i,j])

}

}

#################################################################################

# Write the JSON file

#################################################################################

if( writeJSON==1) {

library(jsonlite)

sink("Expedia/myjson.json")

djson = data.frame(factor(), data.frame(), list())

for (i in 1:num_of_affiliates) {

# compute the top 10 markets

mkt_i = totbkgs_per_id_per_mkt[totbkgs_per_id_per_mkt$affiliate_id == affiliate_id_list[i], 2:3]

mkt_i = mkt_i[order(mkt_i$x, decreasing = TRUE),]

mkt_i_top10 = head(mkt_i, n=10)

# compute the top 10 most similar partners

tmp = affiliates_simfinal[affiliates_simfinal$affiliate_id1==i,2:3]

affiliate_i_top10 = (head(tmp[order(tmp$score,decreasing = TRUE),1:1],n=11))[2:11]

# build a data frame to hold the info

djson_i <- data.frame(partner_name = c(as.character(affiliate_id_list[i])), row.names = i)

tmp <- data.frame(as.list(mkt_i_top10$x))

tmp <- setNames(tmp, as.character(mkt_i_top10$mkt))

djson_i$top10markets <- list(tmp)

djson_i$similar_partners <- list(as.character(affiliate_id_list[affiliate_i_top10]))

# append to final dataframe

djson = rbind(djson,djson_i)

}

cat(toJSON(djson, pretty=TRUE))

sink()

}

#################################################################################

# save plots and tables for similarity matrices

#################################################################################

rgb.palette <- colorRampPalette(brewer.pal(9, "Blues"), space = "rgb")

# matrices for similarity among partners

pdf("Expedia/plots/shared_market_matrix.pdf")

print(levelplot(shaMktMatrix, main="shared market matrix", xlab="partner id", ylab="partner id",

col.regions=rgb.palette(120), cuts=num_of_affiliates))

dev.off()

# matrices for similarity among partners

pdf("Expedia/plots/shared_similarity_matrix.pdf")

print(levelplot(simShaMatrix, main="shared similarity matrix", xlab="partner id", ylab="partner id",

col.regions=rgb.palette(120), cuts=num_of_affiliates))

dev.off()

# matrices for similarity among partners

pdf("Expedia/plots/global_similarity_matrix.pdf")

print(levelplot(simMatrix, main="global similarity matrix", xlab="partner id", ylab="partner id",

col.regions=rgb.palette(120), cuts=num_of_affiliates))

dev.off()