ppt-milestone.Rmd

---
title: "Mall Tenant Synergy Visualization using GraphViz"
author: "Team Algoritma"
date: "Last update: 6th September 2019"
output: 
  revealjs::revealjs_presentation:
    theme: serif
    df_print: paged
---

## Background

```{r echo=FALSE, message=FALSE, warning=FALSE}
suppressPackageStartupMessages(library(dplyr))
suppressPackageStartupMessages(library(tidyr))
suppressPackageStartupMessages(library(DT))
suppressPackageStartupMessages(library(arules))
suppressPackageStartupMessages(library(stringr))
suppressPackageStartupMessages(library(visNetwork))
suppressPackageStartupMessages(library(ggplot2))
suppressPackageStartupMessages(library(RColorBrewer))
```

**Case Study: Mall Tenant Analysis**

In order to manage and sort a highly valuable tenants, we need to be able to identify how much the tenant’s financial value can it brought in, and how does the business itself affect other tenants (synergy). A high synergy relationship will also indicates how the tenants will attract a same group of customers.

## Approach: Customer Shopping Association Rules

By looking at how customer’s probability of going to multiple stores in a mall we can generate its support to measure store’s popularity and its lift to measure the likelihood relationship of two stores visited together.

Main attribute of association rules using market basket analysis can be further breakdown into the following:  

- Support    
- Lift  
- Confidence  

## Dummy Transaction Data

Given the following dummy data, we'll try to implement market basket analysis:

```{r echo=FALSE}
transactions <- read.csv("data_input/transactions.csv")
DT::datatable(transactions, options = list(pageLength=5))
```

## Dummy Tenant and Members Data

The tenant examples presented in this examples are the following:

```{r echo=FALSE}
levels(transactions$Tenant)
```

Currently we use a limited number of members sample:

```{r, echo=FALSE}
levels(transactions$CostumerID)
```

## Transaction Type for Market Basket

```{r}
trans <- transactions %>% 
  select(-DiscountVoucher, -SalesSubTotal)

basket <- split(x = trans$Tenant, # column to be splitted into groups
                f = list(trans$CostumerID,trans$Date), # splitted column to be grouped by..
                drop = T) # drop groups which have no item (Tenant)

trans <- as(basket, "transactions")
trans
```

## Popularity Metrics using Support

```{r, echo=FALSE}
itemFrequency(trans) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  ggplot(aes(x=reorder(rowname, -`.`), y=`.`)) +
  geom_col(fill = "#d63d2d") +
  theme_minimal() +
  theme(
    axis.text.x = element_text(angle = 40, hjust=1)
  ) +
  labs(title = "Support of Tenants (Popularity)",
       y = "Frequency", x = "Tenant")
```

## Popularity Metrics using Support (cont.)

```{r echo=FALSE}
# finding Support of Tenants
tenant.support <- data.frame(Support = itemFrequency(trans))
  

DT::datatable(tenant.support, options = list(pageLength=5))
```

## Extract Apriori Rules

```{r echo=FALSE}
# making association rules
mall_arules <- apriori(data = trans,
                       parameter = list(confidence=0.75, maxlen=2),
                       control = list(verbose = F))
mall_arules
```

```{r echo=FALSE}
# to retrieve the first 10 rules
mall_arules <- DATAFRAME(mall_arules) %>%
  mutate(LHS = as.character(LHS),
         RHS = as.character(RHS)) %>% 
  mutate(LHS = str_replace_all(LHS, "\\{|\\}", ""),
         RHS = str_replace_all(RHS, "\\{|\\}", ""))
DT::datatable(mall_arules, options = list(pageLength=5))
```

## Graph Data Preparation

```{r warning=FALSE}
#Edges
edges <- mall_arules %>% 
  rename(from = LHS, to = RHS,
         weight = confidence) %>% 
  arrange(weight) %>% 
  mutate(color=rev(terrain.colors(nrow(mall_arules))),
         width=weight^5*10,
         arrows="to")

# Nodes
nodes <- tenant.support %>% 
  tibble::rownames_to_column() %>% 
  rename(id = rowname, value = Support) %>% 
  mutate(size=value/max(tenant.support$value)*30)
```

## vizNetwork

```{r echo=FALSE}
library(visNetwork)
set.seed(100)
visNetwork(nodes=nodes,edges=edges, width="100%") %>% 
  visLegend() %>% 
  visOptions(nodesIdSelection = T,highlightNearest = T) 
```

## TODO

// DYNAMIC USER RULES EXTRACTION USING INTERACTIVE WIDGET