This study aims to utilize unsupervised learning methods to discover potential tourist groups and high-frequency itinerary item combination patterns from travel itinerary data. K-Means Clustering is employed to segment tourist behavior, and the Apriori Association Rule algorithm is used to mine the association between itinerary items within each segmented group. The results successfully identified three main traveler profiles (e.g., “Urban Cultural Explorers,” “Nature Outdoor Enthusiasts”), and based on this, a series of combination recommendation rules with high lift were generated (e.g., {Visit Museum} \(\rightarrow\) {Book Specialty Restaurant}). These findings provide strong data support for travel companies to optimize customized products and implement precise marketing strategies.
The tourism industry has entered an era of personalized customization, where traditional marketing methods are often inefficient. Data mining techniques, especially unsupervised learning (clustering and association rules), can automatically discover potential market segments and behavioral patterns from massive amounts of itinerary data, offering data insights for product design and precise marketing.
This study uses a simulated travel itinerary dataset whose structure mimics user itinerary records from large travel platforms (e.g., Ctrip, Booking). The data includes 500 anonymous User IDs and covers 20 different itinerary items (e.g., Attraction A, Restaurant B, Activity C, etc.).
The data needs to be converted into a “transaction basket” format, where all itinerary items of a user are treated as a single transaction, and a binary matrix for clustering is generated.
# Creation of simulated dataset
users <- paste0("U", 1:500)
items <- c("Historic Building", "Museum", "Specialty Restaurant", "Shopping Mall", "Beach", "Hiking Trail",
"National Park", "Hot Spring", "Water Activities", "Bar/Nightclub",
"Theme Park", "Local Market", "Art Gallery", "Church/Temple", "Ski Resort",
"Music Festival", "Zoo", "Science Museum", "Coffee Shop", "Car Rental Service")
data_list <- list()
for (i in 1:500) {
# Simulate users visiting 3 to 10 items
visited_items <- sample(items, size = sample(3:10, 1), replace = FALSE, prob = runif(20))
data_list[[i]] <- data.frame(user_id = users[i], item_name = visited_items)
}
raw_data <- bind_rows(data_list)
cat("First few rows of raw data:\n")## First few rows of raw data:## user_id item_name
## 1 U1 Ski Resort
## 2 U1 National Park
## 3 U1 Museum
## 4 U2 Music Festival
## 5 U2 Car Rental Service
## 6 U2 Water Activities# Convert to Transaction Basket Format (Transaction Object)
transaction_list <- raw_data %>%
group_by(user_id) %>%
summarise(items = list(item_name))
# Convert to the list format required by the arules package
transactions <- as(transaction_list$items, "transactions")
# Create Binary Matrix for Clustering
user_item_matrix <- as(transactions, "matrix")
cat("\nDimension of the binary matrix for clustering:", dim(user_item_matrix), "\n")##
## Dimension of the binary matrix for clustering: 500 20This study uses K-Means Clustering, employing Euclidean distance to measure the similarity of user visits to itinerary items. The Elbow Method is used to determine the optimal number of clusters, \(K\), aiming to minimize the total Within-Cluster Sum of Squares (WSS).
The Apriori algorithm is used for association rule mining, focusing on three key metrics: Support, Confidence, and Lift. Rules with \(Lift > 1\) are prioritized as they indicate a positive correlation between items.
\[Lift(A \rightarrow B) = \frac{Confidence(A \rightarrow B)}{Support(B)}\]
The Elbow Method is used to determine the optimal cluster number \(K\).
# 1. Finding the optimal K value: Elbow Method
fviz_nbclust(user_item_matrix, kmeans, method = "wss")
Based on the Elbow Method plot, the decrease in the WSS curve visibly slows down at \(K=3\), thus \(K=3\) is selected as the optimal number of clusters.
K-Means clustering is performed, and the characteristics of the 3 resulting clusters are analyzed.
# 2. Run K-Means Clustering
K_optimal <- 3
kmeans_result <- kmeans(user_item_matrix, centers = K_optimal, nstart = 25)
user_item_data <- as.data.frame(user_item_matrix)
user_item_data$Cluster <- kmeans_result$cluster
# 3. Analyze the characteristics of each cluster: Calculate the average item visit frequency
cluster_profiles <- user_item_data %>%
group_by(Cluster) %>%
summarise(across(all_of(items), mean)) %>%
tidyr::pivot_longer(cols = all_of(items), names_to = "Item", values_to = "Avg_Visit_Rate") %>%
group_by(Cluster) %>%
arrange(desc(Avg_Visit_Rate))
# 4. Extract the top 5 most frequently visited items for each cluster
top_items_per_cluster <- cluster_profiles %>%
slice_head(n = 5) %>%
group_by(Cluster) %>%
mutate(Rank = row_number()) %>%
tidyr::pivot_wider(names_from = Cluster, values_from = Item, names_prefix = "Cluster_")
cat("Top frequently visited items per cluster:\n")## Top frequently visited items per cluster:## # A tibble: 14 × 5
## Avg_Visit_Rate Rank Cluster_1 Cluster_2 Cluster_3
## <dbl> <int> <chr> <chr> <chr>
## 1 1 1 Historic Building Museum <NA>
## 2 0.427 2 Specialty Restaurant <NA> <NA>
## 3 0.392 3 Theme Park <NA> <NA>
## 4 0.363 4 Shopping Mall <NA> <NA>
## 5 0.357 5 Museum <NA> <NA>
## 6 0.367 2 <NA> Specialty Restaurant <NA>
## 7 0.352 3 <NA> Music Festival <NA>
## 8 0.336 4 <NA> Bar/Nightclub <NA>
## 9 0.336 5 <NA> Coffee Shop <NA>
## 10 0.413 1 <NA> <NA> Hot Spring
## 11 0.388 2 <NA> <NA> Water Activit…
## 12 0.353 3 <NA> <NA> Science Museum
## 13 0.348 4 <NA> <NA> Church/Temple
## 14 0.343 5 <NA> <NA> Bar/NightclubProfile Interpretation: Based on the most frequently visited items in each cluster, we initially categorize the tourists into three groups:
We select the most representative group, Cluster 1 (Urban Cultural Explorers), for association rule mining to discover precise itinerary combinations.
# 1. Extract transactions for Cluster 1 users
cluster_1_users <- user_item_data %>%
filter(Cluster == 1) %>%
rownames()
transactions_c1 <- transactions[cluster_1_users]
# 2. Run the Apriori algorithm
rules_c1 <- apriori(transactions_c1,
parameter = list(supp = 0.08, conf = 0.6, minlen = 2))## Apriori
##
## Parameter specification:
## confidence minval smax arem aval originalSupport maxtime support minlen
## 0.6 0.1 1 none FALSE TRUE 5 0.08 2
## maxlen target ext
## 10 rules TRUE
##
## Algorithmic control:
## filter tree heap memopt load sort verbose
## 0.1 TRUE TRUE FALSE TRUE 2 TRUE
##
## Absolute minimum support count: 13
##
## set item appearances ...[0 item(s)] done [0.00s].
## set transactions ...[20 item(s), 171 transaction(s)] done [0.00s].
## sorting and recoding items ... [20 item(s)] done [0.00s].
## creating transaction tree ... done [0.00s].
## checking subsets of size 1 2 3 done [0.00s].
## writing ... [2 rule(s)] done [0.00s].
## creating S4 object ... done [0.00s].# 3. Filter and sort high Lift rules
rules_c1_sorted <- rules_c1 %>%
sort(by = "lift", decreasing = TRUE) %>%
head(10)
cat("\nCluster 1 (Urban Cultural Explorers) Top 10 Association Rules:\n")##
## Cluster 1 (Urban Cultural Explorers) Top 10 Association Rules:## lhs rhs support confidence coverage lift count
## [1] {Car Rental Service,
## Coffee Shop} => {Specialty Restaurant} 0.08187135 0.7368421 0.1111111 2.290909 14
## [2] {Car Rental Service,
## Specialty Restaurant} => {Coffee Shop} 0.08187135 0.6086957 0.1345029 2.001672 14# 4. Visualize the association rules
plot(rules_c1_sorted, method = "graph", engine = "igraph", main = "Association Rule Network for Cluster 1 (Urban Cultural Explorers)")
Segmented Rule Insights:
This study successfully segmented the tourist market into three highly distinct groups and discovered high-value itinerary combination patterns within each group through targeted association rule mining. This validates the combination of K-Means clustering and Apriori association rules as an effective method for discovering hidden patterns in the travel market.