Transcript CS548S15_Showcase_Association_Rules

CS548 Spring 2015
Association Rule Mining
Showcase
by Cory Hayward and Marcus Moyses
Showcasing work by Ting, Pan, and Chou on "Finding Ideal
Menu Items Assortments: An Empirical Application of Market
Basket Analysis"
Sources
Ting, P., Pan, S., & Chou, S. (2010). Finding Ideal Menu Items Assortments: An
Empirical Application of Market Basket Analysis. Cornell Hospitality Quarterly 51 (4), 492501
Agrawal, R., Imielinski, T., & Swami, A. (1993). Mining association rules between sets of
items in large databases. ACM SIGMOD International Conference on Management of
Data, 207-216
Agrawal, R., & Srikant, R. (1994). Fast algorithms for mining association rules. 20th
International Conference on Very Large Data Bases, 487-499
Whitehorn, M. (2006). The parable of the beer and diapers. The Register, from
http://www.theregister.co.uk/2006/08/15/beer_diapers/
Definitions
Association Rule Mining
● Method for discovering interesting relations between variables in large databases
Applications
● Market Basket Analysis
● Intrusion Detection
● Web Usage Mining
● Medical Research
● Recommender Systems
● Fraud Detection
Market Basket Analysis
Data Analysis and Data Mining technique used to discover co-occurrence relationships
among activities performed by individuals or groups in a retail setting
Has several marketing applications including
● promotional pricing
● product placement
● cross-selling
● up-selling
Classic Example: Beer and Diapers
Popular cross-promotional example using Market Basket Analysis
● cross-promotional, or cross-selling, involves selling an additional product or service
to an existing customer
Through association analysis, a drugstore found that beer had a strong relationship with
diapers
Theories can then be created to understand why these relationships occur (fathers who
buy diapers also buy beer for the weekend) and actions could be taken to improve
product placement
Measurements Example
Transaction 1: Rice, beef, bread
Transaction 2: Beef, potato chips
Transaction 3: Rice, bread
Transaction 4: Beef, milk
Transaction 5: Bread, milk
Analysis for Rice and Bread:
Support: 2 / 5 = 40%
Confidence: .4 / .4 = 100%
Lift: 1 / .6 = 1.667
Taken from: Ting, Pan, Chou (2010)
Measures of Interestingness
Support: probability that a randomly selected set of transactions from a database include
items A and B
Formula: P(A ∩ B)
Confidence: probability that a randomly selected set of transactions will include B given
that they include A
Formula: P(B | A) = P(A ∩ B) / P(A)
Lift: improvement in probability of B occurring in a transaction given that the transaction
includes A
Formula: P(B | A) / P(B) = P(A ∩ B) / (P(A)*P(B))
Showcase Application:
"Finding Ideal Menu Items Assortments: An
Empirical Application of Market Basket Analysis"
by Ting, Pan, and Chou
The Data Set
Taken from: Ting, Pan, Chou (2010)
The Problem
Japanese-style restaurant in Taiwan with fixed-price meals (includes entrée +
appetizer/soup + starch side dish)
What are the ideal combinations of:
● entrée + appetizer/soup
● entrée + starch side dish
● entrée + drink
● entrée + dessert
Why not find the ideal combination of:
entrée + appetizer/soup + starch side dish + drink + dessert ?
Pair-Occurrence vs. Co-Occurrence
Ideal association of menu items is determined by the confidence value of pairs of entrées
and side dishes
● only sets of pairs are important, while individual items have less significance
Certain limitations make pair-occurrence preferable for analysis:
● too many unimportant combinations with co-occurrence
o in a restaurant, if all combinations were considered each would only account
for a small percentage of sales
● limit on free choice for items (such as in a fixed-price restaurant)
● easier to memorize results from pair-occurrence
5 Step Process
1. Retrieve restaurant transaction data
2. Analyze data with Excel’s PivotTable
a. calculate support and confidence for each
item set
3. Identify ideal menu item assortments
4. Validate the findings on real-world transactions
5. Present conclusion and recommendations
Taken from: Ting, Pan, Chou (2010)
Point of Sales Transactions
Taken from: Ting, Pan, Chou (2010)
Data set: data collected in 85 days => 3,727 transactions
Combinations of 24 entrées, 7 appetizers or soups, 2 starch side dishes, 23 drinks and
17 desserts
Association Results
Taken from: Ting, Pan, Chou (2010)
Preferences
Entrée and appetizer/soup: most expensive items
Starch side dish: bread had a strong reputation among customers
Drink: hot coffee had free refill; iced tea had the largest portion size
Dessert: mango cheese was free
Recommendations Based on Results
Taken from: Ting, Pan, Chou (2010)
Overall success: 40/145 × 23.75% + 61/145 × 63.52% + 44/145 × 71.02% = 54.82%
Suggestions for Restaurant Owner
Employee training: memorize ideal menu combinations for each entrée and focus on
giving suggestions to undecided or customers without preference
Test price demand elasticity: increase price of popular items to see if revenue also
increases
Questions?