3 Cases on Business Intelligence MIS

Case 1 : Market Basket Analysis with Association Rule Data Collection The data set for the tests is collected from a convenience store G About 1,000 transactions are selected for a specific item The data has information on Date_Purchased, Time_Purchased, POS, Employee Number, Receipt Number, Item_Name, Quantity, Purchase_Cost, and Description

Rice rolled in dried laver The items in the data set are classified into 21 categories based on their characteristics as below: Category Definition 1 Processed food 12 Household items 2 Health beverage 13 Distilled liquor 3 Sweets 14 Newspaper 4 Rice rolled in dried laver 15 Yogurt 5 Frozen food 16 Milk 6 Tobacco 17 Juice 7 Instant noodle 18 Chocolate 8 Beer 19 Candy 9 Ice cream 20 Coffee 10 Bread 21 Carbonated beverage 11 Spring water  

Association rule Model A typical association rule has an implication of the form A → B where A is an item set and B is an item set that contains only a single atomic condition The support of an association rule is the percentage of records containing item sets A and B together Support represents the usefulness of discovered rules The confidence of a rule is the percentage of records containing item set A that also contain item set B the confidence represents certainty of the detected association rule Association rule technique is implemented with SAS E-Miner 4.0 with minimum support of 0.5%, and minimum confidence of 10%

Health beverage and Bread → Yogurt Support (%) Confidence (%) Explanations   5.5 82 Chocolate → Milk 3.5 86 Processed food → Milk 57 Tobacco → Candy 2 50 Health beverage and Bread → Yogurt 1.5 67 Sweets and Bread → Beer Newspaper → Tobacco 1 100 Milk and Candy → Chocolate Ice cream → Household items Sweets and Frozen food → Instant noodle Health beverage and Frozen food → Coffee Ice cream → Sweets Association rules

Case 2 : Predicting Customers’ Decision on Purchasing in Web Stores This case predicts customer behavior (buy or not buy) in web store On-line customers are likely to move sequentially from item to item in web stores.

Given a session in a website, we examine our model to express detail of clickstream dataset in the session A session means the entire processes from the initial visit to a site to the exit from the site In Figure 2, the number of parentheses is the duration time(in seconds), which indicates how long the user stayed in the web page

An example of a customer session

Results and Discussion We acquired a clickstream dataset and divided it into two, one is a dataset to buy, the other not to buy. ① AvgVisitTime: Total time of a session/No. of items visited ② SwitchCat: Total number of switching categories/No. of items visited ③ Re-visit: Total number of specific pages which are revisited 3 methods to predict customer behavior

Prediction accuracy of the models To investigate the effectiveness of these 3 methods, we conducted experiments with these 3 methods respectively using test data. Table 1 shows the results. Methods Average of Prediction (%) AvgVisitTime Train: 69.9 Test: 70.2 SwitchCat Train: 57.9 Test: 60.8 Re-visit Train: 60.0 Test: 60.0

A suggested method with association rule for predicting customer behavior

Business Intelligence Case 3 : Twitter mood predicts the stock market Twitter Can Predict the Stock Market movement. Behavioral economics tells us that emotions can profoundly affect individual behavior and decision-making. Does this also apply to societies experience mood states that affect their collective decision making? Is the public mood correlated or even predictive of economic indicators? We investigate whether measurements of collective mood states derived from large-scale Twitter feeds are correlated to the value of the Dow Jones Industrial Average (DJIA) over time

We analyze the text content of daily Twitter feeds by two mood tracking tools, namely OpinionFinder that measures positive vs. negative mood and Google-Profile of Mood States (GPOMS) that measures mood in terms of 6 dimensions (Calm, Alert, Sure, Vital, Kind, and Happy). Our results indicate that the accuracy of DJIA predictions can be significantly improved by the inclusion of specific public mood dimensions but not others We find an accuracy of 87.6% in predicting the daily up and down changes in the closing values of the DJIA

Approach to predict stock index (Up or down) Using OpinionFinder and Google-Profile of Mood States, they predict mood for stock market investment. OpinionFinder is an open source tool measuring the opinions in twitter, which are positive or negative. It is combined with the psychological tool, Profile of Mood State, which combines 72 emotional status words with GPOMS, producing 6 mood words: Calm, Alert, Sure, Vital, Kind, and Happy

Mood graph with 6 mood index

12 combinations, positive/negative x 6 mood, are used to analyze 2700,000 twitter’s 9800,000 twitts between Feb. 2008 to Dec. 2008.  The data can be found in http://terramood.soic.indiana.edu/annotated_line.php

Movement of mood

Dow Jones and Calm: correlated gaph

Test Results The test show the up and down of stock market at the prediction accuracy of 86.7%.