IMPLEMENTATION
To prove the quality of the hybrid recommendation mechanism, we implemented the prototype system using the Excel and VBA languages in a Windows XP environment. We call this prototype system CAR (CBR & Association rule-based Recommendation systems). CAR is composed of five components (Figure 3). The five components are: (1) rule generator, (2) knowledge base, (3) inference engine, (4) justifier, and (5) user interface.
Figure 3: The Structure of CAR
Phase I: Association Rule Generation
Web log data, which was used in web data mining, was collected from an Internet-based RC (remote-controlled plastic model) shopping mall. This shopping mall focused on selling remote-controlled products, such as cars , tanks, helicopters, gliders, yachts and ships. The original web log data was contaminated by several types of irrelevant and redundant information including slashes (/, \), file name suffixes (htm, html, gif, jpg, jsp, etc.), and other information for query communications (&, =, <=, ?, etc.).
To mine a meaningful set of association rules from the web log database, the first step is to cleanse the original web log data so that the preprocessed web log data may become more traceable (Lee et al., 2002). Figure 4 shows a preprocessed web log database.
Figure 4: Preprocessed Web Log Database
The web data mining algorithm we adopted here is an APRIORI algorithm (Agrawal et al., 1993a, 1993b), which is known to yield a set of association rules. Based on the preprocessed web log database in Figure 4, the corresponding association rules were extracted with a threshold of 20 percent confidence. Table 2 shows an excerpt of the derived association rules. The association rules shown in Table 2 are straightforward and easy to understand and interpret.
 |
| Pocket Booster (Checker) | <= | RC car guidebook (5:4.673%, 0.2) |
| ACE 2000 | <= | 7.2V low speed charger (5:4.673%, 0.4) |
| 7.2V low speed charger | <= | ACE 2000 (4:3.738%, 0.5) |
| 15% SM15 (1G) | <= | 7.2V low speed charger (5:4.673%, 0.2) |
| Booster | <= | Plus wrench (S) (3:2.804%, 0.667) |
| GP 20(1Q) | <= | Booster charger (3:2.804%, 0.667) |
| |
Phase II & III: Case Generation & Construction of Hybrid Knowledge Base
In this phase, we briefly outline the CBR mechanism, which may help the decision maker in classifying cases which occur in the web log database. The concepts of similarity and similarity relations used in CBR play a fundamental role in many fields of pure and applied science.
The simplest CBR or CBL ( Case-Based Learning) algorithm is CBL1. Its preprocess linearly normalizes all numeric feature values (Aha, 1991). CBL1 defines the similarity of cases C 1 and C 2 as:
where P is the set of predictor features and
The CBL algorithm used in this study is summarized in Table 3.
| |
| |
| |
The prototype system CAR supports a CBL algorithm shown in Table 3, and transforms the case extraction results into a case-based knowledge base. Figure 5 summarizes seven cases extracted from the web log database and the customer profile database.
Figure 5: Case-Based Knowledge Base
After the extracting association rules and related cases, the rule-based knowledge base and the case-based knowledge base are combined using the customer's profile and web log information. At this time, the most important key points are the customer's ID and his web surfing information.
Phase IV: Hybrid Recommendation
The prototype system CAR uses the rule-based knowledge base and the case-based knowledge base concurrently. After the hybrid knowledge base is built, CAR can execute inference. In this phase, CAR may suggest the results of hybrid recommendation to the customer and then wait for the customer's feedback and response. Before the inference, Table 4 shows the web customer's brief profile and preferences to validate our hybrid recommendation mechanism.
| Customer's profile |
| Birth: February 1963 / Sex: Male / Position: Businessman / |
| Experience (career): 7 months / |
| Interest: Car (remote controlled car) |
| Customer's preference : |
| Purchasing the Guidebook for remote-controlled car |
First, the customer will search and select the guidebook for a remotecontrolled car. If this web site is a common shopping mall, however, he can't get additional information about the ability to control the remotecontrolled car. Therefore, the web site may lose this potentially loyal customer. In this case, CAR can present more intelligent and additional information to customers. Figure 6 shows the hybrid recommendation results of CAR.
Figure 6: Hybrid Recommendation Results of CAR
In Figure 6, the customer finds additional information describing other products suggested by CAR. Finally, the recommended products (information) are ˜SuperNova 3000S (re-charger for a worn out battery), ˜Switching Power 15A (high capacity power supplier), and ˜3-Mode Charger (re-charger for remote controller, receiver and battery). These products are the most important and basic goods for controlling the remote-controlled plastic models. As a result, the customer may purchase the product he wants and, at the same time, find additional products.
EAN: N/A
Pages: 171