217 learning to integrate web catalogs with concep

Learning to Integrate Web Catalogs with Conceptual Relationships in Hierarchical Thesaurus: 

Learning to Integrate Web Catalogs with Conceptual Relationships in Hierarchical Thesaurus Jui-Chi Ro, Ing-Xiang Chen and Cheng-Zen Yang Dept. of Computer Science and Engineering Yuan Ze University, Taiwan, ROC 2006/10/17

Outline: 

Outline Introduction Problem Statement The Enhanced Catalog Integration Approach Experiments and Discussion Conclusions

Introduction: 

Introduction Many Web applications are in need of integrating different on-line vendors and Web portals. An integrated Web catalog service can help users gain more relevant and organized information in one catalog, and can save them much time to surf among different Web catalogs.

B2C e-commerce: Amazon: 

B2C e-commerce: Amazon

Web Catalogs: 

Web Catalogs

Introduction: 

Introduction Many Web applications are in need of integrating different on-line vendors and Web portals. An integrated Web catalog service can help users gain more relevant and organized information in one catalog, and can save them much time to surf among different Web catalogs.

The catalog integration problem: 

The catalog integration problem Catalog integration should not be a simple classification task. When some implicit source information is exploited, the integration accuracy can be highly improved. Past studies have shown that the Naïve Bayes classifier, SVMs, co-bootstrapping and the Maximum Entropy model with source information can enhance the accuracy of Web catalog integration in a flattened catalog integration structure.

Motivations: 

Motivations Past studies do not comprehensively study the hierarchical relationships between the categories and subcategories existing in the source and destination catalogs. We are motivated to conduct research on hierarchical catalog integration and study the effectiveness of the source and destination hierarchical information.

Problem statement: 

Problem statement

The flattened integration scheme: 

The flattened integration scheme

The hierarchical integration scheme: 

The hierarchical integration scheme

Enhanced catalog integration: 

Enhanced catalog integration SVM classifiers are used with linear kernel functions: Find the optimal values of w and b such that ||w|| is minimized. Enhanced hierarchical catalog integration (EHCI): (λ = 0.05 in destination catalog)

The weights assigned to different labels: 

The weights assigned to different labels The label weights in the source catalog:

The ECI-F integration scheme: 

The ECI-F integration scheme

The ECI-H integration scheme: 

The ECI-H integration scheme

Experiments and discussion: 

Experiments and discussion Experimental data: Measurement:

The ECI-F result (I): 

The ECI-F result (I) Catalog integration from Yahoo! to Google:

The ECI-F result (II): 

The ECI-F result (II) Catalog integration from Google to Yahoo!:

The ECI-H result (I): 

The ECI-H result (I) Catalog integration from Yahoo! to Google:

The ECI-H result (II): 

The ECI-H result (II) Catalog integration from Google to Yahoo!:

Performance comparison (I): 

Performance comparison (I) The accuracy performance from Yahoo! to Google. (λ = 0.15, ECI-F; λ = 0.2, ECI-H)

Performance comparison (II): 

Performance comparison (II) The accuracy performance from Google to Yahoo!. (λ = 0.15, ECI-F; λ = 0.2, ECI-H)

Results and discussion: 

Results and discussion The best averaged accuracy of the ECI-F and ECI-H is both over 94%. The accuracy improvement of a flattened structure is boosted over 11% on average. The accuracy improvement of a hierarchical structure is boosted over 16% on average. The integration accuracy with the ECI scheme is much better than the original integration scheme.

Conclusions: 

Conclusions Catalog integration is an important issue. We have shown the effects of an enhanced catalog integration (ECI) to boost the integration accuracy in both a flattened and a hierarchical structure. Our ECI approach consistently boosts the SVM classifiers in catalog integration The improvement of a hierarchical structure is more promising than a flattened one.

Thank You for Attention!: 

Thank You for Attention! sean@syslab.cse.yzu.edu.tw