untitled

Size: px

Start display at page:

Download "untitled"

筑蔡
5 years ago
Views:

1 An Improved Semantic Search Model Based on Hybrid Fuzzy Description Logic Ruixuan Li, Kunmei Wen, Zhengding Lu, Xiaolin Sun, Zhigang Wang Huazhong University of Science and Technology

2 Background Semantic search (R. Guha, WWW 2003) Semantic Web + Search Engine finds out the internal knowledge in Knowledge Base (KB) to improve the search results gained by current search engines evolves to next generation of search engines built on Semantic Web Two types of search searching the current Internet searching the Semantic Web portals FCST 2006, Aizu 2

3 Related Work-Tap FCST 2006, Aizu 3

4 Tap FCST 2006, Aizu 4

5 Related Work Swoogle is a prototype system of information retrieval (IR). The search results are physical documents on Semantic Web (such as RDF and OWL files). It has not used the semantic structure information in documents. Turing center in the University of Washington develops the system KnowItAll to extract the information on the Web. Its long-term aim is to replace the search engine by information extraction FCST 2006, Aizu 5

6 Related Fields Description Logic (DL): Tableaux algorithm & many optimized. Classical DLs can only define the certain concepts and properties, which cannot solve the fuzzy problem of ontology system. Fuzzy DL are designed to expand the classic DLs to make it more applicable to ontology system. Ranking the search results: the number of relationships between entities in a KB will be much larger than the number of entities themselves FCST 2006, Aizu 6

7 Our Focus Combine techniques of semantic reasoning and information retrieval Improve semantic search model by supporting imprecise and fuzzy search FCST 2006, Aizu 7

8 Outline Semantic Search Model Architecture of Semantic Search Model RBAC Security Ontology Experiment and Evaluation Conclusion FCST 2006, Aizu 8

9 Basic Model Idea: integrating search and inference Formal DL reasoning: search the resources and relationships Traditional IR: locate the exact resource Basic model: Triple[D, Q, R(d i, Q)] D is the set of documents d i Q is a query R(d i, Q) is the similarity between d i and Q, here d i D Here documents and queries are modeled as individuals and concepts respectively FCST 2006, Aizu 9

10 Basic Model Basic model formal DL (ALC constructors) + traditional IR FCST 2006, Aizu 10

11 Query Form Different users have different privileges for different resources A query is defined as the form Q i = Q i1 Q i2 Q i3 Q i1 means a concept Q i2 is any formal query about resources or the relationships between them Q i3 is a keyword query FCST 2006, Aizu 11

12 Query Form Four typical queries as follows: Q i1 : Concept Query, form as Q i2 = C where means a concept. Q i2 : Relationship Query, form as Q i2 = C1 & C2 where C1 and C2 are concepts. Q i3 : Keyword Query, form as Q i3 = D where D means a keyword which appears in the text. In fact, Q i3 belongs to traditional query. Q i1 Q i3 : Conjunctive Query, form as Q i1 Q i3 = ( A or B ) D where A means a user, B means a role and D means a keyword FCST 2006, Aizu 12

13 Reasoning Based on Description Logic Concept Reasoning. Given Q i = Q i1 where Q i1 means concept, we can get all the sub concepts and all the instances are returned. Relationship Reasoning. Given Q i = Q i2 where Q i2 includes two concepts, we can get the relationship between them or null if there is not any relationship. Conjunctive Query Reasoning. In fact it integrates inference with search by providing both formal query and keyword query. Given Q i = Q i1 Q i3 where Q i1 means user or role and Q i3 is a keyword query FCST 2006, Aizu 13

14 Result Ranking R i = R i1 + R i2 + R i3 for the query Q i = Q i1 Q i2 Q i3. Given Q i1, if d is the instance of the query Q i1 then the corresponding R i1 =1, otherwise the value equals to 0. R i2 is determined by the important value of the relationship. Not deeply discussion here, need to be further explored. Use traditional tf-idf (term frequency inverse document frequency) method to compute the value of R i FCST 2006, Aizu 14

15 Extended Model Extended model formal DL + Type-1 FALC Type-1 FALC Syntax and semantics FCST 2006, Aizu 15

16 Extended Model Classic DL such as ALC cannot deal with the imprecise description Extended model: use type-1 FALC to replace IR model. The fuzzy degree is equal to the ranking value computed by traditional IR FCST 2006, Aizu 16

17 Complete Model Exist imprecise terminological axiom (TBox) and fuzzy individual membership (ABox). It is necessary to maintain trust degree for both of TBox and ABox. Complete model Type-1 FALC + Type-1 FALC FCST 2006, Aizu 17

18 Complete Model Propose type-2 FALC based on type-2 fuzzy sets, Type-2 FALC syntax and semantics FCST 2006, Aizu 18

19 Complete Model The complete model use type-2 fuzzy ALC to replace the formal DL in the extended model. In real world, there are the imprecise terminological axiom (TBox) and fuzzy individual membership (ABox) We apply type-2 fuzzy ALC to deal with the description in ontology for trust degree management (OntoTD) We design a simple ontology named OntoTD for trust management FCST 2006, Aizu 19

20 OntoTD FCST 2006, Aizu 20

21 Outline Semantic Search Model Architecture of Semantic Search Model RBAC Security Ontology Experiment and Evaluation Conclusion FCST 2006, Aizu 21

22 Architecture of Semantic Search Model Query Interface Query Processor Traditional Search Engine Initial Results User Inference Stop Controller Inference Engine Results Interface Result Ranking Engine Knowledge Base Other Ontology Ontology Translator Ontology Base FCST 2006, Aizu 22

23 Outline Semantic Search Model Architecture of Semantic Search Model RBAC Security Ontology Experiment and Evaluation Conclusion FCST 2006, Aizu 23

24 RBAC Security Policy Great demands: Secure Search, such as Intranet search which must satisfy requirements of access control in the background of government or business. Role-Based Access Control (RBAC) Various ways of specifying policy in the last decade Ontology-based approach A security ontology based on RBAC policy FCST 2006, Aizu 24

25 OntoRBAC FCST 2006, Aizu 25

26 RBAC Security Ontology We implement the security ontology using OWL DL as the ontology language < owl : Classrdf : ID = Subject >< rdfs : subclassof > < owl : Classrdf : ID = Entity / >< /rdfs : subclassof > < /owl : Class >< owl : Classrdf : ID = Role > < rdfs : subclassofrdf : resource = #Subject / >.. < /owl : ObjectP roperty >< owl : ObjectPropertyrdf : ID = hasp riv > < rdfs : rangerdf : resource = #Privilege / > < rdfs : domainrdf : resource = #Subject / >< /owl : ObjectProperty > FCST 2006, Aizu 26

27 Instance of Roles Director ProjectLeader1 ProjectLeader2 ProductionEngineer1 QualityEngineer1 ProductionEngineer2 QualityEngineer FCST 2006, Aizu 27

28 Instance of PolicyRules FCST 2006, Aizu 28

29 Outline Semantic Search Model Architecture of Semantic Search Model RBAC Security Ontology Experiment and Evaluation Conclusion FCST 2006, Aizu 29

30 Experiment Settings We implement Ontology Security Semantic Search Engine (Onto-SSSE) in Java Use the Lucene as the traditional search engine based on keyword query Use Jena as the reasoning tool based on RBAC security ontology Only implemented basic model FCST 2006, Aizu 30

31 Experiment Settings No commonly agreed evaluation methodology and benchmark for semantic search The OntoRBAC evaluation dataset OntoRBAC security ontology (including 12 classes, 16 properties and 20 individuals) The set of campus web pages (more then 200MB) FCST 2006, Aizu 31

32 Semantic Search Results FCST 2006, Aizu 32

33 Comparison with Traditional Search FCST 2006, Aizu 33

34 Conclusion Three semantic search models including basic model, extended and complete one. Combine text IR with semantic inference in the model. RBAC security ontology - OntoRBAC. A semantic search system (Onto-SSSE) is implemented based on the basic model FCST 2006, Aizu 34

35 Future Work We plan to achieve improvement in the following aspects: Perform search in a larger dataset. Implement the extended model and complete model. Improve the reasoning efficiency FCST 2006, Aizu 35

36 Thank you! FCST 2006, Aizu 36

University of Science and Technology of China A dissertation for master s degree Research of e-learning style for public servants under the context of

中国科学技术大学硕士学位论文新媒体环境下公务员在线培训模式研究作者姓名 : 学科专业 : 导师姓名 : 完成时间 : 潘琳数字媒体周荣庭教授二一二年五月 University of Science and Technology of China A dissertation for