Outline Software Demo A Simple RuleBased Expert System

Outline • Software Demo : A Simple Rule-Based Expert System • Rule的推導 • 輔助Rule推導的技術 Ø如何自動收集knowledge Ø如何減少人的因素 Ø使用IT技術 ØES的validation

Vehicles Rule Base • • • • Bicycle: IF vehicle. Type=cycle AND num_wheels=2 AND motor=no THEN vehicle=Bicycle Tricycle: IF vehicle. Type=cycle AND num_wheels=3 AND motor=no THEN vehicle=Tricycle Motorcycle: IF vehicle. Type=cycle AND num_wheels=2 AND motor=yes THEN vehicle=Motorcycle Sports. Car: IF vehicle. Type=automobile AND size=small AND num_doors=2 THEN vehicle=Sports_Car

Vehicles Rule Base(cont. ) • • • • • Sedan: IF vehicle. Type=automobile AND size=medium AND num_doors=4 THEN vehicle=Sedan Mini. Van: IF vehicle. Type=automobile AND size=medium AND num_doors=3 THEN vehicle=Mini. Van SUV: IF vehicle. Type=automobile AND size=large AND num_doors=4 THEN vehicle=Sports_Utility_Vehicle Cycle: IF num_wheels<4 THEN vehicle. Type=cycle Automobile: IF num_wheels=4 AND motor=yes THEN vehicle. Type=automobile

A Forward-Chaining Example • Step 1. 使用者輸入已知的facts num_wheels=4 motor=yes num_doors=3 size=medium • Step 2. Automobile : IF num_wheels=4 AND motor=yes THEN vehicle. Type=automobile

A Forward-Chaining Example(cont. ) • Step 3. 更新working memory num_wheels=4 motor=yes num_doors=3 size=medium vehicle. Type=automobile • Step 4. Min. Van : IF vehicle. Type=automobile AND size=medium AND num_doors=3 THEN vehicle=Mini. Van

A Forward-Chaining Example(cont. ) • Step 5. 更新working memory num_wheels=4 motor=yes num_doors=3 size=medium vehicle. Type=automobile vehicle=Mini. Van

A Forward-Chaining Example • Step 1 Vehicle=Mini. Van • Step 2 Mini. Van : IF vehicle. Type=automobile AND size=medium AND num_doors=3 THEN Vehicle=Mini. Van • Step 3 Automobile : IF num_wheels=4 AND motor=yes THEN vehicle. Type=automobile

A Forward-Chaining Example(cont. ) • Step 4. 使用者給予資訊或結束推論 num_wheels=4 motor=yes vehicle. Type=automobile • Step 5. 使用者給予資訊或結束推論 num_wheels=4 motor=yes vehicle. Type=automobile size=medium num_doors=3 vehicle=Mini. Van

Induction Table Example(11. 19)

Induction Table Example(cont. ) • IF population density ≧ 2, 000 – AND density ≧ 4 – AND #(near competitors) = 0 THEN Decision = YES • In real system, multiple induction tables will be used, and factors may appear in several tables. These knowledge chains will be used by inference engines.

Repertory Grid Analysis(11. 10) (RGA) • • Step 1. identify the important objects Step 2. identify the important attributes Step 3: identify the traits and their opposites Step 4: grade each attributes

Example of RGA

Major tasks of the knowledge engineer(11. 11) • Figure 11. 7 • 一堆技術(參照課本)

Deficiencies of manual knowledge acquisition(11. 12) • • • Knowledge engineer難找又很貴 Expert很忙或職位輪調(知識難保存) Knowledge難取得(Expert保留或難描述) Knowledge engineer不懂domain knowledge Expert不懂ES technology • Donald Michie: 「只有當domain knowledge是確定 性高和範圍小, 才適合用interview或observation」 減少人的因素 (Automated Knowledge Acquisition )

Automated Knowledge Acquisition ---Machine learning • Data mining(e. x. , basket analysis菜籃分析) • A rule induction system(只要餵它足夠的 examples, 它就會自動產生rule) • Software package(not only automatically generate rules but also check them for possible logical conflict) • Intelligent agents(e. x. , KQML和KIF)

Automated Knowledge Acquisition ---Machine learning(cont. ) • Case-based reasoning Domain Knowledge很難用 if-then 的規則 方式表達，但卻有大量的案例足以取樣， 則用 CBR的方式來解決問題，應是不錯 的選擇。

Knowledge acquisition from multiple experts(11. 14) • Benefits ---群體思考的優點 • Problems ---群體思考的缺點

The collected knowledge must be analyzed, coded, and documented(11. 16) • Step 1. TRANSCRIPTION verbal report 書面報告(包含無關的資料) • Step 2. PHRASE INDEXING • Step 3. KNOWLEDGE CODING • Step 4. DOCUMENTATION

Numeric and Documented Knowledge Acquisition(11. 17) • Knowledge不是來自expert – Acquisition of Numeric Knowledge – Acquisition of Documented Knowledge Intelligent Agent, Scan database, 其他的ES

Knowledge Acquisition and the Internet/Intranet(11. 18) • Anytime, anywhere擴充溝通管道 • Demo : RGA+Web = Web. Grid • Web search engine + Intelligent agents (e. x. , Yahoo---agents間的合作)

Validation and Verification of the Knowledge Base(11. 15) • Evaluation assessing an ES’s overall Value ( usable, efficient and cost-effective) • Verification building the system right(系統是不是和自己規劃 的一樣) • Validation building the right system(系統是不是符合 real world ) (參照Criterion of Validation Table 11. 8)