1. + About Knowledge September 2014

2. + Teaching Staff Lecturer: Qinmin Hu Email address: qmhu@cs.ecnu.edu.cnqmhu@cs.ecnu.edu.cn Office hour: appointments needed Teaching assistant:

3. + Outline Example systems Knowledge representation and reasoning Information retrieval Recommendation systems Goals / designs of the course

4. + Example Systems Google / Baidu Siri Wolfram alpha Wikipedia Yelp / Dianping

5. + Google From words to real world Real word connection by building knowledge graph Knowledge graph http://www.google.com/insidesearch/features/search/knowledg e.html http://www.google.com/insidesearch/features/search/knowledg e.html

6. + Cognitive Assistant SIRI Demo

7. + Wolfram Alpha Demo http://www.wolframalpha.com/input/?i=10+peanut+M%26Ms

8. + Knowledge Representation and Reasoning

9. + Spotlights For each system, we will look at What knowledge must it represent? What reasoning must it do? What would it take to extend it? Where does it fail? How is it different from (current) Google?

10. + Google From words to real world Real word connection by building knowledge graph Knowledge graph http://www.google.com/insidesearch/features/search/knowledg e.html http://www.google.com/insidesearch/features/search/knowledg e.html

11. + Google (2) What knowledge must it represent? Objects such as person, book or movie What reasoning must it do? Relationships among the objects What would it take to extend it? Add more relative data on the objects, structured and aware data, clear relationship definitions of the objects Where does it fail? No connections to an object How is it different from Google? Made by Google, but different to Google search

12. + Cognitive Assistant SIRI Demo video: What knowledge must it represent? Restaurants, movies, events, reviews, … Location, tasks, web sources, … What reasoning must it do? Nearest location, date for tomorrow, AM vs PM, etc What would it take to extend it? More sources, different sources Where does it fail? Completely different environment, completely different task Differences from Google Dialog driven, task-oriented, location aware, …

13. + Wolfram Alpha (1) Demo http://www.wolframalpha.com/input/?i=10+peanut+M%26Ms

14. + Wolfram Alpha (2) What knowledge must it represent? Different kinds of foods, their nutrition composition, caloric values What reasoning must it do? Mathematical computations based on portions What would it take to extend it? Add more data on foods and nutrition composition Where does it fail? Does not know about recipes, how to combine foods, … How is it different from Google? Data driven as opposed to document driven, mathematical reasoning

15. + Baidu Discussions

16. + Baidu (2) What knowledge must it represent? What reasoning must it do? What would it take to extend it? Where does it fail? How is it different from Google?

17. + Wikipedia

18. + Wikipedia (2) What knowledge must it represent? What reasoning must it do? What would it take to extend it? Where does it fail? How is it different from Google?

19. + Dianping

20. + Dianping (2) What knowledge must it represent? What reasoning must it do? What would it take to extend it? Where does it fail? How is it different from Google?

21. + What is Knowledge? (1) John knows that Mary will come to the party. John hopes that Mary will come to the party. John fears that Mary will not come to the party. John believes that Mary will come to the party. John knows Mary well. What are the differences among the above sentences?

22. + What is Knowledge? (2) Observe we say “John knows that...,” we fill in the blank with a simple declarative sentence. “John knows that Mary will come to the party” Knowledge is a relation between a knower and a proposition (the idea expressed by a simple declarative sentence) The knower like John, and the proposition like “Mary will come to the party”. Propositional attitudes Verbs like “knows,” “hopes,” “regrets,” “fears,” and “doubts” “John hopes that Mary will come to the party.” Same agent and proposition, but the relationship is different When propositions are implicitly mentioned: “John knows Bill well” Nothing further to say, when no useful proposition is involved

23. + What is Knowledge? (3) They all share with knowledge a very basic idea: John takes the world to be one way and not another.

24. + What is Representation?

25. + What is Reasoning?

26. + Benefits of Explicit Representation We can add new tasks and easily make them depend on previous knowledge Enumerating objects vs painting objects Extend the existing behavior by adding new beliefs Assert that canaries are yellow Debug faulty behavior by locating the erroneous beliefs By changing the color of sky we change any routine that uses that information Explain and Justify the behavior of the system The program did X because Y

27. + Benefits of Reasoning Given Patient X allergic to medication M Anyone allergic to medication M is also allergic to medication M’ Reasoning helps us derive Patient X is allergic to medication M’

28. + Relationship to Artificial Intelligence KR&R started as a field in the context of AI research Need explicitly represented knowledge to achieve intelligent behavior Expert systems, language understanding, … Many of the AI problems today heavily rely on statistical representation and Reasoning Speech understanding, vision, machine learning, natural language processing For example, the recent Watson system relies on statistical methods but also uses some symbolic representation and reasoning Some AI problems require symbolic representation and reasoning Explanation, story generation Planning, diagnosis Abstraction, reformulation, approximation Analogical reasoning KR&R today has many applications outside AI Bio-medicine, Engineering, Business and commerce, Databases, Software engineering, Education

29. + Information Retrieval/Search

30. + Spotlights For each system, we will look at Data preprocessing Information extraction / knowledge extraction Index / knowledge storage

31. + Google From words to real world Real word connection by building knowledge graph Knowledge graph http://www.google.com/insidesearch/features/search/knowledg e.html http://www.google.com/insidesearch/features/search/knowledg e.html

32. + Cognitive Assistant SIRI Demo

33. + Wolfram Alpha Demo http://www.wolframalpha.com/input/?i=10+peanut+M%26Ms

34. +

35. + Unstructured (text) vs. structured (database) data in the mid-nineties 35

36. + Unstructured (text) vs. structured (database) data today 36

37. + What is the IR Problem? First applications: in libraries (1950s) ISBN: 0-201-12227-8 Author: Salton, Gerard Title: Automatic text processing: the transformation, analysis, and retrieval of information by computer Editor: Addison-Wesley Date: 1989 Content: external attributes and internal attribute (content) Search by external attributes = Search in DB IR: search by content

38. + Recommendation Systems

39. + Spotlights For each system, we will look at Data preprocessing Information extraction / knowledge extraction Index / knowledge storage

40. + Google From words to real world Real word connection by building knowledge graph Knowledge graph http://www.google.com/insidesearch/features/search/knowledg e.html http://www.google.com/insidesearch/features/search/knowledg e.html

41. + Cognitive Assistant SIRI Demo

42. + Wolfram Alpha Demo http://www.wolframalpha.com/input/?i=10+peanut+M%26Ms

43. + Google

44. + 44 What is Recommendation systems? Items SearchRecommendations Products, web sites, blogs, news items, … Recommendation systems are programs which attempt to predict items that a user may be interested in There are two basic approaches to recommending:  Collaborative Filtering (a.k.a. social filtering)  Content-based

45. + 45 Collaborative Filtering A 9 B 3 C : Z 5 A B C 9 : Z 10 A 5 B 3 C : Z 7 A B C 8 : Z A 6 B 4 C : Z A 10 B 4 C 8. Z 1 User Database Active User Correlation Match A 9 B 3 C. Z 5 A 9 B 3 C : Z 5 A 10 B 4 C 8. Z 1 Extract Recommendations C

46. + 46 Collaborative Filtering Maintain a database of many users’ ratings of a variety of items. For a given user, find other similar users whose ratings strongly correlate with the current user. Recommend items rated highly by these similar users, but not rated by the current user. Almost all existing commercial recommenders use this approach (e.g. Amazon).

47. + 47 Problems with Collaborative Filtering Cold Start: There needs to be enough other users already in the system to find a match. Sparsity: If there are many items to be recommended, even if there are many users, the user/ratings matrix is sparse, and it is hard to find users that have rated the same items. First Rater: Cannot recommend an item that has not been previously rated. New items Esoteric items Popularity Bias: Cannot recommend items to someone with unique tastes. Tends to recommend popular items.

48. + 48 Content-Based Recommending Recommendations are based on information on the content of items rather than on other users’ opinions. Uses a machine learning algorithm to induce a profile of the users preferences from examples based on a featural description of content. Some previous applications: Newsweeder (Lang, 1995) Syskill and Webert (Pazzani et al., 1996)

49. + 49 Advantages of Content-Based Approach No need for data on other users. No cold-start or sparsity problems. Able to recommend to users with unique tastes. Able to recommend new and unpopular items No first-rater problem. Can provide explanations of recommended items by listing content-features that caused an item to be recommended.

50. + 50 Disadvantages of Content-Based Method Requires content that can be encoded as meaningful features. Users’ tastes must be represented as a learnable function of these content features. Unable to exploit quality judgments of other users. Unless these are somehow included in the content features.

51. + Relationships among Knowledge, Reasoning, Retrieval and Recommendations Discussions

52. + Outline Example systems Knowledge representation and reasoning Information retrieval Recommendation systems Goals / designs of the course

53. + Goals of This Course Introduction to techniques used to represent knowledge Introduction to information retrieval Introduction to recommendation system Applications among knowledge representation, information retrieval and recommendation system

54. + Design of the Course(1) Textbooks Knowledge Representation & Reasoning by Brachman & Levesque Information Retrieval by C. J. van RIJSBERGEN Recommendation System and It’s Applications by Liang Xiang Reference papers Be listed after each class Lectures Tuesdays, 10:00 – 11:30 am Grades Paper presentations (20%) + KBA competition (20%) + project (50%) + attendance (10%)

55. + Design of the Course(2) Course website To be announced Topics KBA competition (data processing, tasks) (Knowledge) Object-oriented representation, ontologies (Search engine) Information systems, Information techniques Recommendation systems, recommendation classical algorithms Introduction to probabilistic Graphical Models Tests N/A Project To be announced

56. + Project 1 The TREC Knowledge Based Acceleration Track 2014 20% grades

57. + Project 2 Knowledge Management System for Shanghai Science and Technology Commission 50% grades