7. User Modeling(UM, 1986-2007, 11th) California Adaptive Hypermedia and Adaptive Web-Based Systems(AH, 2000-2008, 5th) Italy UM+AH = UMAP(2009, 17th) UMAP(Adaption, Personalization)
8. Title : [1]Construction of Ontology-Based User Model for Web Personalization (Cited: 9 times) H. Zhang, Y. Song, and H.T. Song, “Construction of ontology-based user model for web personalization,” Proceeding of the 11th international conference User Modeling 2007, pp. 67–76. UM2007
9. Authors: Hui Zhang, Yu Song, and Han-tao Song Motivation: to provide web information that matches a user’s personal interests Purpose: Application: personalized web browsing and search [1]Construction of Ontology-Based User Model for Web Personalization
11. Steps: 1.S-Log(Semantic-log):representing the semantics of the respective URL(from domain ontology) 2.Session analysis algorithm outcome : semantic session include thematic categories 3. IS = user’s new session=outcome B(IS):user ontology(beginning of the visit is empty) S(IS):structure of the site(automatically built) 4.O = B(IS) U O (O:global user’s ontology) [1]How(cont.)
12. :look up :union :ontology [1]How-Imagination of Ontology Global User Ontology
13. Each user has a graph: C_Graph(N, u)=<N, A>, N: nodes, A:arcs, u:user arc(s, t)=>label(s, t) = <dst, rst, hst, Tst> dst: semantic independence coefficient rst: semantic relevance coefficient hst: hit coefficient Tst: time coefficient s,t : concept [1]Pre-defined
14. Duration: 1997-2011 Title [1]Personalized News Recommendation Based on Click Behavior (Cited: 2 times) J. Liu, P. Dolan, and E.R. Pedersen, “Personalized news recommendation based on click behavior,” Proceeding of the 14th international conference on Intelligent User Interfaces, 2010, pp. 31–40. IUI 2010
15. Authors: Jiahui Liu, Peter Dolan, ElinRonby Pedersen(Google Inc.) Motivation: people was burdened with large online information Purpose: to help users find the information that are interesting to read Application: Google News [1]Personalized News Recommendation Based on Click Behavior
16. Click behavior advantage no ratings or negative votes after experiment (picture) news interests do change over time click distributions reflect the news trend different news trends in different locations news interests ↔ news trend in location (a certain extent) [1]How
17. Prediction User’s genuine interests The influence of local news trend Flow predicting user’s genuine news interest from a specific time period t combining predictions of past time periods predicting user’s current news interest recommendation [1]How(cont.)
18. [1]How(cont.) : predicting user’s current news interest : current news trend : past time user’s news interest Nt : all user’s clicks times in t time period G : the number of virtual clicks(smoothing factor)
19. Recommendation: (to rank a list of candidate articles) CR(article): content-based recommendation score CF(article): collaborative filtering recommendation score [1]How(cont.)
22. A Survey on Service Personalization 學生:張維辰 指導教授:劉立頌 時間:2010/09/10
23. Service Personalization Early research Overview of user-profile-based personalization User Profile Purpose Type Process of user-profile-based personalization Outline S.Gauch, M.Speretta, A. Chandramouli, and A. Micarelli, “User Profiles for Personalized Information Access, ” The Adaptive Web, LNCS 4321, pp.54-89
26. Purpose To record interest or habit of the user To filter out irrelevant information from the user To identify additional information of likely interest for the user User Profile
27. Type Static ex: name, age, country, education level Dynamic short-term long-term User Profile
28. Process 1.Collecting information about users user identification user information collection explicit implicit 2.User Profile Representations 3.User Profile Construction User Profile
29. User identification Software agents Logins Enhanced proxy servers Cookies Session ids Collecting information about users
31. Explicit Providing personal information (My Yahoo![110]) Rating (Web pages, Syskill&Webert[68];Movie, NetFlix[62];Consumer, ePinions[24]) Implicit Browsing history (OBIWAN [71]) Browsing activity ([71], Trajkova[99], Barrett[6]) All user activity (Seruku[83], Surfsaver[94]…) Search (Miserach[87], Liu[45]) User information collection
36. More abstract topics (not specific words or sets of related words) Concept profiles
37. User Profile Construction Building keyword profiles Building semantic network profiles Building concept profiles Thank you for attendance! Coming soon…
38.
39. A Survey on Service Personalization 學生:張維辰 指導教授:劉立頌 時間:2010/09/17
40. GediminasAdomavicius , Alexander Tuzhilin, Using Data Mining Methods to Build Customer Profiles, Computer, v.34 n.2, p.74-82, February 2001 (Journal) Building Customer Profiles by data mining methods
46. Key V=(W1, W2, W3, …, Wn) (待修改)Amalthaea’s Ecosystem[61](cont.) Web Pages Stemmer Html2txt filter Removal(commonly used) Html2url filter Hc x TF x IDF Moukas, A.: Amalthaea: Information Discovery And Filtering Using A Multi-agent Evolving Ecosystem. In: Applied Artificial Intelligence 11(5) (1997) 437-457 (Journal, Publisher : Taylor & Francis)
47. WebMate: A personal agent[13] Chen, L., Sycara, K.: A Personal Agent for Browsing and Searching. In: Proceedings of the 2nd International Conference on Autonomous Agents, Minneapolis/St. Paul, May 9-13, (1998) 132-139
48. Definition: 1. Profile set V = { V1, V2,…,VN} (N domains of interest for each user) 2. Document Di -> Vector Vi, i={1,…N} Vi={ e1,e2,…,eM}, ej =TF(wj, Di) x IDF(wj), j={1,…,M} WebMate[13](cont.)
49. Algorithm for multi TF-IDF vector learning: (待修改)WebMate[13](cont.) User marked “I like It” If |V| < N Add in set V T Parse HTML page F Compare every two vectors by (a) Extract TF-IDF vector Combine Vp, Vq with most similarity Vp = Vp + Vq Sort (a)
50. Widyantoro, D.H., Yin, J., El Nasr, M., Yang, L., Zacchi, A., Yen, J.: Alipes: A Swift Messenger In Cyberspace. In: Proc. 1999 AAAI Spring Symposium Workshop on Intelligent Agents in Cyberspace, Stanford, March 22-24 (1999)62-67 Alipes[103] Control
53. A Survey on Service Personalization 學生:張維辰 指導教授:劉立頌 時間:2010/10/22
54. Authors Susan Gauch, Jason Chaffee and Alexander Pretschner Motivation It’s impossible to use one approach to browsing or searching for every user according to preference. Purpose Personalized web browsing and search Application Web sites Ontology-based personalized search and browsing (Cited: 194 times)
55. Reference ontology: Concept, Source Concept To extract top levels of the subject hierarchies (already existing) Source associated web pages from Yahoo, Magellan, Lycos, and the Open Directory Project How-Browsing
64. A Survey on Text Categorization 學生:張維辰 指導教授:劉立頌 時間:2010/11/02
65. Classification supervised learning pre-defined categories ex. credit of consumer Clustering unsupervised learning unknown categories ex. similarity of consumer Preliminary
66. Motivation With the rapid growth of online information, it is difficult and time-consuming to deal with or classify the information by hand. Purpose To manage and use information easily Application Filter(personal portal site, email) Portal site Semantic identifier Image classification multimedia document classification Text Categorization(TC)
67. SVM(Support Vector Machine) Vapnik 1995 kNN(k-nearest neighbor) NB(Naïve Bayes) LLSF(Linear Least Squares Fit) NNet(Neural network) Approaches of TC