pdf - 395 K - Ecole des mines de Nantes

More documents

Recommendations

Info

data like profiles is stored in a SQL database on demand andaccessed through ADO .Net.A. Learning Object Exploration SystemThe architecture is designed to be plug into layered overexisting repositories of Learning Objects. The system providesaccess to repositories through a web server. Users simplyexplore the repository with a web browser and view theLearning Objects as web pages.The system is linked to the architecture through output andinput points on the web server. The output point is a simpleserver log. For each access to a Learning Object, the log isclassically required to record a user identifier (a static IPaddress for instance), an identifier of the Learning Object andthe date and time.The input point consists in adding web content generatedby the architecture to the web page that is viewed by the user.B. The Agent ComponentsUser Agents ComponentA User Agent tracks each user of the system. This agentmay have references to Standard User Profiles (SUP) that suitits user. In addition, the User Agent is in charge of maintaininga personal user profile. This profile contains information aboutthe user’s pathways. The profile is refined over time as theUser Agent Proxy informs the User Agent of the user’sactivities. When a user logs off for a while, his dedicated UserAgent terminates. The profile persists independently from theUser Agent and it can be stored into a SQL database ondemand. It is reactivated when the user navigates again.Node Agents ComponentNode Agents form a virtual layer supplying information abouttraversals made between nodes. This information can then beexploited by others elements, such as the Business IntelligenceComponent. Node linkage information is formulated by thetraversal of users between nodes. The term “node” is a generalterminology used to represent a particular repository artifact.The artifact could be a web page or a relation in a relationaldatabase or it could even be more fine-grained such as a fieldin a relation. In the case study in this paper, a node issynonymous with a unique web page, thus a Node Agent willbe associated with a unique node. In the virtual network.Each Node Agent has the responsibility of:- Capturing a user’s page traversals (destinations). Forinstance a node agent (A) observes a web page (P). Auser (U) accessed page P and then from thereaccessed a new page (Q). The node agent A mustcapture the next web page traverse by the user P,which is page Q in this instance.- Storing the most recent timestamp for eachdestination navigated.- Keeping a count on each destination page accessedfrom the observed page.The Log Monitor gathers the information above is gatheredby analyzing web server log files. Each Node Agent registersits associated node identity with the Log. The Node is thennotified whenever a relevant log line appeared in the webserver log and ‘memorizes’ the information.C. Business Intelligence ComponentThis component is in charge of elaborating explorationassistance dynamically for every given user involved inexploring a Learning Repository. This is achieved bygenerating a directed graph representing navigation pathways.The vertices represent Learning Objects while the edgessymbolize navigational links. All the pathways have thecurrent Learning Object as origin. The edges hold informationabout the relevance of the traversal to the link they represent.Such information comes from 3 sources:(1) the Node Agents that give an indication ofthe general popularity of the link,(2) the SUPs associated with the user,representing users’ interest in somecategories and(3) the personal profile of a user maintained byhis User Agent.This information is computed in order to obtain relevancyindicators.The graph is then formatted to be visualized and sent to theweb server for being displayed to the user. A possibility is thatthe user sees an additional frame on top of the web page bythe mean of a plug-in for his web browser. The frame shows agraph whose nodes are actual web links that can be clicked.The graph provides navigation assistance in that the user isproposed relevant pathways. As an example, the graph couldbe something like this:In this example, “LOX” is the identifier of a Learning Object;the weight and the length of the edges indicate the generalMSc EMOOSE Thesis – http://www.emn.fr/EMOOSE 62
popularity of the link (the bigger the weight, the shorter thearrow) and the thickness of the arrows represents therelevancy of the traversal of the link for the user. This lastinformation comes from the computation of the user profile andthe Standard User Profiles (SUP) associated with the user.D. Standard User Profiles ComponentEvery Standard User Profile (SUP) defines a category ofusers. A SUP contains pathway information for exploringLearning Objects that are interesting for the category of users.Administrators of the repository initially define SUPs, eitherdirectly or by computing similar user profiles maintained by theUser Agents.When a new user enters the system, he has the option ofexplicitly selecting SUPs that suit him via a special web page.His User Agent then references the SUPs. Otherwise, a UserAgent is automatically created for him and initialized with noSUP. Later, the User Agent may infer a SUP after some time,thanks to the knowledge of the exploratory pathways of theuser.E. Exploration System InterfaceA Log Monitor is in charge of reading the server logperiodically. For every line in the log, the Log Monitor notifiesthe User Agent Proxy and the Node Agent Proxy and transmitsthe logged information (Fig.2).LogReadsExploration System InterfaceLog MonitorNotifiesUser AgentsProxyNotifiesNode AgentsProxyAccessesAccessesUser AgentsComponentFig. 2. The Exploration System Interface componentNode AgentsComponentThe proxies act as name servers for User Agents and NodeAgents. In other words, they know all the agents and theircorresponding identifiers, permitting for example access to theNode Agent that manages a given Learning Object. Besides,they transmit the log notifications from the Log Monitor to theagents that are concerned. Thus every User Agent is keptinformed of the exploration of the user it manages, and similarlyNode Agents keep aware of the navigation of all the users.When the User Agent Proxy does not recognize a user thatis referenced in the log, it means that it is a new user. Hencethe User Agent Proxy creates a new User Agent with no SUP.IV. RELATED WORKRecommender systems [1] learn about the preferences ofusers in order to assist them in finding items they areinterested in, like books or movies. These systems aim ataddressing the problem of information overload, particularly onthe Web and in e-commerce. They make use of user profilesthat are built either from explicit or implicit feedback from theuser. The recommendation mechanism is based on comparingitems (content-based), user profiles (collaborative filtering) orboth [2].Our system is comparable to recommender systems in thatrecommends pathways based on user profiles. Like manyrecommender systems it facilitates access to informationwithout requiring the user to formulate explicit queries.Additionally, it uses similar techniques such as passive userprofiling [3, 4] based on server logs [5] and relies on the notionof categories of users in the same manner as collaborativesystems.However, in addition our system provides assistance basedon 3 different dimensions:- the general community of users,- categories specific to the user- and individual user profiles.Reconnaissance agents like those of the MIT Media Lab [6]help users browse the Web. These interface agents are on theclient side to observe the user navigating and generateprofiles. When the user reaches a web page, they proposelinks for further navigation. For example, the well-known Letiziaexplores all the links on the page viewed by the user in order toeliminate irrelevant links, and then recommends the links thatfit best with the user profile.Likewise, our system provides navigation assistancewithout interfering with the normal browsing behavior of theuser. Also, this assistance changes according to the positionof the user in the navigational space. However, ourrecommendations are not only based on the profile of the userbut also on the experience of others.Also, our attempt to build semantic links can be compared tosystems with ontology-based semantics. The idea of theSemantic Web as proposed by Tim Berners-Lee and JimHendler is based on coded ontologies permitting softwareagents to “understand” the relationships between web pages[7].While we are not seeking to replace ontologies, indexes andother repository metadata, we are taking a diametricallyopposite approach to metadata tagging – that is, constructingMSc EMOOSE Thesis – http://www.emn.fr/EMOOSE 63
Page 1:
Vrije Universiteit Brussel - Belgiu
Page 4 and 5:
ContentsAbstract 1Acknowledgements
Page 6 and 7:
(4) Table UP_SUP (matches User Prof
Page 9 and 10:
content providers’ interpretation
Page 11 and 12:
- Active profiling consists in aski
Page 13 and 14: 2 Agents2.1 Overview2.1.1 Definitio
Page 15 and 16: (1) each agent has incomplete infor
Page 17 and 18: MASs need an organization for speci
Page 19 and 20: However, the usefulness of personal
Page 21 and 22: they often provide implementation o
Page 23 and 24: 2.5 Design2.5.1 The DIMA designIn [
Page 25 and 26: 3 MetadataDuring the earlier invest
Page 27 and 28: This requires some agreement and st
Page 29 and 30: 4 PerspectivesRecommender systems m
Page 31 and 32: way Lego blocks should be assembled
Page 33 and 34: usiness intelligence is generated.
Page 35 and 36: important than memory space optimiz
Page 37 and 38: The graph is then formatted to be v
Page 39 and 40: Chapter 4: Implementation1 Implemen
Page 41 and 42: (1) A Name Server (NS) maps every a
Page 43 and 44: the agent terminates, the thread is
Page 45 and 46: 4 Application level4.1 OverviewThe
Page 47 and 48: 4.3 User Agent MakerThis agent has
Page 49 and 50: Thus the profile, node and link cla
Page 51 and 52: BIProviderR : Provide_Business_Inte
Page 53 and 54: to enable users to explicitly speci
Page 55 and 56: ConclusionA “minimalist” approa
Page 57 and 58: Appendix A: Addendum onIndustrial I
Page 59 and 60: Appendix B: SQL Tables Design1 Tabl
Page 61 and 62: Appendix C: Conference PaperA paper
Page 63: very large repositories of learning
Page 67 and 68: Appendix D: Source Code1 Agent infr
Page 69 and 70: public static bool ProcessIsMainHos
Page 71 and 72: string path = MakePath(".", Middlew
Page 73 and 74: System.Messaging.MessageQueue queue
Page 75 and 76: public static NameServer GetNS(){if
Page 77 and 78: }}}table[agentId] = address;}}/// /
Page 79 and 80: ****** NON-STATIC PART ******// Tab
Page 82 and 83: }}}return (AgentId) first;using Sys
Page 85 and 86: {}return Subject.GetHashCode() + Na
Page 87 and 88: name in order to allow for the retr
Page 89 and 90: }// Start agent's main Thread.Threa
Page 91 and 92: }DirectoryFacilitator.GetDF().Unreg
Page 93 and 94: {/// /// Allows for specific conver
Page 95 and 96: }Monitor.Exit(aLock);public overrid
Page 97 and 98: Allows for synchronized Message exc
Page 99 and 100: esource./// /// public string GetMa
Page 101 and 102: }}private void RunTerminationExamin
Page 103 and 104: }private Message RequestUACreation(
Page 105 and 106: }msgQueue.AddFilters(new MessageCat
Page 107 and 108: Message received: interrupt preterm
Page 109 and 110: 19. Class Architecture.DBManagerAg
Page 111 and 112: }}connection.Close();/// /// For sa
Page 113 and 114: using System.Collections;namespace
Page 115 and 116:
namespace Architecture.Profiles{usi
Page 117 and 118:
private UP(DataRow row) : this(newU
Page 119 and 120:
}DataRow[] rows = Row.GetChildRows(
Page 121 and 122:
}adapter.Update(Data);//***********
Page 123 and 124:
internal static readonly string LAS
Page 125 and 126:
}}Row[OCCURRENCES] = Occurrences +
Page 127 and 128:
}get{}return Data.Rows.Find(_name);
Page 129 and 130:
private static SqlDataAdapter adapt
Page 131 and 132:
}foreach (ILink link in links){if (
Page 133 and 134:
}}}private DataRow Row{get{return D
Page 135 and 136:
18. N. Good, et al. Combining colla
Page 137 and 138:
49. D. Derbyshire, et al., Agent-Ba
Page 139:
83. Council for Higher Education Ac
show all

pdf - 395 K - Ecole des mines de Nantes

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?