<< Back
.: Dissertation Area: System Proposal
Get this document in Word format
here
.
The following is a proposal for the Apellicon System. Apellicon: A Proposal for a Semi Autonomous Web Tool for Student Research Carl Nattrass University of Durham Department of Computer Science info@apellicon.co.uk Abstract The expansion of the Internet is making it increasingly difficult for users to locate the information which they are seeking. Search engines have made advances with technology to partially address this issue but still fail to provide a useable querying system which enables a ‘single strike’ access to the sought information. Students and researchers face this problem on a daily basis; Google estimate the web to be in the region on 11.5 billion pages, and IngentaConnect have over 22 million articles. Researchers spend a good proportion of their time browsing through ‘nearly relevant’ articles. This paper proposes a system for Students and Researchers which will assist them in locating the key articles within their field. Keywords Collaborative, Web, Research, Assistant 1 Introduction There will be very few Researchers and Students who would admit they enjoy the process of weeding out those relevant documents they need to complete a dissertation or report. Many students, especially those in high school would prefer to be pointed in the right direction when it comes to researching school projects. Unfortunately, the currently available search engines are still inadequate in homing in on the desired information. There are a number of reasons for this including: 1 Incorrect search criteria given to search engine 2 Search engine out of date. 3 Documents not indexed correctly or wrongly described. 4 Documents not easily accessible by search engine crawlers. Traditionally, a student would learn by approaching a Tutor for information and through a question and answer session, the information would be arrived at. This is similar in concept to a SE where the user refines the search criteria until the correct results are returned. In an ideal world, the user should only need to type in a question and receive the answer, but this is in most cases above the capacity of a computer. An alternative arrangement would be for the question to be forwarded to someone in authority on the subject, for it to be answered. Unfortunately, apart from in a much specialised subject area, this would not happen. This paper proposes a system call Apellicon which keeps this approach in mind and has been created as a basis for further development. To visualise the aim of Apellicon, we have created a target scenario: Student A is 21 years old, lives in Tokyo, Japan, and is studying for a degree titled ‘Metrics within Business Process Engineering’. As part of her research she spends quite a large amount of time surfing the web looking for relevant research documents. At some point, she finds a research paper which includes figures and graphs from a study an organisation has made into several newly proposed metrics. Finding this interesting, she saves a copy to her computer. A week later, Student B, 22, who live in London, England who is also studying for a degree in Business Process Engineering ventures onto the Internet looking for ideas for a section on Metrics for his dissertation. He enters his keywords into a standard web search engine and finds at the top of the returned list, a reference to the document that Student A found useful. Student C, a 16 year old from Raleigh, NC, USA is doing a project on Computer System Design and as part of it, needs to give a brief explanation of what Business Process Engineering is. He uses the web to search for an article, and the returned references do not include the article both Students A and B found useful. He finds what he needs through a normal query process. The key concept is that if Student A finds an article of interest whilst surfing for her key subject, then Student B should also find that article being recommended to him by any search engine he cares to use, so long as his submitted keywords relate to the article. We would not want student B to be looking for video listing and being recommended research articles as a result. The system must be able to differentiate between when a student is studying and when he/she is not. Student C will never be offered a reference to that particular article from a search engine unless by luck of query keyword. We should also examine the purpose of providing assistance to students and researchers; there are a number of scenarios: 1) To make the task of research easier by enabling them to home in on relevant information more quickly? 2) To make the task of research easier by enabling them to home in on known or recommended sources of relevant information more quickly? This implies some sort of recommending system be in place. 3) To assist them in locating new information, not necessarily widely known about within their peer groups. Certain technologies are more suitable to some scenarios than to others. 2 Related Work Much work has gone in to the extraction of information from the web. There are hosts of Agents, Search Engines, Assistants, Collaborative Projects, News Groups, Software Systems, Repositories, Blogs etc. which all achieve a partial amount of success. Most of these concepts are used in conjunction with a form of user profiling. In most cases, the process needs to know what the user is seeking, or at least the field in which it lies. Information held on the Internet must be parsed in some way for it to be categorised. All information is categorised to a degree, and it is this categorising or filtering that these processes are concerned with. The information may be parsed from two locations: Server-side processes and Client-side processes. Server-side Processes Server-side processes (SSP) are those which use the processing power of a web server to locate the desired web pages. In most of the following cases, there will be some form of SSP, be before or after another process has taken place. The most common SSP is the utilisation of a Search Engine (SE). An SE is given a set of query criteria, and checks them against its index of keywords. It then returns the most closely matching pages to the user. Some Servers may do additional processing to the query before submitting it to the indexing mechanism, to help achieve a closer match. They may also do some processing in relation to the returned pages; possibly filtering them to more closely match a user’s profile. In more specialised SSP, a proxy server may sit between the user and the web server. This enables queries to be altered before going to the SE on the web server and to alter the returned results. Client-side Processes Client-side processes (CSP) relate to any software in operation on the user’s PC, normally only concerning Web related systems. CSS are ideally suited to filtering or augmenting queries prior to them being sent to SE’s. Like SSP’s they can also filter results from SE’s before the user can view them. The benefit to CSS is that it uses the user’s PC’s processing power instead of the Server’s processing power. It also has the benefit of not needing to inform the Server of any information about the user with regard to his or her profile or needs. The most popular CSP in this sphere is the use of Agents. One example of a CS assistant is Yarrow [Chen00], a meta-search system which learns from the user with very little feedback. It is capable of augmenting queries before sending them to up to 8 of the popular search engines. Its interface is through a web browse. 2.1 Agents Agents are typically used for CSP. Agents are pieces of software which run on a user’s PC. Franklin and Graesser [Franklin96] offered a good definition of an agent; ‘An autonomous agent is a system situated within a part of an environment that senses that environment and acts on it, in pursuit of its own agenda and so as to affect what is senses in the future.’ Therefore an Agent may be described as a system which has the capacity to: 1) Sense it environment 2) Act upon its environment with a view to altering it. 3) Learn. 4) Own an agenda. Agents can be classified by: 1) Scope: Either locally deployed agent on a user’s computer, restricted deployment on an organisations network, or globally deployed on the Internet. 2) Intelligence: Either those that ‘look over the user’s shoulder’ and learn, those that are given explicit parameters, and those that work in an unsupervised way. 3) Use: Either personal assistants, group assistants or service agents which may work for a community. Because agents are constantly running as the user browses, they have a myriad of uses. One of the most famous Agents called Letizia was created by Lieberman [Lieberman95] and uses the user’s profile to make suggestions about pages in the neighbourhood of the page the user is currently viewing. They can also filter and augment the results from search engines prior to the user viewing them as in the case of WebWatcher [Frietag97]. Agents may be seen as a vehicle on which processes can be organised and implemented. Many of today’s web tools rely on an Agent or Agents to work. By implementing an Agent, many methods can be used to assist a user in surfing the web. Following is a list of the more widely known ones. 2.2 Processing Technologies There are many methods of pre-process and post-processing a data before and after it gets to the Internet. Following is a discussion of the most popular, interesting and the most promising of those methods. Repeated Patterns/Paths Where a user regularly follows a set path to locate a page on a website, a direct link on the main web page can be added which will link to the sought page. For example, suppose a user regularly uses the Java.com main page to browse to the ‘Programming’ page and then onto the ‘Java Manual’ page, and from there to the section on ‘Threads’. An agent would place a link on Java’s main page which would directly link to the ‘Threads’ section. This was proposed by Maglio and Barrett [Maglio96] who observed how users of the Internet spent large proportions of their time repeating processes. Recording Waypoints Maglio also showed that users find the ‘History’ feature in Web Browsers as confusing and tend to go back to major waypoints to find a recently visited site. By recording paths, a graphical notation might be used to illustrate where the user has been. The user could then select from the nodes in the notation to relocate the web page. Trodden Paths A similar concept to the above Repeated Patterns/Paths is the ‘path’ concept, where a well trodden path, in this case, URL’s, between documents can be identified, a link may be established between the contents of the two documents, i.e. there is a good chance that their contents are related. This could be used in an environment such as CiteSeer to link documents of similar subjects. This concept was suggested as being promising by Dean and Henzinger [Dean99] who attempted to establish relationships between documents. They noted that weightings between Nodes (URLs) and edges (links) could be used to form a semantic relationship between documents and web pages. One problem they encountered was that topics dominated by commercial sites are not collaborative in their linking and so key research might be missed through this. A very similar concept was proposed by Jaczynski et al. [Jackz97] with a system called Broadway where comparisons are made between similar user’s URL history. The agent attempts to pre-empt where a user is heading based on comparing the tow set of URL’s. This would only work when the same sites have been visited by both users. Back Links This is a concept which has not yet successfully been implemented in a broad context of the WWW, but Web scripting languages such as Xanadu [Nelson94] show great potential. Consider the scenario where user A follow links from page p1 to page p2 to page p3 to gather relevant information upon a subject. A second user, B initially views p3 from a SE result, the Agent then suggests that p1 and p2 may be of use. It is easy for user comprehend the forward linking progress of user A, but not so the backward linking process of user B because the links are not actually their, they have come from an indexed repository which has previously stored them. Local Page Scouting Letizia, mentioned previously, and proposed by Lieberman is an Agent which uses a user profile to make suggestions about other Web pages in the neighbourhood of the Web page which the user is currently viewing. Using the Best-First search, the agent displays a list in a pane in the Browser with a set of suggested Web site which are nearby to the current Web Page. Letizia ‘learns’ if the proposed links are not of use by monitoring the user’s actions. Web Page Link Weighting Terveen and Hill [Terveen98] considered the idea that if an author of a web page included links to other web pages, he/she must consider the other web page to be of relevance. This was seen as a form of endorsement of the relationship between the two articles. This is a viable idea which could be capitalised on within the boundaries of concept. The problem with this is that if the data we are looking for is not linked to it will not be found. Terveen states that ‘On average, 43% of web sites are isolated”; this means that using this method would not access any of these 43% of web sites. The concept will return high quality knowledge in the immediate sphere of interest, but could suffer from being out of date and missing key concept not covered by those authors. Document Examination Where part of an Agent’s process is to examine a given document to extract some identifying information, the document being proposed or offered to the user must be parsed, this can be done by various methods. Document Examination Agents (DEAs) may look for keywords, Authors, References, headings, images, captions, hyperlinks and alternative text to name but a few. The extracted information is normally indexed and will be used later, for example during queries to augment a search. WebSeer, a Search Engine proposed by Frankel et al [FRANKEL96] uses this method to index documents and images. Personal Web Neighbourhood A Personal Web Neighbourhood (PWN) is virtual zone around where the user is currently browsing. The Agent may look ahead up to 3 links from the current page, and in some case, list them, possibly in what is thinks as priority order in the browser. This is what is known as a PWN. The PWN may be held locally and might be accessible by peers who have the same interests. The process normally utilised some form of User Profiling. Link Personalisation This is where the links available in a web page change according to the users profile/interests. Amazon, www.amazon.co.uk, uses this method to suggest suitable books to a user. This is done by establishing a profile based on the books viewed and bought by the user, and then cross matching the profile with other user’s profiles that have shown with similar interests. Amazon would then suggest the books which the other users considered. Web Page Annotation Fensel et al. [Fensel98] proposed Ontobroker, an HTML extension which could be used to store additional information about the Web Page, its author and the subject covered. The annotations would not affect the view of the Web Page, only the source text which would hold extra searchable information. Ontobroker also supported tools such as a Query Interface to assist the searching of the web for these ontological extensions. The benefits of such extra fields which can be searched are unquestionable, but as Fensel points out, there may be difficulty in getting the extension to the HTML language supported due to the extra work involved. This technology was suggested prior, by Luke et al. [Luke97] in their HTML extension SHOE. Luke pointed out the benefits to ontological knowledge and acknowledged the lack of research done previously in this sphere. Display Content Personalisation This is similar in technique to Link Personalisation but is taken a further step. Instead of the user only being offered a list of promising links, parts or the whole of the web page are personalised to suit the user’s profile. An example of this is WebWatcher [Armstrong97] which adds graphics such as the available command options e.g. search, and highlighted links together with graphics (a pair of eyes) to the web page view. Another Agent which uses this method is called Syskill & Webert [Pazzani98] which learns from user feedback to create a separate profile for each topic the user is interested in. The profile is then used when a search is done against a SE such as Lycos. The resulting view of the returned links is altered to display ‘thumbs-up’ and ‘thumb-down’ to highlight those links which the Agent thought would be more suitable than others. One of the benefits of this method is that it allows the Agent to converse with the user in a discreet manner. Luke et al. [Luke97] proposes an extension to HTML which would allow authors to augment the ontological information held within a web page. An associated crawler was also developed to locate and index this information. Although in most cases, the Web Page content would not be changed the semantic information of the page would be enriched. This would allow a standard search on then keyword ‘Java’ to be augmented with further refining keywords such as ‘PhD research’ or ‘University lecturer’ to better pinpoint the level of information sought. Recommender Systems Recommender Systems (RS) is an umbrella for a group of methods which attempt to make suitable recommendations to the user within the user’s interest area. All RS rely on the system knowing some form of information about the user, without this, they would be hard pressed to be able to make recommendations. In most cases, the system relies on User Profiling (see below). Recommendation by Collaboration: Collaborative Filtering (FC) works by matching users based on their profiles and grouping them into neighbourhoods. This neighbourhood is used to make suggestions to users as to the suitability of papers/docs/web pages. The media is rated in some cases by certain key users, or in other cases, any users. The relationships between the user’s profiles is an approach which matches real-life, and can take into account professional, political, personal and social. This method has been studied in depth by a number of researchers; one such notable is Scott [Scott00] who states that one of the difficulties to overcome is than often, individual’s information needs and tastes differ making recommending difficult. Recommendation by Content Filtering: Content Filtering Recommendations (CF) are made by comparing the content of a webpage with the user’s profile. Users may clustered in to probable classes or profiles [Surflen 57] very similar to neighbourhoods. Depending upon their class, comparisons are made with similarly matching available web pages. Text documents are recommended based on a comparison between their content and the user profile’s key words. Data structures are used alongside word weighting techniques such as TF*IDF to rank documents. One of the downsides is due to the ability for only a shallow analysis of the data, especially movies and pictures. Recommendation by Semantic Based Web Page Filtering: Similar to CF in its concept of comparing the user’s profile to Web pages. In this case, the profile is being compared with the semantic details of Web Pages. Semantic Web (SW) details are those which not necessarily relate to the content of a web page such as links, file names and alt text. More expansive Web scripting languages are being used such as OWL [W3C07] which will enable machines to make more sense of the Web, with the result of making the Web more useful for humans. The benefit of the SW is in the searching of web pages; queries against search engines can be much more clearly defined by using this extra info found in the Web page. Digital Libraries/Repositories & Archiving Systems The easiest solution to the problem of not being able to locate information on the Web is that all existing documents and all future created documents are placed onto a central repository. There are a number of problems with this: 1) Needing incentives so that authors readily contact the repository to notify it of new papers. 2) Copyright infringement issues. 3) Knowledge hiding – some papers may contain knowledge which the author may not want in the public domain. 4) Storage and saleability issues. Cameron [Camer97] in his study on citation databases, pointed out that it would ensure that publications in any form are visible, but not necessarily accessible, and it would allow fairer competition between publication venues. There are a number of useful repositories currently in existence, two of which are; 1) The Berkeley Personal Libraries provide cataloguing and full search capabilities and offers a privacy mechanism. This can be accessed at http://elib.cs.berkeley.edu/pl/. 2) CiteSeer. This does not require the author to actively to list his paper, this is done automatically by the CiteSeer agents. It also provides an API to utilise the database capabilities. CiteSeer an be accessed at http://citeseer.ist.psu.edu/. Sources for indexing literature face three main costs: 1) Entering the data about the documents onto a database; purchase of the document, preparing and key entering data. 2) The addition of the extra data associated with the indexing of the document which would be used to aid search and evaluation. 3) Marketing and distribution. User Profiling A user’s profile may be established by a number of methods, one of the simpler methods is demonstrated by Outride [Pitkow02], which makes suggestions of suitable web pages based on the user’s bookmarks, browsing history, click trails and time spent looking at documents. Dynamic Query Augmentation Dynamic Query Augmentation (DQM) is not in itself a single method, but a concept which may be used in many of the above technologies. It is almost always related to a Search Engine. Where a user has entered a query to be submitted to an SE, the Agent uses its knowledge, however gained, to alter the query in an attempt at getting a better result form the SE. Powerscout [Lieb01] uses DQA to make suggestions of suitable web pages based on user profile and viewed web pages. A more direct example of this came in the form of GOOSE [Lien02], a search engine which augments queries by using common sense statements provided by the Open Mind database. 3 Discussion The selection of methods listed illustrates the expanse of the technologies available to assist researchers in locating information on the Web. However, many of the technologies limit their usefulness by specialising in only one particular paradigm from all those available. It is possible that the reason behind this, as was pointed out by Lieberman that for an Agent to be successful, it must concentrate its resources on a specific task. With reference to what identifies an Agent as an Agent, we see that one must own an agenda. It is proposed here that the level of success an Agent achieves is directly related to its level of atomicity; just as an object in a program is self-contained and should be concerned only with its own goal. Etzioni [Etzioni96] suggested that the creation of an Agent should start with a universal SoftBot which should then be refined to do specialist jobs. Does this mean that in the future, the average computer must run dozens of different Agents in order to make easier the task of accessing the Web? Not necessarily, it is our opinion that Agents and Agent technology will remain an area mainly used by specialists such as researchers. For general use, it is the Domains which need to be classified, not the individual documents. Better classification of Domains and their contents, possibly by using semantic technology would enable a far more indexable Internet that we presently have. Technologies such SHOE and Ontobroker offer a more formalised Web with better searching facilities, but with the expansion of the Internet continuing at its enormous pace, it is likely that the indexation of the entire web is becomes less viable as time goes on. The bigger it gets, the more time it takes to index and the more technology is needed. The future of searching the Internet lies with the formal classification of pages and Domains to central repositories on an ongoing basis. Where does that leave our Student B? He is still searching for information on metrics, will his research ever be mad any easier? With the creation of repositories such as CiteSeer, a good start has been made. Unfortunately, due to their successes, and size, these repositories are becoming increasingly as difficult to search as the Internet. The key to a successful search is for the SE to know more information about what the user is searching for. Is it viable that every user creates his/her own profile for this purpose? With a research student, this is entirely possible, but for a general user who may move from one subject to another within a short space of time, he/she would need numerous profiles and this creates difficulty in the Agent assessing which profile to use. If student B had a profile containing his age group, level of study and specific area of study, would this help him in today’s Internet? Probably not, but if we could match his profile with student A’s profile, we might expect them to be interested in similar web pages. If student A found a document on CiteSeer interesting, student B has a good chance of also finding that document of interest. If we can categorise students accordingly, and learn what users in each category find interesting, we are in a position to give reliable recommendations. One such system which utilises this type of approach is PolyLens [Cosley01] which enhanced the MovieLens system through profiling groups of users, in order to make recommendations to members. One of the limitations of this type of system, as was pointed out by Balabonovic and Shoham [Balab97], is the initial problem having no recommendations on the system, and so nobody uses it. They suggested that initially, recommendations could be given based on using the profile against gathered documents and against SE’s. This is a form of DQM as previously mentioned. 3 The Proposed System We proposed a system based on a Proxy Server and Agent architecture. Processing would be done both on the Proxy side and on the Client side. Users would need to create a ‘profile of study’ which the system would use for key information. The users have the option of remaining anonymous, in which case, only his/her course and level would be needed for the system to operate. Alternatively, they may optionally offer further information in order that they may be identified within the system; this would be useful for Teachers who wish for their input to be visible. Users would be placed into profile groupings, all having the same research subject and study level. It would be possible a School workgroup to set up their own profile for exclusive use throughout a project. A user’s profile would be a member of 1 of 3 grouping domains; 1) A loose grouping of students in similar age-groups studying similar topics such as a 12 year old student doing his history homework on The Great War. In this case, he may find most related articles interesting. 2) A slightly finer grouping where a 16 year old student is studying blood vessels in mammals. He/she would wish to filter out most blood related information not pertaining to mammals. 3) A fine grain search group where a MSc student is researching the effects of ambient temperatures on blood cell degradation. In this case, the student will already know quite a lot about blood cells and may wish to only access specific articles recommended by peers. The user would activate the system from an entry in a Browser menu. The user may have one or more profiles to match the current research activity. The ‘processing’ by the system may be activated or deactivated at any time. Browsing The browsing system would operate in the following manner. Each web-page would have a small pane at the top of the web page showing the system options; we expect to see at least a thumbs-up icon and a thumbs-down icon used for recommending pages. This would be added to the web page view by the Proxy Server. As the user browses, the local Agent will maintain a search and parsation of the immediate browsing environment (BE). The BE would extend to two or three links distant from the current page. The level of the success of the process would depend upon: 1) The time the user spends one a particular page; If the user is quickly flicking through pages, the system would make no attempt for a BE parsation. 2) The number of pages within the BE; if there a many links on a document, the parsation may only extend to one link distant. If there are only a few links, the parsation may extend to its full 3 link distance. 3) The number of pages parsed also depends upon the processor usage at that time on that computer. The Agent must maintain its transparency and must not cause a slowing of the computer’s responses to user requests. The system would offer to the user lists of items of interest such as: 1) A suggested top 10 of the local pages which most accurately match the user profile. 2) A list all downloadable documents such as PDF’s and Word documents within the BE. 3) A list of all other media typed such as pictures and audio files within the BE. Any of the items in the list may be accessed by selection it Searching The user should be allowed to submit a search to any SE. As the search is submitted, the Proxy Server will undertake several processing tasks: 1) Search its own records for recommendations matching the user’s profile and keywords. 2) Attempt to augment the search keywords with known relevant keywords for that profile. 3) Forward the search criteria to the chosen SE. 4) Append the SE results to those found in its own records. These will be added transparently to the search result, normally to the top of the list. For more advanced levels of profiles such as degree courses, the agent could send an additional criteria search to a known repository such as CiteSeer. Depending upon the users preferences, the search could encompass references and citations to matching documents. As these document collections are viewed, it is hoped that user would make recommendations and thus move individual documents up the recommendation list. Recommendations Article and web pages are recommended by the user clicking the thumbs-up or thumb-down icon in the top frame of the web page. These recommendations are recorded by the Proxy Server, and are added to the user’s profile group. The presentation of recommendations would be available to all profile group members through either: 1 Augmented query results: the list of query results brought back from a SE might have a ‘thumbs up’ or ‘Thumbs-down’ icon added to line of those articles which are within the user’s profile. This graphic will be added by the Proxy Server as the data is returned from the SE. There may also be a set of ‘Top-5’ articles added at the top of the resulting list. 2 Recommendation made would always be available in a sub window of the Browser which would be kept up to date in real time. This would allow the user to make a selection form this list. Monitoring of Recommendations Where users have logged in non-anonymously, their recommendations could be monitored. Users may have a value associated against the reliability of their recommendations. In this way, users may have ‘favourite’ peers who they trust for good recommendations. This would give the opportunity for Teachers to make recommendations for starting articles in a study group; all users in the study group would have he teacher as a favourite recommender and would be members of their own profile grouping. Recommenders will build up a history of what they have recommended and the perceived accuracy of their recommendations. It is foreseeable that users may have preferred recommenders. Communication One undecided aspect to the system is whether there will be a mechanism which would allow peers of the same profile grouping to be able to make contact with one another. 4 Conclusion This paper has identified the problems which Students face in locating research material on the Internet which is appropriate to the level and field of study. It has established the technologies available which would be suitable in addressing these issues. A system called Apellicon has been proposed which would allow Students to more readily locate suitable material for study, and could be used by Teachers as part of a working group in collaborative learning. 4 References [Armstrong97] Armstrong R., Freitag D., Joachims T., Mitchell T., WebWatcher: A Learning Apprentice for the WWW, AAAI Spring Symposium on Info Gathering from Hetrogenous, Distributed enviros, 1997 [Balab97] Balabanovic M., Shoham Y., Fab: Content -based, Collaborative Recommendation, Communications of the ACM, March, 1997 [Camer97] Cameron R., A Universal Citation Database as a Catalyst for Reform, First Monday, Feb, 1997 [Chen00] Chen Z., Meng X., Yarrow: A Real-Time Client Side Meta-Search Learner, Proceedings of the 2000 AAAI Workshop on A.I. for Web Search, 2000 [Cosley01] Cosley D., Konstan J., Riedl J., O Connor M., PolyLens: A Recommender System for Groups of Users, Proceedings of the 7th European Conference on CSCW, Kluwer, NY, USA, 2001 [Dean99] Dean J., Henzinger M., Finding Related Pages in the World Wide Web, Computer Networks, Amsterdam, Netherlands, 1999 [Fensel98] Fensel D., Decker S., Erdmann M., Studer R., Ontobroker: Or How to Enable Intelligent Access to the WWW, 11th Banff Knowledge Acquisition for Knowledge Based System Workshop, Banff, Can., April, 1998 [Frankel96] Frankel C., Swain M., Athitsos V., WebSeer: An Image Search Engine for the World Wide Web, Technical Report: TR-96-14, 1996 [Franklin96] Franklin S., Graesser A., Is it an Agent, or just a Program?: A Taxonomy for Autonomous Agents, Proceedings of the Workshop on Intelligent Agents III, Agent Theories, Architectures, and Languages , 1996 [Frietag97] Freitag D., Joachims T., Mitchell T., WebWatcher: A Tour Guide for the WWW, Proceedings of the International Joint Conference on A.I., 1997 [Lieberman95] Lieberman H., Letizia: An Agent that Assists Web Browsing, Proceedings of the Int. Joint Conf. on A.I.., Montreal Aug., 1995 [Luke97] Luke S., Rager D., Ontology Based Web Agents, Proc. of the 1st Inter. Conf. on Autonomous Agents, 1997 [Maglio96] Maglio P., Barret R., How to Build Modelling Agents to Support Web Searches, Proceedings of the 6th International Conf. on User Modelling, 1997 [Nelson94] Nelson T. H., Hyperformance in the Hyperfuture, John Moores University Multimedia Conference, Liverpool, May 1994 [Pazzani98] Pazzani M., Muramatsu J., Billsus D., Syskill & Webert: Identifying Interesting Web Sites, American Association of A.I., Vol 1, 1998 [Pitkow02] Pitkow J., Personalized Search, Communications of the ACM, Vol 45 No. 9 Sept, 2002 [Scott00] Scott00 J., Social Network Analysis: A Handbook, 2nd Edition, Sage Publications, London, 2000 [Terveen98] Terveen L., Hill W., Evaluating Emergent Collaboration on the Web, Computer Supported Cooperative Work, 1998 [W3C07] W3C, OWL Web Ontology Language, Retrieved on 11/07/07 from the World Wide Web: http://www.w3.org/TR/owl-ref/
<< Back
Copyright Apellicon 2007 - 2010, All rights reserved.
Website templates
