* Motivation and Introduction
  This document describes Evolving Assciation Indexes, a method of indexing
  information through typed links between each other. This method tries to
  solve some of the problems found in classical indexing methods like indexing 
  through keywords. The problems of keywords are related to ambiguosity and
  synonymity. Informations associated to a keyword may have different meanings
  due to different meanings of this keyword (e.g. jaguar may be the keyword for
  an article about a rare specie of wildcats, or it could be the keyword for an
  advertisement of an expensive sportscar). On the other hand there may be different
  synonyms for a keyword and the information may not be related to all of them.
** recall and precision
   Information retrieval has two criterias of quality for methods used in this
   domain: recall and precision.
   Recall compares the amount of retieved information due to a query to all information
   that should match to the query (veryfied through human analysis).
   Precision is indirectly proportional to the amount of noise - non relevant information 
   in the sense of the queries intension.
   Therefore ambigous meanings affect the precision and synonyms affect the recall of
   information retrieval in a keyword based system. To improve recall a thesaurus of
   synonyms has to be used. To improve precision keywords must be constrainted to a set 
   of non ambiugous words. Both improvements can not be automated, are error prone and
   produce a lot of work.
   [PENDING: discuss hierarchical structures like catalogues, thesauri...]
   
** What other papers say about topics of information on the web
   The WWW seems to be actually the biggest and fastest evolving pool of information.
   Furthermore due to its decentrlized, unorganized nature it is quite difficult (maybe event 
   impossible) to provide a high quality keyword index according former improvements. 
   Some recent publications about information retrieval specialize on information found
   [PENDING: bibliography]
   on WWW.

*** Terms: topic, ... (?)
    The terms these papers focus on are "topic distillation", "connectivity analysis",
    "outbound links" and "inbound links".
      
** About how these Papers talk about extracting information out of the web as hyperlink network. 
   (distilling topics...)
   The methods described in these papers try to use the hyperlinked structure of the internet
   to retrieve and rank valuable information. The heuristic used is, that links are providet by
   a lot of different authors regarding the linked document as relevant. A often linked document
   seems to be more valuable information than lesser linked documents. Therefore the text of the
   linked document as well as the links text itself may be analysed. 
   A document containing a lot of "outbound" links (links to other documents) is called a "hub".
   A "hub" for a topic is a document being interesting for navigation and a good starting point
   for a topic.
   A document containing a lot of "inbound" links (links from other documents pointing to it)
   is called a "authorative source" meaning that this document is rated (through hyperlinks) by
   a lot of other people as interesting.
   [PENDING: nore precise]

** How a hyperlink network could modelled using relations 
   [PENDING: motivation]
   [PENDING: describe what i am doing here, try modelling pageranking(informal) or so]

        ||arity||               properties  
   ========================================================
   href	||  2  || assymetric, non transitive, non reflexive
   ========================================================

   d1, d2, d3, d4...
   href(d1,d2)
   href(d3,d4)
   ...

   accessible(x,x).
   accessible(x,y) ::=
      href(x,y)
    | href(x,z), accessible(z,y).


** Motivation of typed relations 
   In a hyperlinked network we have one type of associations between documents - the 
   hyperlink. Obviously information retrieval using hyperlinks is often disturbed by
   hyperlink usage not related to the content ( e.g. <a>You need this Software to view 
   our Page</a> vs. <a>This Page has great content!</a> - while the latter link talks about
   the content, and is therefore relevant information about the destinations quality, the
   former links does not provide information about the content of the destination.). 
   But even if one type of links would be dedicated to content linking, this may still be a 
   little bit weak. Some content may have symmetric relations to each other while other 
   relations may be directed. Relations may also provide hints about the meaning of the
   relation and this may lead to other implied relations. 

** Why A small set of basic typed relations 
   ( which are user providet)
   Due to the nessesarity of typed associations we will define them here.
   It seems to make sense to keep the set of association types small. This set should
   be expressive enough to model other relations. Here a preview on the relations I 
   suggest: relatedTo, specializes, generalizes, equivalent, cites. Further base 
   relations are used to represent the publication date of documents. They are called
   before, after and sameDate. The intension is to modell a bigger set of relations
   providing more information about the related documents. If documents for example are 
   connected in a chain through relatedTo and before associations, we could estimate
   this set of documents being a serie. It is likely that all documents diskuss the same
   topic. These kind of derived associations do not have to be providet by the evolving
   association index autor.   
    

** A (bigger?) set of derived relations 
   (based on the former relations.)
   [PENDING: see associations.tex]

      
