Automatic Link Generation

One of the powerful WikiPrinciples used within WikiEngines or WikiLikeThings is to generate links automatically. This enables users to go to related places by means of a link which is created by the WikiEngine. The links are generated by process or processes included in the WikiEngine or locally provided by ClientSideProcesses?. A result of this operation is that if a LinkName? is not found to correspond or equate to a PageName, it becomes an AccidentalLink, waiting to be created.

WardsWiki uses this technique of AutomaticLinkGeneration and so that when a LinkNameIsPageName, it is displayed as UnderlinedBlueText?. When the link does not exist, it is marked for creation by the addition of a QuestionMark which becomes a WikiMethodLink?.

Other WikiEngines and WikiLikeThings may use or facilitate other displays and processes to create, identify and navigate via LinkName?s, either within or provided as associated methods external to the displayed textual information.

Rejection: WardsWiki doesn't have AutomaticLinkGeneration

When I see people using CamelCase or explicit brackets ([[whatever]]), I certainly don't think: hey! that's "automatic"! No... it's better than entering a full URL, certainly, but it still requires an explicit representation of intent via special syntax by the wiki writers. The CamelCase or [[square brackets]] are certainly nicer than pounding out full URLs, but they're very much not automatic. AutomaticLinkGeneration implies that links are formed automatically - that is, without any explicit representation of intent; authors shouldn't have to do anything special to automatically generate a link. E.g. if I write 'automatic link generation' anywhere, it might be automatically linked to this page... unless I prefix it or suffix it with something that implies either a different page or a sub-page anchor (like 'automatic link generation is evil' or 'heuristics for automatic link generation').

I hypothesize that one of the biggest problems with true automatic link generation is the existence of ambiguity in the languages represented upon the wiki, plus ambiguity that naturally exists in object identity and the names of people/places/things. MichaelJackson on this wiki is not the one with the white glove. The presence of ambiguity will require heuristics in deciding where to link, and the need for heuristics will prevent any fixed set of rules from providing perfect linkage. OTOH, one might still acquire "near" perfect linkage by use of advanced heuristics. For example, a very simple and powerful approach would be to look at the context of other nearby links (both implicit and explicit) and somehow count up pages in common in order to automatically 'clump' pages together. If more precision is needed, one can also look at nearby words and look for words in common in linked pages. I.e. so if there are two 'Michael Jackson' pages, one for the singer and one for the software engineer, the page associated with 'Michael Jackson' the singer will be linked to if words like 'singer' or 'white glove' are nearby, and the one with Michael Jackson the software engineer will be linked to with words like 'software engineer' nearby.

Explicit injection is valuable: it gives the writers a degree of control, and it may provide a 'seed' or 'kernel' for heuristic-based AutomaticLinkGeneration. It is not, however, any more "automatic" than explicit description of a full or partial URI in HTTP.

Opinions on AutomaticLinkGeneration

AutomaticLinkGeneration can be visually ugly if it results in all those big, blue, underlined words, especially in the middle of quoted text. Of course, it depends on the domain... if one were designing a wiki for support of FanFiction? or SharedFiction?, it needs different visual characteristics than does a wiki designed to support programming of an OS, a dictionary, or an encyclopaedia. It might be better to, instead, provide a small, unobtrusive search box that can link to pages using the same context-based heuristics one would see with AutomaticLinkGeneration, and/or to provide a means of highlighting a small section of text, right-clicking on it, and getting a 'go to page' option (so highlighting 'automatic link generation' might get a different link than highlighting all of 'automatic link generation is evil'. One might also provide a means of showing or hiding all automatic links (e.g. via CSS or a login preference option) or limiting their visual profile to a slight change in color (e.g. black to dark violet or dark green for automatic links, perhaps using something bolder for the explicit ones).

However, all that is UI stuff, which can be resolved independently of the problems surrounding the AutomaticLinkGeneration. If it is resolved, then the main issue with automatic link generation is in figuring out where to link in the presence of much ambiguity, which is essentially a clustering problem.

AutomaticLinkGeneration for Visual Data

There are some interesting new technologies that associate images and meaningfully (and automatically) link them together. MicrosoftLiveLabsPhotosynth? technology seems to be a very powerful tool for this sort of task, linking 2D camera pictures to form 3D landscapes and terrains, tagging images with the metadata that associates them with other images, etc. I quite look forward to seeing where this is going.

Research

There is at least http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.80.3407&rep=rep1&type=pdf

CategoryAutomated CategoryWikiHelp