Here are some responses to the issues raised so far about using "influence links" to find information in Wikipedia. (original post here http://mail.wikipedia.org/pipermail/wikipedia-l/2004-July/015988.html)
First, Hemanshu brings up a potential POV problem: "do we want endless arguments over whether America influenced the Taliban; George Bush influenced oil markets; oil influenced Gulf war.... to name a few?"
To avoid this, there is one rule: an influence must already be present in the article to be made an "influence link". This way, only assertions that have already passed the wiki test are admissible. The key idea is that influence links don't add new information to the article. Rather, they make explicit the influences that are already present in the article. Then, they can be used to analyze relationships among articles (more on possible uses for this system, below).
On another issue, Austin pointed out: "...for it to be useful for any kind of analysis we would have to work out a system to show the extent and type of relationships involved."
The way around this, is to give users the burden of describing the type of each influence link. Users would include a small description next to each influence link. For example, the article on Copernicus reads:
"His theory about the Sun as the center of the solar system, turning over the traditional [[geocentric theory]] (that placed Earth at the center of the Universe), is considered one of the most important discoveries ever...."
On Copernicus' "influences" page, one influence link would look like this:
Copernicus influenced: - [[geocentric theory]], Copernicus turned over the traditional geocentric theory.
If a user searches for a path of influences between Article "A" and Article "C", the result would look like:
- [[A]] influenced [[B]]. Short description of how A influenced B. - [[B]] influenced [[C]]. Short description of how B influenced C.
In the results, each influence link would have a checkbox next to it, allowing the user to easily glance through and un-check any influence links that aren't of the type he/she is looking for. Then the user could re-run the search, with the un-checked influence links filtered out.
Austin also rightly notes that if there aren't enough influence links, they can't be strung together to show complex paths of influence. But, since the influences are already in the articles, I don't think it would bee too much trouble for users to make them explicit by simply listing them under the influence tab.
Hemanshu brings the discussion back to the mission of Wikipedia by asking what all of this has to do with making an encyclopedia. In response, I say that an encyclopedia needs two things: good content, and a structure for finding the content. With conventional encyclopedias, readers found articles in only one way: paging through, alphabetically. Today, Wikipedia uses a search engine. Or, if you'd rather browse by topic, you can navigate your way through Wikipedia's categories. The more ways you can find information in an encyclopedia, the more useful it is, which is why cross-referencing with hyperlinks in articles is so powerful. None of these tools affect the way articles are written--just how they are found. An "influence map" would be one more tool for finding information that allows you to browse through artiles by tracing their influences, and searching for influence paths between articles. Also, by skimming an article's influence page, to see its outgoing/ingoing influence links, you can quickly put the article in context.
Moreover, influence links will create an incentive for users to contribute more to the Wikipedia as a whole. For example, if you are a contributor to the Copernicus article, not only do you want to make sure that the Copernicus article is high-quality and NPOV, but you'll want to make sure that articles on subjects which claim to have influenced Copernicus do so accurately with a NPOV. An influence map for each article would allow you to quickly see who is claimed to be an influence on Copernicus, allowing you to correct mistakes or POVs in articles that you might not otherwise edit. With more people editing more articles, the whole Wikipedia becomes richer.
Does that fill the gaps? Any more thoughts?
Abe
Abe wrote:
First, Hemanshu brings up a potential POV problem: "do we want endless arguments over whether America influenced the Taliban; George Bush influenced oil markets; oil influenced Gulf war.... to name a few?"
To avoid this, there is one rule: an influence must already be present in the article to be made an "influence link". This way, only assertions that have already passed the wiki test are admissible.
That doesn't work. If people disagree on whether X influenced Y, our NPOV policy dictates that the article should state, "People in group ABC tend to believe that X influenced Y, but other people tend to disagree". Your "influence links" can't do that: They are either there (asserting a definite influence) or not (asserting a definite lack of influence).
Timwi
Timwi wrote:
Abe wrote:
First, Hemanshu brings up a potential POV problem: "do we want endless arguments over whether America influenced the Taliban; George Bush influenced oil markets; oil influenced Gulf war.... to name a few?"
To avoid this, there is one rule: an influence must already be present in the article to be made an "influence link". This way, only assertions that have already passed the wiki test are admissible.
That doesn't work. If people disagree on whether X influenced Y, our NPOV policy dictates that the article should state, "People in group ABC tend to believe that X influenced Y, but other people tend to disagree". Your "influence links" can't do that: They are either there (asserting a definite influence) or not (asserting a definite lack of influence).
We've said nothing yes of negative influences like "I saw the damage that the company was doing to the land, and that influenced me to become an environmentalist."
Ec
It's a neat idea, but...
First, it is already possible to explore relationships among concepts in interesting ways by just browsing the site. I'm sure that the vast majority of contributors are like me -- I start reading about one thing for fun, and end up an hour later reading about stuff that is completely unrelated, just by following links which implicitly describe relationships.
Those existing relationships include causality, influence, proximity, temporality, and much more. Why single out "influence" as a privileged kind of relationship?
Second, your example in your original post was of finding out if there was a causal relationship between the Cold War and 9/11. Wqell, I think this example actually points out the artificiality and arbitrariness of how this would end up being used if it existed as a separate feature.
Nearly everything influences everything else, if you want to be philosophical about it. Introducing a feature which tempts people to link "Marie Antoinette" with "Michael Jordan", well, I don't think it makes sense.
---
What *would* be cool, and might just be a different implementation of exactly what you have in mind, would be a tool to find all the (reasonably short) click-paths between any two concepts. I mean, now that I selected the article titles randomly, I actually wonder how many clicks it takes to get from Marie Antionette to Michael Jordan. And what's intervening?
Basically, this doesn't need a special type of "influence map", but rather just an analysis of existing links.
--Jimbo
Jimmy (Jimbo) Wales wrote:
Those existing relationships include causality, influence, proximity, temporality, and much more. Why single out "influence" as a privileged kind of relationship?
See [[Butterfly effect]]
Second, your example in your original post was of finding out if there was a causal relationship between the Cold War and 9/11. Wqell, I think this example actually points out the artificiality and arbitrariness of how this would end up being used if it existed as a separate feature.
Correlation does not imply causality.
What *would* be cool, and might just be a different implementation of exactly what you have in mind, would be a tool to find all the (reasonably short) click-paths between any two concepts. I mean, now that I selected the article titles randomly, I actually wonder how many clicks it takes to get from Marie Antionette to Michael Jordan. And what's intervening?
Over a year ago I raised the possibility of tracing every article back to the Main Page. None in my random sample was more than five links away. Thus if you trace Marie Antoinette and Michael Jordan back to the Main Page, the sum of their links will be a maximum distance. :-)
Ec
On Tue, 13 Jul 2004 16:24:19 -0700 "Jimmy (Jimbo) Wales" jwales@wikia.com wrote:
What *would* be cool, and might just be a different implementation of exactly what you have in mind, would be a tool to find all the (reasonably short) click-paths between any two concepts. I mean, now that I selected the article titles randomly, I actually wonder how many clicks it takes to get from Marie Antionette to Michael Jordan. And what's intervening?
I'm afraid that the shortest connections are often not the one that are most interesting. Going through the links, one short path from Marie Antoinette to Michael Jordan is the three-step:
Marie Antoinette -> October 16 -> October 30 -> Michael Jordan
Michael Jordan got back to the NBA in the same month of the year that Marie Antoinette was executed. Not really something deep...
Andre Engels
wikipedia-l@lists.wikimedia.org