MeatballWiki | RecentChanges | Random Page | Indices | Categories


A news medium must be able to provide you with the possibility to hear about many, many news events each day, but must also allow you to filter in some way since you will not have time to read all of the news.

Conventional newspapers

Conventional newspapers filter by two criteria; subject matter and importance (as judged by the editors). The most important stuff is on the front page. The next most important stuff is on the front page of the various subject-specific sections. Finally, there is a lot of less important material carried inside the various subject-specific sections. In addition, there are various "summary" types of sections; most newspapers have a sidebar or small column with one or two line summaries of some of the most important daily news per subject section (personally, I find The Economist's "The world this week" section is the best concise weekly summary of big events).

My biggest problem with conventional newspapers is that this method of filtering requires them to grade importance "on a curve"; every day, there must be the same amount of stuff on the front page. I have noticed that if I read every front pages of even a few periodicals (and finish some of the stories), I do not have enough time to get my work done.

The general form of the solution

The solution would be some sort of space where news only appears if it is VERY important. Then you would read THAT space (for each periodical that you follow) every single day, but stories would appear there infrequently enough that you would have time to do so.

Ideally, I would like to set two major configuration options. The approximate amount of news that I have time to receive on a given day, and a critical importance threshold above which I am shown any news, even if it goes over my amount. Some sort of software would then rank each news item that it has (according to my criteria, or to others'), and present to me the desired amount of news. Effectively, news items which fall above a certain importance threshold will be displayed, and others will not.

The need for a ratings system

You might ask why I don't just read the front page of one single popular newspaper every day? The problem is that I am also interested in specialized topics. For example, I might like to hear about every milestone in the passage of the DMCA. A paper like the New York Times won't put this on the front page. A specialized site like Slashdot will, but I don't want to spend time reading the other stuff on their front page every day.

News sources generally do not provide the required importance information for this task. Hence, some sort of rating system must be used to allow people to create the importance information (the ratings).

It seems to me that the closest thing (technologically speaking) we currently have is KuroShin. KuroShin allows the community to choose which stories make it to the front page. The problem here is that the importance threshold here is chosen not by the reader but by the community. For me, too many stories make it to the KuroShin front page to justify my checking it often.

The proposed technological architecture

So, I feel that the ideal news system needs the following components:

All of these components should be as modular as possible. They should communicate with each other through RSS or something similar (for lists of news items), and with some sort of rating communications standard.

The ratings user interface

There should be a web site which allows people to enter ratings. Any item in (some set of?) RSS feeds should be "rateable". Ratings are then sent to the ratings database using the rating communication standard. People should be able to rate these news items with the system even if the actual items are on another web site which does not support ratings (for example, the New York Times).

The ratings database

The system should be able to "import" ratings from other services (such as the ratings entering frontend) and to export its ratings. The import functionality allows Ratings Groups to implement (on their own, externally) whatever decision procedure they want to produce their ratings, which are then imported into the database. The export functionality allows ratings to be syndicated to other, similar services.

The data type of ratings should be extendable. Ratings can take the form of numbers, or of categorization into various subject hierarchies, or anything else. Of course, in addition to this "free form" data type, there should be a few "more standard" rating types with well defined semantics (like a standard numerical rating).

The filterer

The next component, the filterer, gets ratings as exports from the rating system component, and then applies some filtering procedure. For instance, it might look at all the news items reported by all of your subscribed Rating Groups, re-weight the importance scores according to how much you like each group, sort the results, and returns the top ten. The result can be emitted as something like RSS.

The end result

Finally, the user interface part of the FilteredRecentChanges displays the filtered items. If the filtered items were emitted as RSS, a standard ChangeAggregator can be used for this step.

-- BayleShanks

Alternative to explicite rating

I think a very good thing is to base the rating on some existing economic market - this is the idea behind google page rank and I believe this is very underexplored subject. People try to create artificial scarcities as basis for the voting system - but this unevitably leads to great difficulties in balancing the system.

Back to the practical part - there is a bounded amount of web information sources any human can (and want to) watch regulary. So how about measuring the reputation of a web page by the number of people that subscribe to this page? -- ZbigniewLukasiak

2 remarks: - It is very hard to provide a unique identifier for an interesting resource. eg: a web page might provide the same information than another webpage like different blogs linking the same source. In the best case, we want a single identifier for that source in order to use all the ratings we have efficiently. - Alternative to explicit rating? This is utopia because computers cannot tell the meaning of your implicit moves. eg: is keeping the browser open an indication that you left the desktop and left or that you are reading (and enjoying?) the content. -- Ky Vinh

This is a kind of straw man argument, I provided allready an example (the Page Rank) of implicite automatic rating. I am now thinking about an ChangeAggregator that would provide as a byproduct a kind of implicite rating in [Active Bookmarks] --ZbigniewLukasiak

I see. Then we don't have the same meaning for "implicit rating". Make a quick search for "research implicit rating" and you'll find what I meant. i.e: subscribing to a bookmark = explicit rating. Moreover, the problem with using such basis (page rank, ...) is that the filtering become community filtering instead of personalized filtering. So I'd rather call that "objective" rating, since it doesn't vary among users. --Ky Vinh

I use the term 'implicite' here because the rating is a kind of byproduct - the user does not explicitely rate the sites, he just expresses his will to be informed about updates, or in the case of Google Page Rank he just links to it. Talking about community and personalized filtering I believe what we need is something in between, something network based (think FOaF?), where the opinion of people close to you weight more than those of the people farther away. -- ZbigniewLukasiak


See also TimeStream, ChangeAggregator, GroupLens.

[CategoryRatingSystem] [CategorySyndication] [CategoryWebTechnology]


MeatballWiki | RecentChanges | Random Page | Indices | Categories
Edit text of this page | View other revisions | Search MetaWiki