Open Journalism & the Open Web

I am going to look at this story from the Guardian as it touches on international relations, health issues and sport:

http://www.guardian.co.uk/sport/2010/sep/23/commonwealth-games-scotland-...

I may well create a small application to go with it!

My partner is David Medinets and we're still going back and forth on our assignment. However, I was inspired by his semantic wiki and wanted to put something out there that I'm messing with. I'm looking for an efficient way to do source analysis for the news site I'm working with--and I'm talking source in the hack sense.

What I want to do is take a period of time and look at every quoted source in every published article. For my little experiment, I took a feature I wrote in 2009 (because I have the copyright) and created a semantic wiki page for it. I labeled every quoted source using a simple taxonomy:

Source Class 1: In Person Interview
Source Class 2: Phone Interview
Source Class 3: Email Interview
Source Class 4: Observed

Here's how it looks (scroll to the bottom for a breakdown by source class):
http://sourceanalysis.referata.com/wiki/Main_Page

I didn't create a class for facts gleaned from off-the-record sources or documents but you get the idea.

So how is this useful? In this example, I'm satisfied with the quantity of sources named in the piece. Nearly all of my interviews were done in person or over the phone. However, ALL of my quoted sources were men. Since my sources were mostly sports writers who were 50+, this is not a huge surprise. But a look at my original source list shows that I reached out to five women sources. We weren't able to connect by deadline, but that's my fault for not being more persistent.

Looking at articles published by the news organization I'm working with now, I'd create a class for spokespeople (our reporters seem to lean heavily on them) and quoted press releases.

I can also see a utility to creating a class for the quotes themselves to have a look at just how substantive a reporter's quotes are--though this could be misleading when a reporter sets up a simple quote with something of substance they pen themselves.

As a hack, this little experiment comes from a concern I have that reporters consumed with feeding the beast are making fewer phone calls and doing fewer in-person interviews--which means they are not building the sorts of relationships that make a reporter effective.

I have no doubt there is a more efficient way to do this sort of analysis for a large number of articles, but I have no idea what that is.

Hackers, you have any thoughts?

And to the hacks: what other classes would you create?

Interesting. Also, I want to make a shirt that says: "2011: The Year of the Semantic Web"--just because.

With semantic mediawiki, I did not see a way to mark something in multiple categories. For example, in order to mark someone as both a crook and a donor, you'd need to mention that person twice which will lead to awkward sentences. Another approach is generic tags (like Person) which leads to disambiguation pages.

I haven't had a chance to use GATE yet. Looking at its web page was a bit daunting.

David Medinet: The class approach was really just a matter of keeping it generic for the experiment--helped me to better imagine other classes while looking at the results.

I looked at this story at The Independent

http://www.independent.co.uk/news/business/news/richest-dynasties-back-i...

* Title seems an The Independent editor's interpretation: "Richest dynasties back in the money after crunch" not really developed in the article.
* The top list of US billonaires came from Forbes 400 annual survey: 272 of the 400 richest Americans are self-made.
* The distribution of the "wealth of millonairs" proposal has no direct source.
* Using Forbes 400 as a "recruiting rool" is "even told" by Mr Buffett
* Mr Buffett quote "The way I got the message..." is direct, but It doesn't said where did they took from. Internet?
* The "rebourse of the economy" has unknown source.
* Facebook owner information has not defined source, but reader can presume that is Forbes 400 also.
* "Mr Moskovitz left Facebook in 2008" has not known source.
* Biggest looses source, should be an interpretation from reading Forbes 400.
* David Koch direct quote "One day, my father..." is unknown. The article said that it's something that Koch said "in 2003".
* The Forbes Rich Top List: 25, placed after the article has of course Forbes as unic source.

My conclusion is that there is one "main source" that is Forbes 400 list, but it is not clear when and where the journalist use the Forbes list information, and when he is interpreting the information by him self.
The second "main" source of this article should be something like "Internet". The two direct quoted are no specified when and where they had been told, and the interpretation about the trends of the economy has no source. Neither the information about the idea of distributing the wealthy from millonairs is sourced in any way. For the journalist It seems that is "something that people already knows".
One extrange thing: the "financial" word in the text has a not related link to an advertisement.

Here is the story I pulled.

http://news.yahoo.com/s/ynews_excl/ynews_excl_pl3712;_ylt=AkHluPzqww2_86...

I chose a story from Yahoo! News because it is destination many people use for news. Yahoo! News brings together a nice variety of content from news outlets I view mostly as reputable.

This particular story is based on a phone interview with the primary source. There are several statements of fact sprinkled into the story that are not sourced but seemed to be based on common knowledge or can be inferred from quoted material with the primary source.

The article also contains links to other articles that add facets to the story without bogging down the primary reporting.

Source: Shepard Fairey, phone interview

Facts
Unnamed Sources
“The poster of Barack Obama became a rallying image during the hope-and-change election of 2008.”

“It's been reproduced countless times on the Internet, and a parody version, with Obama as The Joker and "Socialism" in place of "Hope," is a favorite at Tea Party rallies.”

“Fairey's blue-and-red image was altered from an Associated Press photograph of Obama, , and the artist is embroiled in an ongoing lawsuit over use of that picture. (He didn't discuss the case with National Journal.)”

My partner (Sarah Laskow) and I took a story from the NY Times about the egg crisis, and each broke the attribution down, separately.

I broke it down this way:
-- single attribution
-- multiple attribution
-- no attribution (but presumably from older stories)
-- assumed information (based on info that is widely available)
-- & linked attribution, which means there was no direct attribution in the story but there was a link to an archived story or source material. (this overlapped with some other categories.)

Which looked like this:

Type of attribution Number
Single attribution 11
Multiple attribution 15
No attribution in this story 31
Assumed 4
Linked attribution 5

& was also translated into a pie chart, which is here:
http://bit.ly/cPCMiR)

I worked with Terri on the same story: http://sports.espn.go.com/espn/eticket/story?page=100725/stadiumconcessions
We both made an independent spreadsheet of facts and sources. Mine is here: https://spreadsheets.google.com/ccc?key=0AmQ6_0IhhTX8dElya2FwZWxGRFQ2Qm1...

It's interesting to observe how similar our resulting datasets were, which I guess suggests that this kind of data extraction could be somewhat standardized. Differences I noticed were that I identified more "facts" (perhaps doing a more literal reading of the text), and Terri noted significantly more detail and context about each fact. I think Terri's dataset is an example of what you get when a professional journalist organizes data by hand, while mine is more like what a computer algorithm might produce.

I pulled out this story: "Latest prize is a tweet, not a celebrity meet-and-greet" It was published in yesterday's Wall Street Journal.

I pulled the data by creating a Word document to analyze the facts and sources of the story. I also used the document to suggest other ways I could have gathered and shared the facts and figures mentioned in the story. I have started on some of those ideas, which I will share as I get them completed. Those ideas include using Storify, which some of you saw with Dave Goodchild's piece.

Here is a spreadsheet that shows all the people mentioned in the story, some information about them and how they were included in the story. I think this could serve as a practical exercise to keep track of how I use sources in my stories.

Some thoughts:
** You can hide bad reporting (or shortfalls in reporting) in a story if you write it in a certain way. Of course, I knew this (I've had to write around a fact I didn't have from time to time), but doing this exercise made me realize how easy it is to expose our sourcing and facts if people took the time to deconstruct it like we're doing here. On the flip side, in this story, I was impressed at how well the writer did in compressing a whole lot of facts into one easy-to-read story!

** That said, as journalists, trust from the reader is a very important thing. I wonder if providing the raw information in some sort of data platform could help in building that trust.

** Forming a database, of either the people sourced or the facts themselves, is also great as boiler plate material for future stories. I find myself searching my newspaper's archives for background information when I need it. Having it in an organized structure would be much easier and faster, which is very much needed these days.

** My partner, David Mason, is working on a separate story, but we have plans to follow up and evaluate each other's assignments and perhaps get feedback on how to further improve our respective assignments and perhaps provide more ideas on how to pull and organize data.

Story: http://bit.ly/9Pe1pS
Analysis: https://spreadsheets.google.com/pub?key=0Agbu2FfNye2GdEpPNnY1MDY0NWFlZEo...

Nothing too shocking or revelatory about my analysis here, though it did lay bare how biased our daily newspaper is in Vancouver. The story had three sources: a government minister, a doctor and the executive director of an organization hired by government who also happens to be the brother-in-law of our premier (shocking!), and a bureaucrat/doctor. No sources from health care advocates or patient advocates, etc.

What excited me, though, is the potential for tagging stories in this way. Not only could you tag the source, but could further tag the source's descriptors: doctor, executive director, brother-in-law of the premier. That would be one powerful database.

So, yes, excited to learn more about semantic mediawiki and what David Mason and others were discussing. I especially liked:

"The point here is you're creating a flexible, re-usable, co-developed database with something closer to text than traditional programmer's tools, with the transparent culture of wikis, and it "speaks" the distributed knowledge friendly RDF."

Fascinating and looking forward to learning more.

For this assignment I started to read an article about the Sudan that appeared in Newsweek but I soon determined that the article was more of an opinion piece that would not work well for this assignment. Sure there were facts and sources but overall it was longer history type piece about the Sudan.

I shifted my sights to an article that appeared in my local paper that was much more of a just the facts ma'am type article that would work better for this assignment.

The article I read was entitled "Colorado has worst ethnic education gap in the Country that appeared in the Boulder Daily Camera.

All of the sources in this article were named and are as follows: a report commissioned by the Governor entitled "Colorado's Strategic Plan for Higher Education"; David Aragon, director of student success in the Office of
Diversity, Equality and Community Engagement at CU; John Poynton, spokesman for the school district of St Vrain; and CU Regent Michael Carrigan.

The facts were as follows: CO has a 39 percentage-point ethnic achievement gap compared to 19.3 nationally; Hispanics are the fasatest growing demographic in the state but only 6% have an associate's degree and 8% a bachelor's degree; in-state and out-of-state Hispanic enrollment at CU has increased; Hispanic graduation rates have increased and the St. Vain Valley School District offers additional classes for at-risk students.

I was surprised how many of the facts were not tied specifically to sources. None of facts tied to CU were sourced.

This is in response to Jeff Severns Guntzel post here: http://sourceanalysis.referata.com/wiki/Main_Page
Aside from everything else, I really enjoyed this piece. And the thing I liked about it illustrates one problem I see with the idea that information equals data and is somehow extractable and codable. By focusing on this one guy and his world, the article beautifully describes a certain time and place in journalistic culture. That multi-dimensional portrait only emerges when the "data" in the piece is viewed as it was intended. You have to read it and let the details (aka data) click together in your mind in order to see the whole scene. So you could tag it 'journalism, culture, newspapers, sportswriter,' etc., but is tagging what the semantic-mediawiki-XML-Ruby-Python thing is about? Or are tagging and semantic markup stuff two totally different things? This is a hack knowledge gap. Right now I can search for whatever I want. How does the linking/markup thing improve on that? Do the links make it easier to extract the data? Anyway, cool piece.

Wow, thank you! Between your reply to my questions and a conversation I had today with my hacker collaborator, my understanding of this has increased exponentially. Still miles to go though. There is one sentence here I don't totally understand: "The articles both describe data (via markup) and store data (as articles)." I'm not sure how you are using the word 'articles.' I really appreciate the explanations. Thanks.

I chose a story I wrote a year or two ago for the Atlantic Monthly, partly because I have the copyright, and partly because I think the editors there did the internet side of this story particularly well. (E.g., the "Darwin's Revenge" sidebar.) You can see the original at: http://www.theatlantic.com/magazine/archive/2008/09/heart-of-darwin/6932/

I have put up a partly annotated copy on Google Docs:
https://docs.google.com/document/pub?id=1GdeBSAWEKArBsxNb2qca4DwKFtAksdO...

A couple of general thoughts: A lot of people in this class have said something like "we're in the information business." I don't agree. Humans are not a data-mining species. We are a story-telling species, and a lot of us are in the story-telling business.

Data can of course be useful. And sometimes you can just put it out there and say, "ok, do me." (I liked some of the neat wiki things readers are doing with raw data in that Guardian article someone posted.) But most of the time, for most readers, an editor's job is to make the data get up off the bed and dance with the reader a little. We need to make it tell a story.

I also disagree with the suggestion that stories lacking lots of internal web links are a form of "bad journalism." In my Atlantic story, those links work fine in the "travel advisory" sidebar, which is primarily information. But within the body of the story, links would just invite the reader to break the spell, step out of the story, and go somewhere else. Like footnotes in a book (which I also avoid; endnotes are less distracting).

When I submit a manuscript to the editor, however, I always include those links, along with source names and coordinates, page numbers in books referenced, etc. I do this for fact-checking, because magazines generally insist that I back up every assertion with a source. For the most part, I do this in Word comments. (Which is not great: My manuscripts tend to be polka-dot pink with sidebar comments and dotted lines, and it's sometimes hard to figure out which comment goes with which fact.) The URL links are a godsend to me--particularly when something like the Darwin Correspondence Project puts almost every letter Darwin wrote and received at my finger tips. I could see that this might also be a godsend to readers in an online story. But it would only appeal to me as a writer if the information was turned off as a default, and available for interested readers to turn on as a follow up. The editors I work with at places like The NYT online, Yale Environment 360, etc, also seem to want to avoid too much clutter and protect the mood of the story.

By the way, I take it the shortcut Chris alluded to was OpenCalais at API Playground. If not, what am I missing?

Have you tried the entity extractor located at http://viewer.opencalais.com? You can paste into the form and the OpenCalais service will percolate and extract information like places and people. For example, it found NOPD (New Orleans Police Department) but incorrectly identified 'Body of Evidence' as a movie.

Very late here, my apologies. My partner is Steve Myers, and we wound up going solo a bit. He picked the story, and it's a good one for these purposes.

I decided to read the assignment literally, and set up two classes of data: Sources and Facts, which I organized as the X and Y axes of matrix.

I manually culled the data, and chose a matrix to visually display their interrelations.

In choosing a matrix format, I found myself wishing I could make one with a "Z" axis, where I could list the assumptions/assertions/analysis that emerged out of the data.

But I lack the 3d skills, and instead stuck with two classes of info, "Datum" and "Source," and added an asterisk * whenever a particular datum lacked anything immediately verifiable other than the reporters' own good word.

I like the "meta" approach Steve took of color-coding types of data -- an easy way to visually test the strength of a story's sourcing.

For this piece, I noticed that anything that didn't have a direct source but was presented as factual had the reporters as the source. From that reading, the reporters were the most frequently used sources for their own story.

Of course, they're doing a news analysis, so unattributed assertions are easier to present.

Here's the matrix ... it seems like a useful way to analyze the strength of a story based on its sourcing:

https://spreadsheets.google.com/pub?key=0AhPQrzxcJophdDNIRF8tX2QyOWZrWWR...