To parse or not to parse, that is the question: Top, CKP Newspaper Systems has provided the ability to take classified ads in a fielded manner. Bottom, Harris Publishing Systems shows what a parsed ad would look like after XML encoding.

Help wanted: find new ways to take and transmit classifieds

SAN FRANCISCO -- In the 1980s, the classified-ad manager at The Times-Picayune in New Orleans looked at the flats of classified advertising lined up on the make-up banks, smiled and said, "That's money, my boy, pure money."

Nowadays, those banks are undercut by the currents of the Internet, washing classified ads -- and their attendant revenue -- along in its path as it remakes businesses across the globe. The surge of auction sites, spread of on-line communities and ability of the Internet to cut out the middleman has put extreme pressure on the traditional newspaper revenue generator.

Newspapers trying to respond with their own on-line initiatives found that their best weapon, the existing classified database, had to be bludgeoned into a format acceptable to both Internet users and Internet computers. Worse yet, even when extracting all the information from some proprietary front-end classified system, newspapers discovered that the sorts of information expected in the Internet world just did not exist. What worked in print did not work on the Web.

Add to that the problems faced by newspapers and their suppliers in getting information out of classified-ad systems and onto the Web, reformatted for wireless devices, sent to national ad aggregators or reformatted for niche publications.

"You can't do some of this if you've gathered the information by listening to a less than four-minute phone call designed to optimize space," said Jim Hitchman, manager for newspaper technology at the Tribune Co. in Chicago.

Getting ads to the Web was a major topic of conversation at Newspapers 2000, held here June 15-21 in conjunction with NEXPO, both in the exhibit halls and in the conferences, with words such as parsing, XML, adex, Castf and Crest being tossed around.

"Back in 1998, most newspapers were requesting from classified systems vendors the ability to publish on-line," said Kevin McCourt, the Newspaper Association of America's director for real estate and on-line classifieds. "Many of the on-line companies wanted to publish them. Every time they did, the vendor had to write an export program."

Enter stage left, Vienna, Va.'s NAA. Many newspapers and their suppliers are riding the wave, updating an old standard to make things work smoother in the new world.

Even earlier, back in 1988, the ANPA (the NAA's predecessor) had a specification, called Crest (http://www.naa.org/technews/standards/crest.pdf), that detailed how to remotely send ads into classified systems so real-estate agents, automotive dealers and the like would not have to learn a different system for each newspaper. Ten years later, the reformulated NAA looked at updating that standard to take advantage of the Internet and Internet protocols.

"The original version of Crest was an outgrowth of what the [Newark, N.J.] Star-Ledger was doing," said John Iobst, vice president of technical research at NAA. "The original one was the handshake, how to transmit the ad, some typography and special characters. That's all ancient history. For the new stuff, we do the meta data -- the whos, whats, hows and whys. The only thing it has in common [with the original] is the ad itself."

The Crest challenge
The latest incarnation of Crest is the result of a series of meetings between suppliers, ad aggregators, newspapers and NAA staff members, who began meeting in 1998 to address their problems. The initial meeting, held in Houston, outlined the broad plan to move classifieds into the Internet era.

Jack Stanley, vice president of operations and technology at the Houston Chronicle, said committee members initially thought they would be looking at something akin to the old wire service standards, but to transmit the ads. It turned into something completely different.

"The wire protocol turned out to be a non-event," Stanley said. "We made that decision right off, to go to XML and Internet standards for the transmission. The real event is to change the way in which classified ads are taken and massaged and passed out."

"XML is a way to describe fielded data, to effectively have fielded data in your content," said Pat Stewart, general manager at CKP Newspaper Systems of Bedford, N.H. "Its value is it standardizes the way your ads look and the content and allows it to be repurposed. Whoever is repurposing it can choose which of the content they want to use and how it looks."

Taking classified ads traditionally has been done in phone rooms by people whose job it has been to suck enough information out of the caller to write the ad, get the billing information and then go on to the next call. That worked fine for print, but often leaves gaps in the information needed to put an ad on the Web and allow customers to search for it.

Going with XML, the text in an ad needs to be tagged so the different mediums and different search engines have the information they need.

To rectify that, suppliers have taken multiple approaches, often in a single product.

"If you're doing ad entry through the Web, where the end customers are logged on, they typically understand the value of the information; people are willing to enter lots of things in fields and put more information in an ad," said Jim Woodfin, who works on classified-ad projects for Melbourne, Fla.-based Harris Publishing Systems Corp. "But in a typical ad salesroom they are tasked with getting the ad, the credit information and getting off the phone."

Harris uses a variety of methods to ensure that the information needed for other media is available. Although the company's products can take in fielded data, for phone rooms, Woodfin prefers keeping the people doing what they know best, and parsing the information on the fly.

"We can have as many parsers, as many algorithms, as there are types of ads," Woodfin said. "In autos for example, we can look for make, model, year or whatever. I can add the missing information, if needed, but that's unlikely because the person is off the phone."

What makes a legend most?
Using algorithms keyed to the ad classification, Woodfin said he could look for missing information and fill in the blanks. "We look for synonyms to determine make and model, then we do inference. If I know it's a Legend, we know it's an Acura and probably an automatic," he said.

But by parsing on the fly, Woodfin gets a second chance. "When I do the pricing [of an ad], I do the parsing, and it will flag the ad for things that might be missing," he said.

Prompting the ad-taker to get that information also allows for an "upsell," too, he said, referring to the practice of persuading classified customers to buy more space or features. The parser also looks for expressions, such as square feet or bedrooms, to use when parsing real-estate ads.

"Synonyms and expressions work directly off the text and typically are the first things we look for," Woodfin said. "And, knowing them, I can infer other things. It may not be in the text at all, but I can infer it. And the things I infer may be important so I can search for text on the Internet. I can have fielded data, but the field data is redundant if the algorithm is correct."

Parsing has other advantages for Harris, too. Newspapers that are not ready to junk an existing classified system can use a back-end parser, grabbing a dump and parsing that to generate the XML code, maintaining their equity but also moving existing ads out to the Web or to aggregators.

At Mactive Inc., also in Melbourne, fielded entry is the method of choice for generating XML tags.

"We decided that private party people were not going to write their own ads," said Mactive's Lars Bjorn. "They would fill out the fields and we would generate the ads." The Mactive model uses subfields, with the person doing the ad entry choosing the make of a car from a checkbox or drop-down list. Once that choice is made, options come up for year, model, options or whatever, based on the values available for a particular car.

When complete, the system generates the ad in versions for print or the Web, based on what the customer wants. The underlying data is stored in a database -- a structured query language database such as Oracle -- so it can be sliced and diced however which way is needed for the medium. "Once you've got that information in a database, you can do whatever you want with it," Scott Roessler, Mactive's general manager, said.

A bit of both
Edgil Associates' Linda Gagnon was quite up front when asked if they parsed text ads or did they use fielded data. "Yes," she said. "And yes. All of the above."

Like Mactive, Edgil of North Chelmsford, Mass., uses fielded data. And like Harris, Edgil parses text. A bit of both techniques.

"We use fields on-line" when taking ads, Gagnon said. "For transient advertisers, we take them through a wizard, click, click, click, click and we create the ad, create the keywords and create the XML tags."

For ads coming from regular customers, Edgil goes the parsing route. "We take feeds from [auto] dealer inventories, MLS [multiple listing real-estate services], employment companies and so on," she said. "We can take that feed, parse it, evaluate the text and extract the fields relevant to publish."

Whether by parsing on the fly or using fielded text, the object is to get standardized XML tags into the text so it can be moved around to different devices.

"It used to be 'how many nouns could you cram in an ad,'" said Glenn Cruikshank, formerly with Idaho's Lewiston Morning Tribune and now with Kpmg Consulting. "That's not an issue any more, you can take more time to describe a product and sell it better."

Cruikshank said with XML output, you could take one ad and repurpose various parts of it to the Web or wherever. "You could take that the old '6 rms w vu' for print, and expound on the quality of the neighborhood for the Web and put it back to the '6 rms w vu' for the WAP [wireless application protocol] version," he said.

Cruikshank wasn't all that concerned about whether an ad was parsed or used fielded data to get to XML. He did say, however, "the sooner you convert to XML, the better it will be for you. You can have the long form of the ad and pull out the parts you want when you want."

The task force working on the new Crest document type definition (DTD) decided not to try to do the complete lineup of classified categories, focusing instead on transportation, employment and real-estate listings.

Houston's Stanley, who claims he was roped into being the task force chairman at the initial meeting, said the DTD is essentially complete, and now they have moved into the testing and implementation phase, hunting for holes and looking for improvements.

Tribune's Hitchman has been running some tests through a soft launch the company did at its Florida paper, the Orlando Sentinel (https://classxact.orlandosentinel.com/), comparing ads that come in through a web site with the ads that come in through the paper's front-end classified system. "Volume is increasing 50 to 70 ads a day, and we now have a dataset we can use to compare things."

Hitchman is pushing the new Crest specification along for a very simple reason. "Tribune now has 11 newspapers to send the data out. For our interactive group, we don't want to have to figure this out 11 times. And imagine the aggregators. Classified Ventures has 150 data feeds."

Leon Williamson of Harris, also a committee member from the beginning, said that while the testing is going on, the committee needs to do several other things, things that require help from the outside.

"What we're tying to do [with the tests] is to determine what advertising management should be doing with this information," Williamson said.

"When we find missing information, someone from the advertising side of the fence needs to be telling us what is a key field. For the tests, we're going to parse seven fields. It may turn out that four of them are not needed. We may need to add a different four and maybe add six more."

Papers also need to stress the importance of getting phone room staff to go a little slower and get a little more information.

"There is still a philosophical battle going on in the classified phone room," Iobst said. "To field or not to field. They sit there with phone on and fingers on keyboard. It comes in and goes right out their fingertips because they get paid by the number of lines or calls.

"Newspapers are still getting their minds around the fact that they can take ads slower and get a better ad."

-- Steven E. Brier, seb@colepapers.net

CKP Newspaper Systems,
(603) 472-5825,
e-e-mail: Millennium@CKP.com;
Edgil Associates Inc.,
(978) 251-9932,
e-mail: jkosiorek@edgil.com;
Harris Publishing Systems Corp.,
(407) 242-5330,
e-e-mail: hpscmktg@harris.com;
Mactive Inc.,
(407) 254-5559,
e-mail: info@mactiveinc.com;
Newspaper Association of America,
(703) 902-1600.

From THE COLE PAPERS, August 2000, Copyright © 2000, All Rights Reserved.