A description of a document is really useful if you’re dealing with large amounts of information. If I’m searching through emails, or even when they’re coming in, I can usually decide whether a particular message is worth reading based on a short description. Unfortunately, creating a full human-quality description is an AI-complete problem, since it requires an understanding of an email’s meaning.

Automatic tag generation is a promising and practical way of creating a short-hand overview of a text, with a few unusual words pulled out. It’s somewhat natural, because people do seem to classify objects mentally using a handful of subject headings, even if they wouldn’t express a description that way in conversation.

If you asked someone what a particular email was about, she’d probably reply with a few complete sentences; "John was asking about the positronic generator specs. He was concerned they wouldn’t be ready in time, and asked you to give him an estimate." This sort of summary also requires a full AI, but it is possible to at least mimic the general form of this type of description, even if the content won’t be as high-quality.

The most common place you encounter this is on Google’s search results page:

The summary is generated by finding one or two sentences in the page that contain the terms you’re looking for. If there’s multiple occurences, usually the sentences earliest in the text are favored, along with ones that contain the most terms closest together. It’s not a very natural-looking summary, but it does do a good job at picking out the relevant quotations to what you’re looking for, and giving a good idea whether it’s actually talking about what you want to know.

Amazon’s statistically improbable phrases for books are an interesting approach, they try to identify combinations of words that are distinctive to a particular book. These are almost more like tags, and are found by a similar method to statistics-based automatic tagging, by spotting combinations that are frequent in a particular book, and not as common in a background of related items. They don’t act as a very good description in practice, they’re more useful as a tool for discovering distinctive content. I also discovered they’ve introduced capitalized phrases, which serve a similar purpose. That’s an intriguing hack on the English language to discover proper nouns, I may need to copy that approach.

The final, and most natural, type of summary is created by picking out key sentences from the text, and possibly shortening them. Microsoft Word’s implementation is the most widely used, and it isn’t very good. There’s also an online summarizer you can experiment with that suffers a lot of the same problems.

There are two big barriers to getting good summaries with this method. First, it’s hard to identify which bits of the document are actually important. Most methods use location, if it’s a heading or the start of a paragraph, and statistical frequency of unusual words, but these aren’t very good predictors. Even once you’ve picked the ones you want to use, here’s also very little guarantee that the sentences will make any sense when strung together outside the context of the full document. You often end up with a very confusing narrative. Even MS in their description of their auto summary tool acknowledge that at best it produces a starting point that you’ll need to edit.

Overall, for my purposes displaying something like Google or Amazon’s summaries for an email might be useful, though I’ll have to see if it’s any better than just showing the first sentence or two of a message. It doesn’t look like the other approaches to producing a more natural summary are good enough to be worth using.

I’ve been thinking hard about painful problems that email analysis could solve, and one of them is the use of a corporation’s email store to discover internal colleagues who have existing relationships with an external company or person you want to talk to. For example, if you want to sell to IBM, maybe there’s someone in your team who’s already talking to someone there. Or internally, you might want an introduction to someone in another department to discuss a problem, and it would be good to know who in your team had contacts there already.

I was discussing these thoughts with George Eberstadt, co-founder of nTag, and he pointed me to a couple of successful companies who are already mining email to do this, Spoke and Contact Networks.

Spoke are interesting because they’re entirely client-based, rather than running on an organization’s whole message store. They work by taking data from everybody who’s running their Outlook add-on, along with information pulled from publicly available sources, and feeding into their own database. You can then search that yourself to find information on people you’re interested in contacting, and people you know who’ve already been in touch with them. It’s effectively creating a global social network largely based on the email patterns of everybody who belongs.

Technically, it sounds like they’re doing some interesting things, such as exchanging special emails to confirm that two people really do know each other, but when I tried for my own name, I didn’t get any useful information. I also am surprised that companies would allow the export of their employees email relationships to a third-party. It may just be that this is happening under the radar, but it seems like the sort of thing that a lot of companies would want some safeguards on. The service encourages individual employees to install the software themselves, without any warning that they might be opening up the organization’s data to third-party analysis. I know at a lot of the companies I deal with would frown on this, to say the least.

Contact Networks seem much more focused on selling to corporations as a whole, rather than individual employees. They build a social graph pulling from several sources internal to the company including email and calendars, CRM, marketing databases, HR and billing systems. They use this to identify colleagues who know a particular individual, which is a succinct description of a ‘painful problem’ that companies would be willing to pay money to solve. They seem to have been very successful, with lots of big-name clients and they were just bought out by Thomson.

It’s good to see how well Contact Networks have done, it’s proof there’s demand for the sort of services I’m thinking of, even if they have already solved the immediate problems I was considering.

I flew into Denver this morning, and even though Defrag doesn’t officially start until tomorrow, I’ve already had a couple of early meet-ups with some of the local folks. It was fun seeing Rob and Josh from EventVue in the flesh for the first time, and hearing about all their hard work. They’ve been running at full steam since May I hope they get a chance for a break soon.

I also made an interesting discovery; Denver has two Hyatt hotels just a couple of blocks from each other, the Grand Hyatt and the Hyatt Regency. I only found this out after I’d dropped my car at the Grand’s valet parking and tried to check in! Luckily I was able to make it to the right one without further misadventure.

Funhouse Photo User Count: 2,097 total, 111 active. Ticking up gradually, with some good weekend active numbers.

Event Connector User Count: 109 total, 4 active.

With the mass of raw data I’m getting from a couple of years of my own email, I’m looking around for a good way to turn that into information. A simple ranking of my closest contacts is a good start, but I want to also see how much of the real-life groupings between others can be revealed. I’m working on a basic force-directed graph implementation, but that still leaves a lot of display choices.

VisualComplexity.com is one of my favorite places to find inspiration. They’ve done a great job collecting some of the most striking methods of presenting graph data visually. I also enjoy the Data Mining blog. Matthew’s a great resource and he’s good at reminding me to focus on getting something useful from my visualizations, not just pretty pictures. He’s headed to Defrag, so I hope I’ll get a chance to say hello.

Funhouse Photo User Count: 2,042 total, 52 active. Steady growth, but a low active count.

Event Connector User Count: 106 total, 13 active. A miniature growth spurt over the last day or two, with a comparatively large number of engaged users.