Visualization - CorpBlawg

VBDU and MWTTR

Jul 14th, 2008 | Coding, NLP, Visualization | No Comments

Disclaimer: this blog entry is concerned with certain aspects of natural language processing and automated text analysis and may therefore appear excessively nerdy to the non-initiated. Read at your own risk.

VBDU and MWTTR - don’t worry, those aren’t breakfast cereals, government agencies or contagious diseases.

Every once in a while, I feel the need to brush up on my programming skills. Lately, most of what I’ve been doing has been centered around writing human-readable text (as opposed to machine-readable text, i.e. code) and therefore I felt a little PHP practice was in order.

The result of yesterday’s half-day coding session is this script. Read on for an explanation of what exactly it does.

Thinking of what I could possibly code, I remembered an interesting paper by Eniko Csomay that I came across a while ago. In it, Eniko suggests a methodology for segmenting texts into smaller units according to their internal structure*. How can you teach a machine (even in general terms) where one section of a text ends and a new one begins? In her article, Eniko suggests the following approach: if a text moves from one part to another (say the transition from analysis to conclusions in a scientific paper) it is plausible that the lexical material used changes. To simplify a little, one section is likely to use a specific bundle of recurring words, while another will use different terms. Eniko calls these sections vocabulary-based discourse units (VBDUs) and she has shown that variation between VBDUs can be used to find topical and argumentative shifts in a text.

How does it work in practice? VBDUs can be measured by taking a snapshot of N words from a text and then comparing it with another window of the same size that follows the first one.

Let me give an example:

text = Mary had a little lamb, John had a little pony

window size = 5 words

window 1 = Mary had a little lamb

window 2 = John had a little pony

The calculated difference between the two windows equals 2, because John and pony differ from Mary and lamb, while the rest of the words are identical.

How can we calculate this variation for a text in its entirity? By moving through it, word by word.

If we move window 1 forward by a single word and do the same with window 2, the difference between the two windows may change. The example above isn’t terribly well-suited to demonstrate this, simply because the windows are very small, but if you boost window size to 50 or 100 words, you can get an idea of how this works.

Another thing that I decided to implement in my little script is a measure called Moving Average Type-Token Ratio (MATRR)**. The terms types and tokens are used in computational linguistics to differentiate between unique words and total words in a text. To use the example from above, the sentence Mary had a little lamb, John had a little pony consists of 10 tokens (actually 11 if you count the comma), but only 7 types, because the words had, a and little occur twice and we only count each unique word once when looking at types.

Comparing the ratio of unique words to total words is useful for several reasons. Generally, we can expect written texts which convey a lot of information to have a higher type-token ratio than (for example) spoken conversation, where certain material is likely to occur again and again (say, the pronouns I and you). This difference is not absolute, but there is a strong tendency for information-dense pieces of discourse (scientific papers, legal texts) to have a higher TTR than less dense material (casual conversation, probably most blogs).

However, there’s a minor methodological issue. TTR is tied to text length and tends to decrease the longer a text is - the amount of lexical material at our disposal is simply not infinite and therefore the ratio inevitably goes down.

The solution to this problem can be integrated into our approach to VBDU analysis: compare two windows, then move forward by a word and repeat the process.

Right, so what’s the result of all this? Lo and behold

The VDBU Difference and Moving Window Type-Toke Ratio Calculator (and no, that is probably not hyphenated correctly)

Go ahead and try it. Simply paste a text into the window, preferably over 1.000 words, and hit submit. A value of 100 for the window size seemed like a good idea to me - values of under 50 and over 250 appear to work less well.

The resulting chart is drawn using Google’s Visualization API and I think it looks quite spiffy. Here are two examples

A news report from the New York Times (source text, visualization)
The first chapter of Edgar Allan Poe’s short novel Arthur Gordon Pym of Nantucket (source text, visualization)

How can the results be interpreted? The x axis represents the progression of the text - essentially we are moving through the text word by word from left to right. On the y axis three normalized scores a represented: the word-based variability between our two windows (VBDUdiff, light blue), the type-token ratio of the first window (TTR1, red) and the type-token ratio of the second window (TTR2, orange).

Great, so what does it all mean?

By itself, probably not too much. You’re unlikely to find a clear-cut correlation between shifts in topic or section transitions by looking at VBDU_diff peaks (those places where difference between the two windows is highest) only. Language is just too tricky for something that simple. But I can imagine there being interesting shifts in word class percentages and the like from one part of a text to the next. Integrating a part of speech tagger would be interesting, but that’s something I’ll save for another day.

In the mean time, try the script and let me know if you find something interesting. Visualizations are stored on the server for now and you can retrieve them later by using the URL at the bottom of each page.

Oh and to be a bit meta, here’s the analysis of this blog entry. Hmmm….

* I need to note several things regarding my implementation of VBDU analysis:

I’ve reproduced the procedure from memory, meaning it is likely to differ from the original implementation in some form and may incorporate infelicities or errors
in addition to possible methodological flaws, simple programming bugs are also imaginable
as a result, use this at your own risk and do not cite or use this script in a serious context (i.e. publishing) without contacting me first

** Michael A. Convington published an implementation of MATTR for Windows last year and explained the method to me at a conference. Essentially, I’ve just recreated MATTR in PHP, hopefully without any significant bugs.

Why don’t we put the Google N-gram corpus on the Web?

Jul 8th, 2008 | Google, IBM, Linguistics, Many Eyes, Visualization | 3 Comments

Two years ago, the news that Google was going to make available the largest collection of n-grams to the global research community that had ever been compiled sparked a lot of interest. I was among those who immediately ordered those six DVDs… and ever since they have been resting dutifully on a shelf in my office, collecting dust and reminding me that I need to bring them into a more accessible format. Alas, so many things to do, so little time.

Something led me to look for information on that corpus this morning and I came across this. Sadly, the link to Chris Harrison’s site no longer seems to work, but when I saw his visualization I immediately thought of Many Eyes.

My reasoning goes a little something like this:

Google N-gram corpus hosted on Google Palimpsest servers + IBM’s Many Eyes = Fantastic web-based tool for linguists

To elaborate: Google has a gigantic database of word collocations that can be used as a baseline for all sorts of interesting analysis, but you can’t really do any of these things unless you have a user interface and enough computing juice to sift through almost 100 gigabytes of text data on the fly. On the other hand, solutions like Many Eyes are amazing, but currently there’s no way you can use it with a really big data set like the n-gram corpus and therefore the research utility is limited.

But it must be possible somehow to bring together

the data to analyze
the computing power required and
the user interface needed to allow a non-technical person to interact with the data

and to put the whole thing on the Web. It’s Google’s stated intention to host data for us and they are the owner of the n-gram dataset, so I can’t imagine there being any licensing issues. And, as if to put a cherry on that sundae, here’s the announcement of a joint project by IBM, Google and the NSF to do exactly that kind of stuff. Put the 6 DVDs on a cloud, throw in a tweaked version of Many Eyes (think the word tree vis with a few extras) and construction grammarians everywhere will absolutely love it.

What do you think?

Brief screencast on f-score in blogs

Jul 1st, 2008 | Corporate Blogging, Delta, Johnson & Johnson, Marriott, Screencast, Style, Visualization | No Comments

Just because the subject came up in several contexts recently, I decided to make a screencast of me explaining the concept of f-score and applying it to some data from my corpus of company blogs. I tried to embed it in a blog post, but that caused several problems because the clip would neither fit nor scale for some reason.

Click here to view the screencast in a separate window. You can also download (right-click, save) and watch it in your favorite video player, which gives the additional luxury of being able to pause.

The three blogs I look at in the clip are Marriott on the Move, JNJ BTW and Delta Air Lines. Here’s the link to the cited article and to the presentation with the example.

And apologies for my lapse of memory towards the end (which blogs am I comparing again?), but it was a long day and organizing a conference occupies a lot of brain cells. I hope it’s still informative.

A taxonomy of corporate blog subtypes in map format

Jun 16th, 2008 | Corporate Blogging, Research, Visualization | 2 Comments

Sometimes a picture says more than a thousand words - especially when the picture is rather fussy and complicated. I’ve created a map of corporate blog subtypes, the functions they realize and the audiences they address. It’s clearly idealized, but I think it captures the essentials reasonably well.

Have a look at it here. I couldn’t fit it into a blog entry because, as you can see, it takes up quite a bit of screen space.

Thoughts?

Edit: don’t miss the comments.

Organizing a panel on how blogs are changing academic publishing at the Berlin 6 Open Access Conference

Mar 4th, 2008 | Academic Publishing, Open Access Publishing, Open Science, Visualization, Web 2.0 | 1 Comment

Wow, I think I’ve never had a post title as long as that one.

As some of you might know, I’m very much engaged in the Open Access movement and involved in several projects related to making scholarly information more accessible. In light of this, I am enthusiastic to announce that I will be organizing a panel with the working title New Forms of Scholarly Communication: Blogs, Wikis and Web 2.0 in Academia at the Berlin 6 Open Access Conference in November. The event is the successor to previous Berlin conferences organized by the Max Planck Society and its partners and will take place here in Düsseldorf.

What exactly is behind the title of the panel? Essentially, I envision a bundle of presentations centered on these interconnected aspects:

research publishing beyond e-books and e-journals - what new forms of publishing (if any) has the Net brought us?
new ways of dealing with data - how do platforms such as IBM’s Many Eyes and MIT’s SIMILE library affect how we can look at data and, consequently, how we publish?
new ways of collaborating - how do new means of communication and collaboration affect us - for example, the use of social bookmarking tools to create shared bibliographies, use of wikis to collaboratively write books etc?
new ways of evaluation and discussing - how do approaches such as open peer review affect our view of science and the way in which we evaluate research results?

I am pinging the institutions and individuals listed below, which I believe could contribute greatly to making this an interesting and diverse panel. Please do let me me know (via blog or email to puschmann@gmail.com) if you are interested in contributing, or if you have suggestions for subtopics or speakers.

Tools/Technology

Research into eScience and blogs/wikis/social networks in an academic context

Virtual Knowledge Studio

HUMlab

Lilia Efimova

Eszter Hargittai

New concepts and approaches in publishing/reviewing

Nature Peer-to-Peer

Living Reviews

Note that these are just a few names that popped into my head spontaneously - there are many more.

I also realized this morning that one immensely interesting speaker on the changing forms of information and on how we share it, disseminate it and evaluate its usefulness would be JP Rangaswami. About 1,5 years ago, I read this fascinating post by JP about what he called “livebrarians”. The post, in which he sketched out differences between the Net and physical libraries, ignited a debate about what role information “professionals” (in other words, librarians) can play in a read-write environment where retrieval happens via keyword search and semantic information is annotated automatically or by amateurs. I particularly liked this quote: “my problem is I really think that any damned fool can be a librarian.” I fully agree. JP has also recently posted about Many Eyes, a project that I very much want to integrate into the discussion.

One might think that open access publishing is a very specific issue, relevant only to academics and librarians, while what we generally call Web 2.0 is just a bundle of trendy buzzwords and an opportunity for tech companies to make money, and that the two issues have little to do with each other. But I believe that means not seeing the big picture. Ultimately, open access publishing is about making information accessible to anyone with an interest in a given area of research, because it is assumed that what can be created as a result of the information being free is worth more than what can be earned by selling it. Open access is to research what open source is to software and for that reason it should be every bit as relevant to companies.

What corporate blogs look like: JNJ, Chrysler, Palm, Marriott

Nov 1st, 2007 | Chrysler, Corporate Blogging, Johnson & Johnson, Many Eyes, Marriott, Palm Inc, Visualization | 1 Comment

If blogs were people, this would be a little bit like a beauty pageant. I’ve taken four blogs from my corpus of company blogs and analyzed them using IBM’s Many Eyes. Many Eyes is a hosted software tool for quick and simple data visualization - you should try it out if you ever have something statistical to present.

Here are the four (randomly picked) candidates.

1. JNJ BTW

Posts: 52

Words: 17077

Sentences: 729

Average Word Length (AWL): 4.8

Average Sentence Length (ASL): 23.4

Average Words per Post (AWpP): 328.4

Word Cloud:

Word Tree:

2. Chrysler Blog

Posts: 59

Words: 13341

Sentences: 780

Average Word Length (AWL): 4.6

Average Sentence Length (ASL): 17.1

Average Words per Post (AWpP): 226.1

Word Cloud:

Word Tree:

3. The Official Palm Blog

Posts: 46

Words: 9262

Sentences: 446

Average Word Length (AWL): 4.5

Average Sentence Length (ASL): 20.8

Average Words per Post (AWpP): 201.3

Word Cloud:

Word Tree:

4. Marriott on the Move

Posts: 60

Words: 4937

Sentences: 305

Average Word Length (AWL): 4.5

Average Sentence Length (ASL): 16.2

Average Words per Post (AWpP): 82.3

Word Cloud:

Word Tree:

All four candidates have around 50 entries, with word counts ranging from roughly 5,000 (Marriot on the Move) to about 17,000 (JNJ BTW). I’ve picked different starting terms for the word trees, depending on the the respective company’s industry, but you can easily search inside a tree for any word that occurs in the blog.

ManyEyes and a HuffPo word cloud

Jul 21st, 2007 | Linguistics, Other Stuff, Visualization, Web 2.0 | No Comments

It’s amazing what kind of great data visualizations you can create with IBM’s web statistics tool Many Eyes (I’ve used it before). The Many Eyes team has recently added a simple concordancing function so that you can see in what context a given word is used. People doing literary studies can do some interesting things with such a tool, as this word cloud from the ME site demonstrates.

While I was already at it, I decided to create a word cloud for HuffingtonPost.com using 2175 entries made in the last six months. You get a fairly clear idea of the topics that were central in that time by looking at the cloud. In case you were wondering - the terms appear so large because I used the top 50 words with their individual frequencies instead of a raw text.

What McDonald’s is saying

May 9th, 2007 | Corporate Blogging, Linguistics, McDonald's, Style, Visualization | 1 Comment

Thought it was time to share a few blog-related statistics with you once again. I’ve looked at three different things in McDonald’s Open for Discussion blog using my corporate blogging corpus.

a) f-score (for details on what that is, read this post)

b) most frequent nouns

c) collocates of the noun PACKAGING

For visualizing f-scores and noun frequencies I’ve once again used IBM’s nifty Many Eyes tool. Have a look.

For the third step (collocates) I’ve used what’s called a concordancer in linguistics to look at the contexts where the noun PACKAGING is typically used.

Concordance for PACKAGING

1. Designing Packaging With the Environment in Mind in Open for Discussion

asked how we address sustainability issues in designing our packaging. I’m happy to jump on this question because we’ve b

2. Designing Packaging With the Environment in Mind in Open for Discussion

at McDonald’s as the manager for initiatives to reduce our packaging impacts. The short answer is that we study the pote

3. Designing Packaging With the Environment in Mind in Open for Discussion

hat we study the potential environmental impacts of any new packaging design. We work hard to ensure that our packaging w

4. Designing Packaging With the Environment in Mind in Open for Discussion

f any new packaging design. We work hard to ensure that our packaging will be environmentally responsible while also meet

5. Designing Packaging With the Environment in Mind in Open for Discussion

onable cost. Our main environmental priorities for consumer packaging include:- Minimizing use of materials. - Favoring m

6. Designing Packaging With the Environment in Mind in Open for Discussion

ls made from renewable resources, like wood fiber. - Having packaging that can be recycled or composted. - Incorporating

7. Designing Packaging With the Environment in Mind in Open for Discussion

ckage was featured awhile ago, as an example of sustainable packaging, in a trade magazine called packaging World. I wi

8. Designing Packaging With the Environment in Mind in Open for Discussion

mple of sustainable packaging, in a trade magazine called packaging World. I wish I could take you back to the late 198

9. Designing Packaging With the Environment in Mind in Open for Discussion

d take you back to the late 1980’s so you could compare our packaging then to what it is today. There are big differences

10. Designing Packaging With the Environment in Mind in Open for Discussion

ese solutions–and many others–by working with our primary packaging supplier and Environmental Defense. At the end of

11. Designing Packaging With the Environment in Mind in Open for Discussion

the 1990’s, we calculated the results of our collaborative packaging efforts. They showed that we’d eliminated 300 milli

12. Designing Packaging With the Environment in Mind in Open for Discussion

rts. They showed that we’d eliminated 300 million pounds of packaging during the decade. And that was in the U. S. alone.

13. Designing Packaging With the Environment in Mind in Open for Discussion

next time you eat at McDonald’s, take a closer look at our packaging. And let me know what further questions you have. -

14. Saving the Earth and Saving Money in Open for Discussion

he decade, we’d eliminated a total of 300 million pounds of packaging. And it didn’t cost us a penny. We also worked with

15. Greening Our Supply Chain in Open for Discussion

out our sustainable fisheries program and our work with our packaging supplier. Let me tell you about another initiative-

16. Engaging in the Global Obesity Dialogue in Open for Discussion

rant chain to begin providing nutrition information on food packaging. We are using a simple icon-based format because we

17. We Want You… To Critique Our Worldwide Corporate Responsibility Report in Open for Discussion

ompany?- Limited consumer interest?- Limited technologies - packaging…- Established food production systems - price set

18. We Want You… To Critique Our Worldwide Corporate Responsibility Report in Open for Discussion

know more about your plans for new nutrition information on packaging to be in 20, 000 retaurants worldwide by end of 200

19. Spinning Green in Open for Discussion

me period, McD USA spent more than $1/2 billion on recycled packaging With a leading environmental organization’s help, ou

= 19 matches in 6 blog posts for your query

I’ll omit a detailed commentary this time, just take this as a sort of text-statistical doodle. Oh and I picked OFD for no particular reason - the blog just happened to pop up in my records.

Visualizing blog language data

Feb 9th, 2007 | Corporate Blogging, IBM, Jonathan Schwartz, Microsoft, Robert Scoble, Style, Thesis, Visualization | 13 Comments

I’ve been playing around with this great little tool for several days now and thought I’d share some of the results with you.

But first, here’s a brief recap of what I’ve been doing before I start throwing statistics at you.

I am in the process of building a textual database (or corpus, as linguists call it) of corporate and enterprise web logs. The purpose of this corpus is to investigate corporate blogs as a text type. In the current phase of my research, I am especially interested in the following questions

- how do corporate blogs compare stylistically with non-corporate blogs, news texts and other types?

- is there a typical ‘corporate blogging style’ in terms of how people write?

- are there recognizable differences in style that correspond with differences in purpose or authorship (in other words, do CEOs, marketers, software developers, etc have distinct styles?)

- how much variation is there stylistically between different blogs, different bloggers in the same hub (e.g. MSDN) and between different posts by the same blogger?

- are there patterns of change in style over time?

You might wonder what such a description is good for (well, apart from furthering the pursuit of knowledge and all that). I think that, on the practical level, it will enable us to better understand what people are trying to achieve with blogs and how they do it. Ultimately blogging is about good writing. The trouble is, neither is ‘good’ easily defined, nor is it always the same to everyone on any occasion. Blogging styles are highly dynamic and situation-dependent and I think the most successful bloggers very consciously adapt different styles to address different people and issues.

Right, so what do I have so far?

One of the first measures I’ve implemented into my database is a relatively simple formula for calculating how formal/informational or (on the other end of the scale) involved/context-dependent a text is. This is done by adding the frequencies of certain types of words together and subtracting others, under the assumption that (for example) nouns are more numerous in texts which are primarily informational, while a high frequency of pronouns indicates involvement. The formula looks like this:

0.5 * ((NOUNS + ADJECTIVES + PREPOSITIONS + DETERMINERS) - (PRONOUNS + VERBS + ADVERBS + INTERJECTIONS) + 100)

(see Heylighen and Dewaele 2002)

As you can guess, the results are potentially ambiguous - in other words, texts can have a very high or low score for a variety of reasons - and should be used with care. That being said, the measure produces some pretty interesting results.

This is a chart of f-scores from Robert Scoble’s blog

Each data point in the graph is the f-score for a single post, or the average for several posts made on a single day. As the graph shows, Scoble’s posts are fairly consistently in the 50s in August and September. They surge to over 100 in mid-October and make overall gains in November and December, though these gains aren’t really as significant as they might look at first. The more notable change is the high degree of variation in these months compared to the time span before that.

You might wonder which posts exactly get a high or low f-score. Here are the entries with the highest score, by date.

Comparing new TailRank/DiggTech/TechMeme to Google Reader, 16 October 2006 (f-score 102)

Grapes on a Plane, 29 October 2006 (f-score 97)

The highs and lows of CES, 15 January 2007 (f-score 93)

Photo “training”, 21 January 2007 (f-score 106)

If you have a look at those posts, you’ll probably notice that they aren’t really in any way more formal than Scoble’s other writing. The difference is that they tend to be more informational, i.e. have more and more condensed information crammed into to them than most entries. Lists and enumerations will immediately lead to a high score (because they usually translate into a high noun count) and for Scoble those entries which are written in a sort of telegraph style to convey information about a photowalk or CES thus have a high score. This doesn’t really demerit the f-score as a metric - it simply means that it’s context-sensitive. What’s important is that, with an overall mean score of 60, Scobelizer ranks on the extreme low end of the formal/informational vs involved/contextual scale. To Scoble, blogs really are conversations, not just metaphorically but in a quite literal stylistic way.

That’s the score for one source over time. Let’s compare a bunch of sources.

If you have trouble seeing anything on the chart, look for a little dropdown menu on the lower right hand side labeled dot size. Change it from ‘posts’ to ‘no selection’ and all the dots will be changed to have the same size, which should make the whole thing a lot easier to read.

The chart is a representation of scores for 137 different blogs, computed from data collected during the last five months. Each dot represents a single blog and its average f-score on the x axis. The position of a dot on the y axis indicates the standard deviation of values inside of that blog, i.e. the degree of internal variation

The vast majority of the sources I’ve used are corporate blogs - after all that’s what my research is about. But in addition I’ve also thrown in a few non-corporate sources, simply to be able to compare one type of blog with another one. Thus the list contains 17 personal blogs randomly found via blogger.com, 1 a-list professional blogger (Scoble), 1 political blog hub (huffingtonpost.com) and 3 non-blog sources, namely editorials from the New York Times, the Washington Post and the LA Times collected in the course of this week (see below for a full list of sources).

The first thing likely to catch you eyes are the outliers. On the far right hand side, there is one source simply tagged “Blog” (informative, I know) with a record f-score of 195 and and a standard deviation of 92. That’s Ray Ozzie, Chief Software Architect of Microsoft. Now, if you have a look at his blog you might find that the best description for his writing is not so much formal, but rather “technical” or maybe “information-oriented”. The reasons for the high scores are the many compound nouns (things like development ecosystem, application components, clipboard data formats, etc) coupled with the overall significant length of entries. Like the other outlier, Irving Wladawsky-Berger of IBM, Ozzie also produces very long posts. Ozzie’s longest has 1,700 words, while Wladawsky-Berger is a close second with 1,500. Length tends to coincide with somewhat higher f-scores, however, there are counter-examples. Heather Hamilton has one post with a whopping word count of over 2,000 and an f-score of only 105. Generally brief posts tend to coincide with lower scores, but, as the example shows, there are exceptions.

Overall it is important to consider a few things, especially in regards to the those sources with a high standard deviation and a high f-score:

- the deviation is often high simply because there aren’t many posts (for example, Ozzie only has 6 entries)

- several of the high-deviation blogs are hubs, i.e. they aggregate a number of individual blogs (e.g. MSDN and HuffPo)

But the cool part is that the remaining sources usually contain very conscious stylistic variation (Jonathan Schwarz is a prime example). I other words, they write differently to address different people and achieve different things and this - at least to some extent - stylistically visible. Compare that with the scores for the three newspaper editorials grouped together in the lower right area of the plot. They are surprisingly consistent if you consider that we’re looking at texts published in three different papers, written by an even larger number of journalists. Which just shows that the editorial is a pretty solidified type of text in terms of style, while the (corporate) blog isn’t - at least not yet.

Anyway, I’ll wrap it up for now and save the more in-depth look for another post.

Sources

iUpload InSights
http://hopper.iupload.com/default.asp

Time Leadership
http://www.jimestill.com/

I Love Me, vol. I
http://www.michaelocc.com/

Simply Albert
http://simplyalbert.blogspot.com/

ChristianLindholm.com
http://www.christianlindholm.com/christianlindholm/

PR Thoughts
http://www.prthoughts.com/

Occam’s Razor
http://mgoldberg.typepad.com/occams_razor/

Loic Le Meur Blog
http://www.loiclemeur.com/

CTO Blog
http://www.capgemini.com/ctoblog/

Lakattack
http://spreadlog.net/