Lak11 Week 3 and 4 (and 5): Semantic Web, Tools and Corporate Use of Analytics

Two weeks ago I visited Learning Technologies 2011 in London (blog post forthcoming). This meant I had less time to write down some thoughts on Lak11. I did manage to read most of the reading materials from the syllabus and did some experimenting with the different tools that are out there. Here are my reflections on week 3 and 4 (and a little bit of 5) of the course.

The Semantic Web and Linked Data

This was the main topic of week three of the course. Basically the semantic web has a couple of characteristics. It tries to separate the presentation of the data and the data itself. It does this by structuring the data which then allows linking up all the data. The technical way that this is done is through so-called RDF-triples: a subject, a predicate and an object.

Although he is a better writer than speaker, I still enjoyed this video of Tim Berners-Lee (the inventor of the web) explaining the concept of linked data. His point about the fact that we cannot predict what we are going to make with this technology is well taken: “If we end up only building the things I can imagine, we would have failed“.

[youtube=http://www.youtube.com/watch?v=OM6XIICm_qo]

The benefits of this are easy to see. In the forums there was a lot of discussion around whether the semantic web is feasible and whether it is actually necessary to put effort into it. People seemed to think that putting in a lot of human effort to make something easier to read for machines is turning the world upside down. I actually don’t think that is strictly true. I don’t believe we need strict ontologies, but I do think we could define more simple machine readable formats and create great interfaces for inputting data into these formats.

Use cases for analytics in corporate learning

Weeks ago Bert De Coutere started creating a set of use cases for analytics in corporate learning. I have been wanting to add some of my own ideas, but wasn’t able to create enough “thinking time” earlier. This week I finally managed to take part in the discussion. Thinking about the problem I noticed that I often found it difficult to make a distinction between learning and improving performance. In the end I decided not to worry about it. I also did not stick to the format: it should be pretty obvious what kind of analytics could deliver these use cases. These are the ideas that I added:

  • Portfolio management through monitoring search terms
    You are responsible for the project management portfolio learning portfolio. In the past you mostly worried about “closing skill gaps” through making sure there were enough courses on the topic. In recent years you have switched to making sure the community is healthy and you have switched from developing “just in case” learning intervention towards “just in time” learning interventions. One thing that really helps you in doing your work is the weekly trending questions/topics/problems list you get in your mailbox. It is an ever-changing list of things that have been discussed and searched for recently in the project management space. It wasn’t until you saw this dashboard that you noticed a sharp increase in demand for information about privacy laws in China. Because of it you were able to create a document with some relevant links that you now show as a recommended result when people search for privacy and China.
  • Social Contextualization of Content
    Whenever you look at any piece of content in your company (e.g. a video on the internal YouTube, an office document from a SharePoint site or news article on the intranet), you will not only see the content itself, but you will also see which other people in the company have seen that content, what tags they gave it, which passages they highlighted or annotated and what rating they gave the piece of content. There are easy ways for you to manage which “social context” you want to see. You can limit it to the people in your direct team, in your personal network or to the experts (either as defined by you or by an algorithm). You love the “aggregated highlights view” where you can see a heat map overlay of the important passages of a document. Another great feature is how you can play back chronologically who looked at each URL (seeing how it spread through the organization).
  • Data enabled meetings
    Just before you go into a meeting you open the invite. Below the title of the meeting and the location you see the list of participants of the meeting. Next to each participant you see which other people in your network they have met with before and which people in your network they have emailed with and how recent those engagements have been. This gives you more context for the meeting. You don’t have to ask the vendor anymore whether your company is already using their product in some other part of the business. The list also jogs your memory: often you vaguely remember speaking to somebody but cannot seem to remember when you spoke and what you spoke about. This tools also gives you easy access to notes on and recordings of past conversations.
  • Automatic “getting-to-know-yous”
    About once a week you get an invite created by “The Connector”. It invites you to get to know a person that you haven’t met before and always picks a convenient time to do it. Each time you and the other invitee accept one of these invites you are both surprised that you have never met before as you operate with similar stakeholders, work in similar topics or have similar challenges. In your settings you have given your preference for face to face meetings, so “The Connector” does not bother you with those video-conferencing sessions that other people seem to like so much.
  • “Train me now!”
    You are in the lobby of the head office waiting for your appointment to arrive. She has just texted you that she will be 10 minutes late as she has been delayed by the traffic. You open the “Train me now!” app and tell it you have 8 minutes to spare. The app looks at the required training that is coming up for you, at the expiration dates of your certificates and at your current projects and interests. It also looks at the most popular pieces of learning content in the company and checks to see if any of your peers have recommended something to you (actually it also sees if they have recommended it to somebody else, because the algorithm has learned that this is a useful signal too), it eliminates anything that is longer than 8 minutes, anything that you have looked at before (and haven’t marked as something that could be shown again to you) and anything from a content provider that is on your blacklist. This all happens in a fraction of a second after which it presents you with a shortlist of videos for you to watch. The fact that you chose the second pick instead of the first is of course something that will get fed back into the system to make an even better recommendation next time.
  • Using micro formats for CVs
    The way that a simple structured data format has been used to capture all CVs in the central HR management system in combination with the API that was put on top of it has allowed a wealth of applications for this structured data.

There are three more titles that I wanted to do, but did not have the chance to do yet.

  • Using external information inside the company
  • Suggested learning groups to self-organize
  • Linking performance data to learning excellence

Book: Head First Data Analytics

I have always been intrigued by O’Reilly’s Head First series of books. I don’t know any other publisher who is that explicit about how their books try to implement research based good practices like an informal style, repetition and the use of visuals. So when I encountered Data Analysis in the series I decided to give it a go. I wrote the following review on Goodreads:

The “Head First” series has a refreshing ambition: to create books that help people learn. They try to do this by following a set of evidence-based learning principles. Things like repetition, visual information and practice are all incorporated into the book. This good introduction to data analysis, in the end only scratches the surface and was a bit too simplistic for my taste. I liked the refreshers around hypothesis testing, solver optimisation in Excel, simple linear regression, cleaning up data and visualisation. The best thing about the book is how it introduced me to the open source multi-platform statistical package “R”.

Learning impact measurement and Knowledge Advisers

The day before Learning Technologies, Bersin and KnowledgeAdvisors organized a seminar about measuring the impact of learning. David Mallon, analyst at Bersin, presented their High-Impact Measurement framework.

Bersin High-Impact Measurement Framework
Bersin High-Impact Measurement Framework

The thing that I thought was interesting was how the maturity of your measurement strategy is basically a function of how much your learning organization has moved towards performance consulting. How can you measure business impact if your planning and gap analysis isn’t close to the business?

Jeffrey Berk from KnowledgeAdvisors then tried to show how their Metrics that Matter product allows measurement and then dashboarding around all the parts of the Bersin framework. They basically do this by asking participants to fill in surveys after they have attended any kind of learning event. Their name for these surveys is “smart sheets” (an much improved iteration of the familiar “happy sheets”). KnowledgeAdvisors has a complete software as a service based infrastructure for sending out these digital surveys and collating the results. Because they have all this data they can benchmark your scores against yourself or against their other customers (in aggregate of course). They have done all the sensible statistics for you, so you don’t have to filter out the bias on self-reporting or think about cultural differences in the way people respond to these surveys. Another thing you can do is pull in real business data (think things like sales volumes). By doing some fancy regression analysis it is then possible to see what part of the improvement can be attributed with some level of confidence to the learning intervention, allowing you to calculate return on investment (ROI) for the learning programs.

All in all I was quite impressed with the toolset that they can provide and I do think they will probably serve a genuine need for many businesses.

The best question of the day came from Charles Jennings who pointed out to David Mallon that his talk had referred to the increasing importance of learning on the job and informal learning, but that the learning measurement framework only addresses measurement strategies for top-down and formal learning. Why was that the case? Unfortunately I cannot remember Mallon’s answer (which probably does say something about the quality or relevance of it!)

Experimenting with Needlebase, R, Google charts, Gephi and ManyEyes

The first tool that I tried out this week was Needlebase. This tool allows you to create a data model by defining the nodes in the model and their relations. Then you can train it on a web page of your choice to teach it how to scrape the information from the page. Once you have done that Needlebase will go out to collect all the information and will display it in a way that allows you to sort and graph the information. Watch this video to get a better idea of how this works:

[youtube=http://www.youtube.com/watch?v=58Gzlq4zSDk]

I decided to see if I could use Needlebase to get some insights into resources on Delicious that are tagged with the “lak11” tag. Once you understands how it works, it only takes about 10 minutes to create the model and start scraping the page.

I wanted to get answers to the following questions:

  • Which five users have added the most links and what is the distribution of links over users?
  • Which twenty links were added the most with a “lak11” tag?
  • Which twenty links with a “lak11” tag are the most popular on Delicious?
  • Can the tags be put into a tag cloud based on the frequency of their use?
  • In which week were the Delicious users the most active when it came to bookmarking “lak11” resources?
  • Imagine that the answers to the questions above would be all somebody were able to see about this Knowledge and Learning Analytics course. Would they get a relatively balanced idea about the key topics, resources and people related to the course? What are some of the key things that would they would miss?

Unfortunately after I had done all the machine learning (and had written the above) I learned that Delicious explicitly blocks Needlebase from accessing the site. I therefore had to switch plans.

The Twapperkeeper service keeps a copy of all the tweets with a particular tag (Twitter itself only gives access to the last two weeks of messages through its search interface). I manage to train Needlebase to scrape all the tweets, the username, URL to user picture and userid of the person adding the tweet, who the tweet was a reply to, the unique ID of the tweet, the longitude and latitude, the client that was used and the date of the tweet.

I had to change my questions too:

Another great resource that I re-encountered in these weeks of the course was the Rosling’s Gapminder project:

[youtube=http://www.youtube.com/watch?v=BPt8ElTQMIg]

Google has acquired some part of that technology and thus allows a similar kind of visualization with their spreadsheet data. What makes the data smart is the way that it shows three variables (x-axis, y-axis and size of the bubble and how they change over time. I thought hard about how I could use the Twitter data in this way, but couldn’t find anything sensible. I still wanted to play with the visualization. So at the World Bank’s Open Data Initiative I could download data about population size, investment in education and unemployment figures for a set of countries per year (they have a nice iPhone app too). When I loaded that data I got the following result:

Click to be able to play the motion graph
Click to be able to play the motion graph

The last tool I installed and took a look at was Gephi. I first used SNAPP on the forums of week and exported that data into an XML based format. I then loaded that in Gephi and could play around a bit:

Week 1 forum relations in Gephi
Week 1 forum relations in Gephi

My participation in numbers

I will have to add up my participation for the two (to three) weeks, so in week 3 and week 4 of the course I did 6 Moodle posts, tweeted 3 times about Lak11, wrote 1 blogpost and saved 49 bookmarks to Diigo.

The hours that I have played with all the different tools mentioned above are not mentioned in my self-measurement. However, I did really enjoy playing with these tools and learned a lot of new things.

Lak11 Week 2: Rise of “Big Data” and Data Scientists

These are my reflection and thoughts on the second week of Learning and Knowledge analytics (Lak11). These notes are first an foremost to cement my own learning experience, so for everybody but me they might feel a bit disjointed.

What was week 2 about?

This week was an introduction to the topic of “big data”. As a result of all the exponential laws in computing, the amount of data that gets generated every single day is growing massively. New methods of dealing with the data deluge have cropped up in computer science.  Businesses, governments and scientists are learning how to use the data that is available to their advantage. Some people actually think this will fundamentally change our scientific method (like Chris Anderson in Wired).

Big data: Hadoop

Hadoop is one of these things that I heard a lot about without ever really understanding what it was. This Scoble interview with the CEO of Cloudera made things a lot clearer for me.

[youtube=http://www.youtube.com/watch?v=S9xnYBVqLws]

Here is the short version: Hadoop is a set of open source technologies (it is part of the Apache project) that allows anyone to do large scale distributed computing. The main parts of Hadoop are a distributed filesystem and a software framework for processing large data sets on clusters.

The technology is commoditised, imagination is what is needed now

The Hadoop story confirmed for me that this type of computing is already largely commoditised. The interesting problems in big data analytics are probably not technical anymore. What is needed isn’t more computing power, we need more imagination.

The MIT Sloan Management Review article titled Big Data, Analytics and the Path from Insights to Value says as much:

The adoption barriers that organizations face most are managerial and cultural rather than related to data and technology. The leading obstacle to wide-spread analytics adoption is lack of understanding of how to use analytics to improve the business, according to almost four of 10 respondents.

This means that we should start thinking much harder about what things we want to know that we couldn’t get before in a data-starved world. This means we have to start with the questions. From the same article:

Instead, organizations should start in what might seem like the middle of the pro-cess, implementing analytics by first defining the insights and questions needed to meet the big busi-ness objective and then identifying those pieces of data needed for answers.

I will therefore commit myself to try and formulate some questions that I would like to have answered. I think that Bert De Coutere’s use cases could be an interesting way of approaching this.

This BusinessWeek excerpt from Stephen Baker’s The Numerati gives some insight into where this direction will take us in the next couple of years. It profiles a mathematician at IBM, Haren, who is busy working on algorithms that help IBM match expertise to demand in real time, creating teams of people that would maximise profits. In the example, one of the deep experts takes a ten minute call while being on the skiing slopes. By doing that he:

[..] assumes his place in what Haren calls a virtual assembly line. “This is the equivalent of the industrial revolution for white-collar workers,”

Something to look forward to?

Data scientists, what skills are necessary?

This new way of working requires a new skill set. There was some discussion on this topic in the Moodle forums. I liked Drew Conway’s simple perspective, basically a data scientist needs to be on the intersection of Math & Statistics Knowledge, Substantive Expertise and Hacking Skills. I think that captures it quite well.

Data Science Venn Diagram (by Drew Conway)
Data Science Venn Diagram (by Drew Conway)

How many people do you know who could occupy that space? The How do I become a data scientist? question on Quora also has some very extensive answers as well.

Connecting connectivism with learning analytics

This week the third edition of the Connectivism and Connective Knowledge course has started too. George Siemens kicked of by posting a Connectivism Glossary.

It struck me that many of the terms that he used there are things that are easily quantifiable with Learning Analytics. Concepts like Amplification, Resonance, Synchronization, Information Diffusion and Influence are all things that could be turned into metrics for assessing the “knowledge health” of an organisation. Would it be an idea to get clearer and more common definitions of these metrics for use in an educational context?

Worries/concerns from the perspective of what technology wants

Probably the most lively discussion in the Moodle forums was around critiques of learning analytics. My main concern for analytics is the kind of feedback loop it introduces once you become public with the analytics. I expressed this in a reference to Goodhart’s law which states that:

Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes

George Siemens did a very good job in writing down the main concerns here. I will quote them in full for my future easy reference.

1. It reduces complexity down to numbers, thereby changing what we’re trying to understand
2. It sets the stage for the measurement becoming the target (standardized testing is a great example)
3. The uniqueness of being human (qualia, art, emotions) will be ignored as the focus turns to numbers. As Gombrich states in “The Story of Art”: The trouble about beauty is that tastes and standards of what is beautiful vary so much”. Even here, we can’t get away from this notion of weighting/valuing/defining/setting standards.
4. We’ll misjudge the balance between what computers do best…and what people do best (I’ve been harping for several years about this distinction as well as for understanding sensemaking through social and technological means).
5. Analytics can be gamed. And they will be.
6. Analytics favour concreteness over accepting ambiguity. Some questions dont have answers yet.
7. The number/quantitative bias is not capable of anticipating all events (black swans) or even accurately mapping to reality (Long Term Capital Management is a good example of “when quants fail”: http://en.wikipedia.org/wiki/Long-Term_Capital_Management )
8. Analytics serve administrators in organizations well and will influence the type of work that is done by faculty/employees (see this rather disturbing article of the KPI influence in universities in UK: http://www.nybooks.com/articles/archives/2011/jan/13/grim-threat-british-universities/?page=1 )
9. Analytics risk commoditizing learners and faculty – see the discussion on Texas A & M’s use of analytics to quantify faculty economic contributions to the institution: http://www.nybooks.com/articles/archives/2011/jan/13/grim-threat-british-universities/?page=2 ).
10. Ethics and privacy are significant issues. How can we address the value of analytics for individuals and organizations…and the inevitability that some uses of analytics will be borderline unethical?

This type of criticism could be enough for anybody to give up already and turn their back to this field of science. I personally belief that this would a grave mistake. You would be moving against the strong and steady direction of technology’s tendencies.

SNAPP: Social network analysis

The assignment of the week was to take a look at Social Networks Adapting Pedagogical Practice (better known as SNAPP) and use it on the Moodle forums of the course. Since I had already played with it before I only looked at Dave Cormier‘s video of his experience with the tool:

[youtube=http://www.youtube.com/watch?v=ZHNM8FWrpLk]

Snapp’s website gives a good overview of some of the things that a tool like this can be used for. Think about finding disconnected or at-risk students, seeing who are the key information brokers in the class, use it for “before and after” snapshots of a particular intervention, etc.

Before I was able to use it inside my organisation I needed to make sure that the tool does not send any of the data it scrapes back home to the creators of the software (why wouldn’t it, it is a research project after all). I had an exchange with Lori Lockyer, professor at Wollongong, who assured me that:

SNAPP locally complies the data in your Moodle discussion forum but it does not send data from the server (where the discussion forum is hosted) to the local machine nor does it send data from the local machine to the server.

Making social networks inside applications (and ultimately inside organisations) more visible to many more people using standard interfaces is a nice future to look forward to. Which LMS is the first to have these types of graphs next to their forum posts? Which LMS will export graphs in some standard format for further processing with tools like Gephi?

Gephi is one of the tools by the way, that I really should start to experiment with sooner rather than later.

The intelligent spammability of open online courses: where are the vendors?

One thing that I have been thinking about in relation to these Open Online Courses is how easy it would be for vendors of particular related software products to come and crash the party. The open nature of these courses lends itself to spam I would say.

Doing this in an obnoxious way will ultimately not help you with this critical crowd, but being part of the conversation (Cluetrain anybody?) could be hugely beneficial from a commercial point of view. As a marketeer where else would you find as many people deeply interested into Learning Analytics as in this course? Will these people not be the influencers in this space in the near future?

So where are the vendors? Do you think they are lurking, or am I overstating the opportunity that lies here for them?

My participation in numbers

Every week I give a numerical update about my course participation (I do this in the spirit of the quantified self, as a motivator and because it seems fitting for the topic). This week I bookmarked 37 items on Diigo, wrote 3 Lak11 related tweets, wrote 5 Moodle forum posts and 1 blog post.