This is the second part of the evaluation on VideoJS 3 that we will potentially upgrade to in the video player for Vialogues. See my previous post on the background information on the video player technology used in Vialogues.

My plan has two steps. Firstly I will implement a minimal setup for both VideoJS 2.0.2 (the version used in the current Vialogues player) and 3.1.0 (the latest VideoJS release), and compare their performance and cross-platform compatibility for HTML5 video playback (Flash video is not supported in v.2 so we cannot compare). Secondly I will implement a minimal setup for VideoJS 3.0 and Flowplayer 3.2.6, and compare their performance and cross-platform compatibility for Flash video playback. The results from the first step will tell us if it is worth considering upgrading to VideoJS 3, while the second step will tell us if we want to replace Flowplayer with the integrated Flash component in VideoJS 3.

I finally found an English blog of the robot reference librarian called Xiao Tu implemented at the Tsinghua University, China. Here's the link.

Xiao Tu is a web chatting bot designed to help library users who need information but not totally sure what they are looking for. Xiao Tu was implemented with artificial intelligence technologies to mimic human capability in conducting conversations. In a large way it's an intelligent tutor for library information retrieval. Below is a screenshot from the above blog of the English translation of a conversation with Xiaotu.

The blog doesn't cover an interesting episode that happened a week ago. Xiaotu was taken down from the web because it had been mis-taught by bad people.

I was browsing the Harvard Business Publishing for Educators and came across this award-winning simulation called Everest V2. This simulation places students in a dramatic scenario of expedition to Mount Everest and train them about group dynamics and leadership.

I haven't played the sim but learned about it in general by watching this video. In the sim, 5 students will work together to conquer Mount Everest in 6 simulated days in 1.5 actual hours. Each teammate is assigned a different role such as the captain, physician, marathon, and so on. And some roles have special responsibilities. For example, the physician needs to decide when and what medical supplies to be distributed to the teammates. On each day every teammate will make individual and collaborative decisions in order to complete the mission of the day. Decisions include, for example, how many oxygen canisters to carry for oneself, and what to do in case of heavy rain. In order to make collaborative decision, teammates can talk face-to-face if they sit together, or through instant messaging in the sim.

Skillicorn, D. (2007). Understanding complex Datasets: Data Mining with Matrix Decompositions. Boca Raton, FL: Chapman & Hall/CRC

Description: This book provides a solid, in-depth review of various data mining techniques involving matrix decomposition, and how these techniques can be used to analyze complex datasets. It does not dig too deep in terms of the mathematics, but explains exceptionally well the general concepts, the procedures, the application with scalable examples, and the implication of the results. It covers the most commonly used and well established techniques including general matrix decomposition, Singular Value Decomposition (SVD), Semi-discrete Decomposition (SDD), using SVD and SDD together, Graph Analysis, Independent Component Analysis (ICA), Non-Negative Matrix Factorization (NNMF), and so on. I haven't looked into every chapter but found the discussions of SVD, SDD, and their combinations extremely useful for the project I'm working on. It even comes with Matlab code examples which I haven't had a chance to test them out.

RStudio is an open source integrated development environment (IDE) for the R statistics software. If you don’t know about R, R is an open source programming language and software environment for statistical computing and graphics. R is widely used by researchers and enterprises in statistical analysis and data mining.

I had been playing with R in the past two weeks in a data analysis project for Vialogues. As a long-time user (>5 years) of SPSS, I found R highly powerful and flexible being a open-source software. However, I also felt frustrating in learning R because it would take some time to grasp all the functions and syntaxes. In SPSS you can simply import the data, click a few menu items, set the parameters, and click “run” to do a t-test or regression. But in R, you have to do ten times more things. Few established statistical procedures are built into R. You have to write pretty complex scripts even for tasks as simple as importing the data and creating bar graphs. And it’s especially frustrating in trying to figure out what functions to use to do seemingly simple tasks and guessing what are wrong with your scripts if your desired outcome is not seen.

A side product of my current work on the Vialogues data mining research is the statistical summary of Vialogue discussions. I'm sharing the stats here as many of you may find it interesting.

There are totally 1,457 Vialogues with 16,821 comments in total by 3:38 PM, January 19, 2012. Averagely each Vialogue has 11.54 comments. The highest total of comments in a single Vialogue is 214.

In today's EdLab Development & Research meeting, Manav talked about Latent Semantic Analysis (LSA), an approach to analyzing discussions by clustering similar discussions. It is not difficult to grasp the concept of LSA but it is difficult to imagine a real example of how LSA is applied to discussion analysis even after reading this paper and this website.

I found an excellent tutorial on how to do a mini LSA on search results of book titles on Amazon.com.

The most helpful thing that I learned from this tutorial is that we can use a modified version of the Python code example as shown in the tutorial to construct a visualizable graph on how Vialogues discussions are clustered in a semantic space, similar to the graph below:

Sclater, N. (2008). Web 2.0, Personal Learning Environments, and the Future of Learning Management Systems

Article Review

Learning Management Systems (LMS) have been widely adopted in almost every institution to host e-learning for students. LMS works well in that it supports a wide range of learning activities and the institution possesses centralized data for every student. However, LMS also has the bad reputation of being inefficient in every learning activity due to its inflexibility, non-personalizability, and lack of communication features. Therefore many students and lecturers choose to use online tools for certain learning activities instead of totally relying on LMS. It is increasing questioned whether LMS should incorporate social networking features, whether using tools other than the LMS should be allowed, and whether students should be allowed to choose whatever tools for learning.

I came across this nice magic show with iPod on TED

I met Rajesh Jha in California a few weeks ago at Irvine Spectrum Center in Irvine, California. It was raining non stop and there were no seats in the restaurant. So we sat outdoors with a tiny shade above our table. My hair got wet and so did my burrito. But I was very impressed by the SimNewton project Rajesh's company was working on.

SimNewton is a collaborative web application for mechanics simulation designed for use in introductory physics courses at high school and college freshman levels. Teachers can create a challenge by building a mechanical system in the simulation with customized variables and learning scaffolds such as texts, images, videos, and dynamic charts. Then the teacher can export the scenario and share with groups of students. The students in turn will work alone or in the group to solve the problem. Students can also view and comment other groups' work online. Once the problem is solved, the students can export their solutions and share with the teacher for assessment and feedback.