However, social media consumerism can be fickle with how consumers accept and process advertisements. In many cases, advertisements are considered obtrusive and annoying; and so it falls to business and marketing to provide advertisements in a targeted but natural manner. Last quarter Joshua Miller (an ITS program student) and I embarked on a study to better understand how one might go about collecting data about marketing tools using these new social media platforms. This post outlines some of the wisdom we gathered.

Understand the terms: Just like when you decide to visit another country, you probably think about communicating with the locals, its a good idea to familiarize yourself with the language basics of social media.  We will not dive into common terms such as posts and tweets here but a simple search can lead a novice to several sites that do the job.

Be aware of Influence: Influence is an important factor on social media, specifically on Twitter. One of the more useful break downs of how influence plays a role can be found in a good ICWSM 2010 paper here. Influence on Twitter was analyzed and broken into three main segments: indegree influence, retweet influence, and mention influence. Indegree influence is the number of followers someone has. Examples of accounts with high indegree influence are CNN (1,762,856 followers), the New York Times (3,001,515), president Barack Obama (6,990,413) as well as Shaquille O’Neal (3,620,162). Users with retweet influence are mostly content aggregators. Their retweet influence extends out past their immediate followers, meaning that their tweets can be traced out to users that have no direct connection to them. Mention influence is occupied mostly by celebrities.  This list is where other users on twitter tweet to and mention posts by other users and is good for name and product recognition. There is a bit of overlap between the users on the lists but users can generally be sorted into these three categories.

Once Classified always Classified: Some users on social media have trouble changing their ‘category’. An example is News Networks. They are often retweeted by other users but are not often mentioned since they do not have a platform to specifically communicate with their followers. On the flip side, you see that celebrities are often the opposite, getting much name recognition but not much for retweets from people. A third group exists that does moderately well in both areas due to engaging their followers in conversation. It should also be noted that any of these users’ tweets are in danger of being considered spam should they tweet too often. Spam gets quickly ignored so avoid getting classified as spam!

Note the key players: There are companies that dedicate huge amounts of time and effort into understanding social media. One specific platform Josh and I tested was powered by Davai (I’m a technical advisor). It has two major facets. On the front end they help businesses organize polls, sweepstakes, and contests via Facebook, Twitter, and their own campaign pages. The second facet is their backend where they process the data they collect from these campaigns and map it out to try to understand it.

On the surface the campaigns were very customizable, allowing for unique looking landing pages that can be created with the intention of gathering data and easily becoming a place where people go to to chat about different things going on within the campaign. This helped customers become engaged with the products and the brands they were engaging with. We also noticed that people who receive ads or spam immediately after campaign participation often reacted to them but the response rates decreased as time from engagement increased. In fact an interesting study we read cites that people receiving such ads “confirmed to never have followed the ad links and ignore the ad posts.” (The Effects Of Advertising In Twitter On The Regular 20-30 Year Old User) A twitter poll is actually much less invasive to get the word out. Everyone who is friends with that person sees the tweet where a poll is voted on or a sweepstakes entered. This can generate more contestants and traffic to the company’s website.

Moreover, there is tremendous advantage of such an advertising campaign. This network effect of participants can be leveraged due to their social network. To put it simply, one can understand who has more influence and provides more traffic or creates more awareness of your company, product, or contest. It is possible to map out this network of people so that one can know about the users who have the most followers, the most retweet influence and the most mention influence.

Avoid the Social Shotgun: This collection and mapping of influence on Twitter can help companies avoid what could only be now called the social shotgun method of advertising. On Twitter, constant re-tweeting and hiring interns to message people somewhat at random is a social shotgun and it backfires! People who are actively mentioning your business account may still not be the best people to get word about your product or company out. It can be discovered who in their network is really important and these larger influence nodes can propagate your brand value through connections between you and current or potential customers. If targeted marketing can get someone with, say, high retweet value to tweet about your product, the likelihood of it reaching out to an audience beyond your followers is much higher and this is much better for business.

An example of this kind of marketing can be seen when Old Spice (124,036 users) did their online marketing campaign. People could tweet questions to the Old Spice man and he would do his best to answer all the questions he could via video. However, the Old Spice man made sure to answer plenty of celebrity questions. Paris Hilton (3,065,567) and Alysaa Milano (1,433,384) both got tweets and in the case of Alyssa Milano, multiple replies. The campaign was hugely successful but there was a level of certainty that more people would see the tweets when they go out to people with higher name recognition and high retweet value. Good fun and good marketing got mixed up together and the advertisements are suddenly sought after on Twitter instead of ignored.

This is exactly the kind of service that some noteworthy companies are trying to provide on a smaller scale to small and medium businesses. You might have not have Barrack Obama tweeting about your shop in Seattle, but if Bill Gates (2,310,411), Paul Allen (11,555) or The Seattle Times (29,005) does, you know local people will be visiting and mentioning it in their own tweets and Facebook accounts. This kind of traffic proves useful in a local market as well as on a country-wide or global market and can open up a business to new customers that may not have even known about it prior to seeing it on social media.

End Note: All this makes Twitter and companies such as Davai that leverage social media analytics an invaluable resource for businesses looking both to benefit from social media and to people looking to tap twitter’s active user base. Initial failures to launch successful Twitter and Facebook campaigns should not be immediately thrown out by businesses.

We leave you with a quote from the ICWSM 10 paper: “[we found that] influence is not gained spontaneously or accidentally, but through concerted effort. In order to gain and maintain influence, users need to keep great personal involvement.” (Measuring User Influence in Twitter: The Million Follower Fallacy)

An initial campaign may only generate a handful of new business, but if attention is given to your Twitter and Facebook accounts businesses will see that they get more product and brand recognition than they ever imagined through other media advertising. (Jointly authored with Joshua Miller, ITS student, Institute of Technology, UWT)

When an area is blossoming and growing its tentacles as fast as data mining is today, keeping up with all the noise can be overwhelming. Graduate students in particular can often find it frustrating and overwhelming to track various conferences, workshops and journals. In this post I outline some of the lessons that might be useful to new graduate students looking for ideas and advice on process for reading papers in data mining. It might be useful for students and young researchers in other areas of computer science and informatics as well.

Before we launch into how to read ‘a’ paper, lets reflect on a few quick things that need to be in place before you can embark on reading the paper.

1. Online vs Offline – Some of us are very savvy at reading and grasping papers on the computer/tablet/smartphone(?) screens. Others need the printed copy. Try doing both. I often first read the paper’s abstract and conclusion ‘online’ without saving a copy on my local machine. Why waste memory on crappy papers? If its something I need to read for a class or assignment or if I find that its of initial interest to me, I download it on my machine or save it to my ‘dropbox’ (so I can access it from anywhere).

2. Using your library – Every major university at least in the US will have ‘off-campus’ access to its library system. The University of Washington as an excellent system for allowing students and faculty to access papers from anywhere there is an internet connection using their university account. This proves to be a life-saver if you lead a nomadic computing lifestyle. Learn how to get access to your library systems from both ‘on-campus’ and ‘off-campus’.

3. Time of day – Our brains are best at reflecting and assimilating new ideas when we are most physically relaxed. If these moments are accompanied by a cup of hot tea or coffee in the morning then that adds to the ambience. Some researchers find that this works best for them late at night. Pick a time that works best for you, but once you decide to ‘read’ at that time, put everything aside and just do it! New grad students may need to setup a reward matrix for their first 10 papers.

4. Handling distractions – In our always connected age its easy to find attention suckers all over the place! I’ve already thought about checking my email 3 times so far while writing these 4 points!! You just have to close your mind to external thought once you get a paper in hand. A few years ago I used to fly across the country a lot. Thanks to the no-devices, no wifi, no power outlets, no blankets, no food, and no leg-room policy of the airlines; I found the flight to the best place to take a printout of what I need to read and just read. These days there are times when I force myself to think that i’m on another such flight if I want to get any reading or writing done! So close your mind to external thought and think you are on your way to Hawaii or someplace nice as soon as you finish this paper.

Once the basics are taken care of you must begin the reading.

Becoming a good reader takes practice and skill. We can’t afford to completely read all the papers that we encounter. In todays day and age there are just too many conferences and workshops in areas like data mining and social networks which are of interest to me.

Adopt a 3 step process:

First step: Check if the paper is worth reading. Abstracts are supposed to tell you what’s in the paper but these days the authors seem to be paying less and less attention to writing good abstracts, so they rarely are good indicators of quality/area of the paper. So use the random surfer model. Jump around. Read the  conclusion or future work, and do read the introduction.

Second step: If you find the paper mildly interesting by now, download it and save it. Next, start going through the evaluation section. If the author has been diligent and their algorithm or framework is indeed worthy of attention, they would have spent a good amount of time explaining how they measured the success of their own technique while trouncing the competing previous work and proving themselves to be the new king of the hood. This is often the place where the good stuff is hidden that can lead to getting new ideas.

Third step: Read the entire paper. Use a tool that helps you add comments and highlights. If I’ve reached this step in my reading of the paper, I’ve already downloaded it and printed a hard-copy. Ignore the sections you don’t understand or are poorly written. Start from the abstract and plunder on as if there is no tomorrow. Whack away at the text with a pen or your keyboard inserting comments and highlights. Again focus most of your attention to what you think is the good stuff in the paper. Often your idea of the good stuff and the author’s idea may not match. No worries!

A few things to remember when you read in the third step. Always keep the mind open to how you can use this idea in your work. Think about the ‘what ifs..?’. Think if the idea is well placed, is the author sounding motivated? Is the experimentation reasonable? Are the datasets openly accessible or closely held? Are the experiments done those that one would naturally perform or are they construed to suit the results the author obtained? Good papers will give you ideas and outline assumptions. Most good papers will also make very simple assumptions that others can later relax. As you read more and more, many of these things will come naturally to you.

March on! Build your portfolio of favorite papers and then follow their styles to write your own!

Usually when you travel overseas your favorite US or UK site may not be available, Yet there is good news: iPad can be easily setup to use a VPN connection via PPTP, L2TP or IPSec protocols.

There are two main technologies that you need to be aware of PPTP and OpenVPN. Check out the differences here.

Bottomline, for iPad you’ll essentially need PPTP.

Both http://www.supervpn.com and http://www.hideipvpn.com will offer great free VPN accounts. Super VPN Service is the first one I intend to try out.

A good tutorial on setting up the iPad after you get an account can be found here.

I’m going to test drive them next and give you a holler on which worked for me next week. Stay tuned.

According to Microsoft StreamInsight’s Mohamed Ali, learning to catch a blue fish is central to the understanding of how data streams work and I attribute the story that follows to his vivid imagination!

Dev pondered on the problem of catching the blue fish for a while. Being a relational database developer in his past life, Dev knew that the relational model should be a good one to solve any query in the world so why not this one too! He decided to build a dam on the stream just like a disk resident database. He would need a solid set of bricks just like the disk manager, and then he’d need an army of fishermen with fishing nets, just like a query engine so he could catch the blue fish. Heck, once he had the dam, he’d be able to catch all sorts of fish; red, gold, and blue! That’s the relational model for you; build a dam, stop the stream, catch the fish.

While he was designing the dam, a noble saint was taking a quiet stroll along the banks of the stream. He saw Dev alternating between deep thought and poring over his notes and stopped to inquire what Dev was doing. Dev mentioned the need to build a dam. The saint exclaimed in despair! There had to be another way he said. Why stop the stream just to catch some of the blue fish and destroy the whole ecology and balance of nature? Why not let the stream be a stream and monitor the stream for some blue fish and catch them as they flow by? This was a revelation to Dev, as much as it is a revelation to the traditional database designer.

The key distinction between a data stream model and the traditional database model is the nature of the data and the queries. In the stream model, queries appear much before the data does, whereas we all know well that traditional databases first gather and store the data and then allow for any queries to be processed. This is also termed as the continuous query paradigm since the queries are registered ahead of time and are running continuously over the data stream.

The second distinction between a data stream model and traditional databases is the ability of the stream model to start answering queries in real time as the data comes in. Consider the fact that we know an event happened between 9 am to 9 pm. The relational model records the event, and tells you that the event occured after the fact. In the stream model you can treat the event to occur from 9 am till infinity and learn that somethings up at 9 am itself. This is a more speculative, imperfect model for data delivery, yet it is more efficient and reactive. Tools that handle data streams today often include mechanisms to retract issued triggers and the ability to change their decision when new data becomes available. This makes the streams model extremely attractive for monitoring applications in a variety of domains including healthcare, web services, and the smart-grid, to name a few.

The consequence of adopting a stream model is the realization that consistency comes at a cost. The more amount of main memory you are ready to set aside to retract event and issue updates, the more consistent your eventual system will be. Barga et. al. discussed this proposition to expose consistency versus throughput as control knobs for the continuous query processing paradigm in their ‘Consistent Streaming Through Time‘ paper. When consistency is black and white, the streams approach gives a speculative result along with an exact time when it is black and when it is white. When consistency is a spectrum of colors and tunable, the approximate stream processing concept kicks in to give a confidence score corresponding to each decision. This combination of consistency schemes allows todays data stream engines to address a diverse set of needs and applications.

Who said to catch a blue fish, you first need a dam!

ps: Additional details about the data streams course can be found here.

Led to the development of Twitority.

Several bloggers followed up on the thread and suggested initial ideas notably by twittermaven and seoptimise. 

Later, several sharp devs followed up and here are some interesting threads:

Mayank proposes his algorithm: http://mayank.name/2008/12/27/an-algo-for-twitter-authority/

Denis Florent proposed an improvement: http://www.denisflorent.fr/twitter-authority-algorithm-proposal

Thenextweb has a good set of comments on it here.

Most of the algorithms are one shot computations and don’t take into account an iterative solution so the field is wide open for new proposed solutions.

tweetRank anyone?

I on the other hand think twitter is just fine the way it is. Twitter needs to grow more. Needless to say the Twitter founders and co will be looking to sell to the highest bidder but there are times when companies should necessarily not participate in the auction.

Twitter as a brand is getting enough subscribers and traffic to become a key player in the social media market. In addition to that it is gaining equal importance with the local newspaper and the local NPR station except on a  much more granular, personalized scale.  In this connected web where the local church group now tweets the gospel, and the world wakes up to action on a single tweet on being arrested; you know this is different. When a local non-profit asked me how they could integrate twitter to provide better organization for their volunteers, I knew tweets are becoming as pervasive as search. Search is a way to import knowledge and grow wiser, but search doesn’t let you yet prove how wise you’ve become. Twitter does just that!!

The power of Twitter is the power of information cascades. You don’t need to be in the know anymore, you are the knowledge-base and you aggregate knowledge at a level you can and from the sources you trust.

No search engine can match Twitter in these two respects: personalization and trustworthiness.

In fact integrating twitter with powerful behind the scene algorithms for doing useful things like improving search relevance and advertising is a good idea.

Twitter folks, be bold. Hold on to your horses. Command your potential and stabilize your API. A wonderful ecosystem is growing around you API platform and you should make it more open and accessible. In addition to the plain old tweet, start providing analytics through the API and just see how many data miners jump on the wagon to harvest the crop.

If you can resist the temptation of $1.3 billion for that’s what it will be, right?