Research-Driven Ideas for Social Data


Attending the European Conference of Information Systems in Tel Aviv last week, I was happy to find that areas such as data visualization, predictive analytics, social CRM and ROI etc found their way into the conference. These areas have been somewhat excluded in the past by more traditional information systems research. This shift in mindset also shows a warming towards potential industry topics on the program that can be far more relevant to practice. Industries such as marketing now rely heavily on a new field of social data science to make sense of the large quantities of information about their brand community.

In the following, I will present three new research initiatives from the conference that in my opinion ought to be credited for their contribution to the field social business intelligence. They offered exceptional concrete value to practitioners of social business intelligence:

1. Predictive Modeling of Social Activity

A project of particular practitioner interest came from researchers at the University of Freiburg who have contrasted different prediction models of social data. Sebastian Wagner presented several techniques for facilitating a “virtual neighborhood watch” by investigating crime-to-Twitter relationships in cities. This primarily demonstrated the extent to which Twitter activity reflects social activity. Since some crime types relate heavily to social activity, Twitter can thus enhance prediction of certain criminal activities. For example, bank robberies are less predictable, but certain crimes such as car theft can be correlated strongly by both place and time. Sebastian’s research team has already been in contact to exchange data collection techniques and is keen to collaborate on social business intelligence projects.

modelling social activity

2. Social CRM & Risk Management – How to put out a firestorm? 

The second project that caught my attention was the German experiments by Anastasia Mochalova at Katholische Universität Eichstätt-Ingolstadt. She has specifically investigated “restricting the spread of firestorms in Social Networks” and how exactly to best put them out. This is an all too familiar fear of social media managers who defend increasingly vulnerable brand reputations online. The question then becomes who, when, where and how to intervene when things go south. Facebook data was collected from over 60k users and 1.5m links to see how we can become better firefighters by answering these questions not only with passive solutions (contingency plan, monitoring tools, etc) and into intelligent active solutions. To find who exactly to target, the research implemented an independent cascade model to look for how influential a neighbor is. It then searches for both early infectees and the largest infectees of negative activity as it spreads like wildfire. Similar to firefighters, the combined technique can pinpoint the outer ring where a fire is spreading and put stop to it there with the network neighbors of infectees. Unsurprisingly the delay must be minimal for users to be saved and the fire to be put out. So having such intelligent detection calculations in place can be of critical importance to brand managers. Ultimately the research showed that targeted firefighting results can be large in effect (High Stopping probability) even with reversals made to just a few specifically targeted people.

3. Tracing Patterns in Online Social Media Artifacts

The third project has built a data visualization tool that identifies patterns of posts when visualizing the footprint of an article. A tool developed by Holon Institute of Technology called TRACE has leveraged the programming language Processing and API access from Wikipedia to make the invisible and complicated authorship data both strikingly visible and elegantly useful. Thus far, the tool’s creator Daphna Levin has been zooming into this quantitative data behind Wikipedia article changes to visually detect these four types of post edit patterns, including:

  • Outliers/vandalism: are identified to reveal added text that is rejected by wikipedians.
  • Stable drop: reveals the evolution as new articles break off and form their own entity (the magic of Wikipedia stigmergy).
  • Stable jump: big increases in valuable text of an existing topic
  • Significant event: lines are thickened and brighter – possibly due to an edit war between two authors

These patterns are intriguing. The obvious next step would be to detect topical relationships to patterns in different types of postings, and perhaps combine such metadata with other patterns from language and sentiment analysis. Daphna highlighted that we can now begin to explore the lifeline of an article that grows and evolves like an organism, and notes that if a Wikipedia article is not continually updated, it loses life and essentially dies (relevance wise). However, Daphna agreed that other social channels may be a more natural embodiment of our life. Facebook for instance, is more of a daily part of our lives, and a system by which we collectively construct the discussion around topics and “vote-up” it’s acceptance with the Like button. Likewise research could be done into what types of posts are edited by their offers, and what types of language content are edited most in terms of topic frequencies in the text.

the footprint of a wikimedia article visualized by @DaphnaLevin at ecis2014

The footprint of a Wikimedia article visualized by @DaphnaLevin at ECIS2014

The last project is also very much in line with tools of the future that we are trying to develop at Mindjumpers. For this reason I met up with Daphna in Tel Aviv after the conference to continue discussing how to use similar techniques for extracting and visualizing Facebook conversation evolutions and ultimately detect patterns and trends by the masses. Here she explained her research as one that digs into visual complexity using systems perspectives. Like the human body or the weather, a social network can’t be broken down since the Internet is bottom up, organic with redundancies almost like a city. The Like button thus makes Facebook the ultimate human example of stigmergy as we collectively vote up elements of our lives, co-creating relevance trails.

Needless to say there has been incredible food for thought in my stay in Tel Aviv and several new contacts made with eager researchers who share a passion with me to develop data science techniques for social business intelligence. You can find the full proceedings to the conference here, which includes the full PDF downloads of all the research presented at the conference.

> You can follow Chris on Twitter @socialbeit



ECIS2014 – New Views on Social Networks

Chris ZimmermanOur industrial PhD Christopher Zimmerman has been attending this year’s European Conference of Information Systems (#ECIS2014) in search of freshly revealed research on social media, and in particular contributions to the emerging field of social business intelligence. This is what he had to report. 

As an industrial PhD working in this field at Mindjumpers, I was not only down in Tel Aviv for the academic networking and the sun, but to hunt down the social media topics of interest to social media marketing agencies like our own.   (more…)