We are currently researching on a dataset obtained from a Virtual World environment. We have complete social network data (similar to Facebook), more or less detailed tracking data of the users and know who bought what from whome. The dataset includes 300,000 users, over the three domains in a period of one year. Since we are using the dataset currently mainly to research on the link prediction problem, or tie strengths between users utilizing this cross-domain data, I would be interested in what other problems one might be interested in to research on, if she had the access to such a dataset.
Web-Mining is a field in the Data-Mining subject where a lot of research aspects exists on such data. According to the mined user behavior we can build our websites more user-friendly, can put important information in some particular section of webpages, or can maintain a cloud on the basis of time when load on website increases with the number of users on a particular time interval each day, or on some particular day. We can analyse that what kind of a profile likes another kind of profiles, for example - suppose a user's hobby is playing cricket - but in his firend's list there are persons with hobby "cooking" - then suppose in 100 cricketers 80 likes profiles with "cooking" then we can target those people by sending them to buy a "pizza" from their nearest shop on 10% discount rates. since the person likes "cooking or he is a foody" then he will go there to take advantage of this discount and sales increases and so the profits.
Thank you
Dear Christoph Trattner,
I would probably go with the topic of "Online friendship dynamics". I belief that people know little about social ties and friendship connections. I just have read http://www.hindawi.com/journals/ddns/2011/765640/ which utilizes fuzzy logic and some social-interaction theories in an interesting way. (But please note that I am rather new to this topic and I am not aware of all the other research papers about it – I am just dropping an idea here).
Best regards,
Nik
investigating effects of positional attributes such as centrality, betweenness and closeness in a micro level on buying behavior is my recommendation. your domains are too limited to recommend spatial analysis of data, otherwise it was possible to study the effects of location on buying behavior and to find purchase hotspots and micro, meso and macro differences of network structure in and out of hotspots.if your data permits consider time series or panel analysis of purchasing behavior over time. Anyway the possible types of analysis depends on when, how and what you have recorded.
I would try to test social contagion in buying behaviour. Do friends buy the same things from the same suppliers? This is a major marketing issue. Up to now marketing targets potential customers according to their individual attributes (age, income, location, etc) rather than their social network. Some industries are upfront in this (like telecome, pharma), most use traditional ways.
Best
Could you share the data for research? We have extensive event-centric data in our Twitris system [1] but Twitter's data sharing policy do not allow free or easy sharing, making it hard for providing data along with evaluations. There are also several advantages of using data from non-real world! If yes, will contact by email.
[1] http://twitris.knoesis.org
- Badges
- Science topic