A collaboration with TU-Berlin sees an accepted paper on context-aware user behaviour modelling at RecsysTV (Recsys 2015)

With over 2,000 registered users, Lancaster’s Vision IPTV system is a purpose-built research platform that allows experimentation of new media and web technologies. In a collaboration between Lancaster and TU-Berlin’s Distributed Artificial Intelligence Laboratory (DAI-Labor), Jing Yuan and her colleagues at DAI-Labor investigated a novel approach in analysing a large corpus of TV programme and user interaction data extracted from Vision’s statistics database. We use a LDA-based topic model, with a tailored extension to capture 2-dimensional contextual information, to infer latent user preference and user interaction distributions. In the next step, we’ll generalise the model and integrate it in Vision’s production environment for user evaluation.

Context-aware LDA: Balancing Relevance and Diversity in TV Content Recommenders
Yuan, J., Sivrikaya, F., Hopfgartner. F., Lommatzsch, A., and Mu, M., to appear in 2nd Workshop on Recommendation Systems for Television and Online Video in conjuction with the 9th ACM Conference on Recommender Systems (ACM Recsys). 09/2015

Won the best paper award at IWQoS

It was a great fun visiting Portland, Oregon, the city of roses (and beers!), and getting to know a great research community on QoS. Winning the best paper award was certainly one of the highlights of the trip. If you are ever around Portland, don’t forget to visit Washington Park. I particularly enjoyed the hiking on those great forest trails (over 1000 species of trees), the International Rose Test garden, and the Japanese Garden. For the beers, don’t forget to try the DEAD GUY ALE.


I am working on a JSAC and a TOMM submission.

Setting up OpenFlow testbed for user-level resource allocation experiments using a number of HP 3800 switches. We use VLAN to bridge OpenStack VMs to physical ports (hence the little connections). After that, its all controlled by software.


A Dell server running OpenStack virtualization. We used to run it on 5 PCs but R720 is certainly a better choice to scale up your experiments.