I do a fair bit of research when I was in university and at work, broadly in the areas of Machine Learning, Statistics, and their applications. Here are some of the works I've done:

Illustration of a Facebook Multicell Lift Study

Experimentation - While I work with the marketing teams at ASOS, we have to figure out how we can effectively measure and compare the incrementality of ad campaigns on Facebook. This is no ordinary A/B test due to several features unique to Facebook, necessitating new experiment designs. The paper detailing our work is published in AdKDD'18 workshop.

[Paper | Code | Slides]

Customer-centric predictive models - Together with the team, we also built a number of customer-centric models, including one that predicts the lifetime value of customers. Having such knowledge will help us to optimise our engagement with customers and hence reduce churn. This is featured in KDD'17 paper, the conference being one of the best in data science.

[Paper | Citation]

We also collaborated with Imperial to produce a model that predicts how long will it take for a user to return to the website. It is featured in an ECML PKDD'18 paper.

[Paper | Citation | Code]

Generating customer embeddings using their product views.
Illustration of the change in predicted probability on successive days with unstable vs stable predictions.

Practical machine learning - On a more theoretical side, I looked at how tuning the parameters for a random forest, an established machine learning technique, could change the predictions. In particular, I am interested if successive predictions are "strong and stable" and if the training is cost-friendly enough. This results in a ECML PKDD'17 publication.

[Paper | Citation | Code | Slides]

Social network - In my Master's thesis, I look at how social networks develop, in particular how they accommodate many overlapping communities/groups and why some of the individuals have so many connection and far more other have so little.

Awarded the best thesis of the cohort, part of the thesis - on how I sped up a popular community detection algorithm by 4.5 times - is then expanded as a ICCSW'18 paper.

[Thesis | Paper | Citation | Code]

Bipartite graph showing community affiliations.