Demographic prediction is a key technology which supports the multi-billion dollar on-line advertising industry. Demographic information can come from many sources including on-line registration records, inferred from site visitation and form entry. However one of the highest quality sources comes from panel companies such as Nielsen and Comscore. These companies send the demographics in an aggregated and anonymized format. Although this aggregated data is more challenging to use for prediction, it would be useful to be able to harness these batched data sets to learn relationships that would be able to predict the demographics of general Internet traffic.
Predicitng user future behaviour is the building block for optimizing ad allocation.
I am leading the reseach edge of the predicitve segments project in AOL Platforms. Working with the engineering team,
we have build very scalable machine learnign tool for predicting user behavior using historical data.
We have to address challenges such as:
• Feature egineering and extracting meaningfull features from logs files.
• Fitting model in presence of more than half a million features.
• Modifying Random Forest to achieve 99% of the original performance while it is 10x faster.
In contrast to the power generation, Electricity load historically has been a strong function of weather. As we are installing more and more wind towers, and solar panels, weather condition is becoming a significant factor in realtime energy generation. I have been using machine learning tools to predict energy prices in MISO market.
Efficient allocation of impressions to advertisers in display advertising has a significant impact on advertisers' utility and the browsing experience of users. The problem becomes particularly challenging in the presence of advertisers with limited budgets as this creates a complex interaction among advertisers in the optimal impression assignment. We study online impression allocation in display advertising with budgeted advertisers. That is, upon arrival of each impression, cost and revenue vectors are revealed and the impression should be assigned to an advertiser almost immediately. Without any assumption on the distribution/arrival of impressions, we propose a framework to capture the risk to the ad network for each possible allocation; impressions are allocated to advertisers such that the risk of ad network is minimized. In practice, this translates to starting with an initial estimate of dual prices and updating them according to the belief of the ad network toward the future demand and remaining budgets. We apply our algorithms to a real data set, and we empirically show that Kullback-Leibler divergence risk measure has the best performance in terms of revenue and balanced budget delivery. More Info
Advertiser are always try looking to optimize their budget allocation cross various advertising chanels such as TV, Inernet, and offline. Measuring ROI of each channel indepenedently is a challenging, and it just becomes more complicated when we are considerig inter-dependencies of channels. That is, a customer can be exposed by TV, online, and offline ads. Each channel has its own unique impact on the user, and optimizing budget allocation across channels can significantly boost ROI. We modeled utility of the advertiser as general monotone concave function of the budget. Imperical observed data is used to model utility functions in all channels, and then we maximize the model for ROI.