Former KTP Associate – now senior data scientist at Carbon – Ryan Jessop, and current KTP Associate Ela Oftadeh, recently presented some of their KTP work at the Royal Society for Statistics (RSS) conference 2020. Diving into their use of statistical models and decision support for personalised advert and content delivery, it’s further reflection of Carbon’s innovative approach to solving key challenges in the adtech space.
Ryan Jessop was part of Carbon’s first KTP in a collaboration with Durham University for which he gained national recognition as “Best Future Innovator” as well as a Data Scientist of the year title at the inaugural AI & ML Awards. Ryan has since become a core part of Carbon’s data science team, with his intent scoring algorithm an integral part of the platform.
KTP 1: Audience creation based on unique intent scoring
Publisher content such as newspaper and magazine articles are usually categorised by user interests: when a user views a number of articles an online profile is built based on their journey around a website and assigned interests. Using the interest data we can subset a publisher’s online audience into more valuable sets of profiles, which can be then targeted for advertising. However, whilst many data platforms use a binary score of ‘interest’ or ‘intent’ it doesn’t necessarily give a fair reflection of the user’s true intent.
In Ryan’s talk he discusses a real case study from a publishing client aiming to improve their advertisers’ ROI by finding more of their ideal customers. He discusses how interests in real-time are assigned to articles on publisher sites, and how the intent scoring algorithm allows us to find highly engaged users. The results from the advertising campaign found a 300% increase in page dwell time, with an Increase in sales and sales revenue for the advertiser and improved traffic volumes.
Check out Ryan’s talk at 50mins 20 seconds of this recording.
Knowledge Transfer Partnerships (KTPs) are collaborations between Universities and businesses to help academic expertise find its way into the business world. Partly funded by the UK government, individuals are jointly recruited by the University and business involved to be KTP Associates. Academic and company supervisors help develop expertise – the associate forms a bridge between academia and industry. There is a substantial personal development and travel budget for the KTP associate to spend. The associate is encouraged to attend events such as conferences and courses to develop related skills.
KTP 2: Leveraging page content for contextually relevant ads
Ela Oftadeh joined Carbon in early 2020 as a KTP associate on our second data science project collaboration between Carbon and Durham University. With a PhD in Statistics, her KTP work has been focused on using machine learning methods such as classification and clustering, and Bayesian statistics where probability expresses a degree of belief in an event.
In Ela’s talk she explains the process of generating advertising content by leveraging page content to make it contextually relevant. The idea here is to fill in the unsold advertising spaces on publisher sites implementing an in-house ad-generating model. The model uses page content and the page’s interest category to choose the best keywords to be shown to a user. We apply natural language processing combined with some statistical methods to identify and rank keyphrases and find their interest category.
Check out Ela’s talk at 1:03:40 of this recording.