Posted 27th January 2015 by APSIS

Personalisation – A Technical Perspective

My name is Shahab Mokarizadeh and I have a PhD in Computer and Information Systems from the Royal Institute of Technology (KTH), Sweden. A major part of my PhD thesis work involved developing methods for privacy aware recommendation systems and as a result I have an understanding of the web personalisation field.
This passion for developing intelligent and easy to use personalisation systems was the driving motivation to join Innometrics. Not to mention the great atmosphere for excellence and brilliant colleagues, of course.
I am going to write a short series covering different areas of predictive, recommendation and personalisation. These are terms that are used a lot; I want to look again at each of them and offer a glimpse of how Innometrics are approaching these opportunities with our customers.
So, first off – the obvious question but it’s worth revisiting.
What exactly is personalisation?
Generally speaking, personalisation  refers to the process of customising content or structure to the needs of  each individual user.  A personalisation system is a user-adaptive system where the ultimate objective is to provide users with what the need without them asking for it explicitly [1]. Personalisation is done by collecting information and/or acquiring knowledge from the analysis of user behaviour data (e.g. usage data). In this context, prediction is also a kind of personalisation where the objective is to predict the future behaviour or interest of the user and then take appropriate action upon the predicted behaviour or desire. A well-known use case of prediction is on next basket recommendation [2] where the goal is to recommend items to the user that she might want to purchase in her next visit.
Personalisation and generally predictive analysis have recently gained significant attention from the e-business sector ( shops). A driving motivation behind this rise is the fact that personalisation can enhance their retention by providing them with information they need, without any extra explicit effort form the user. This implies that the power of a personalisation system depends on how well the system tackles three key challenges:
  1. How to find out what information the user is seeking right now
  2. How the system can obtain, collect or infer relevant knowledge about its users
  3. How the personalisation decision is then adapted by the front end system and presented to the user.
While the former issue is related to user behaviour prediction, the latter is conventionally described as user profiling.  We at Innometrics have already developed a scalable adaptive user profiling system, which can be easily integrated to the current system and is proven to be an efficient user profiling solution. The system also allows us to implement some personalisation functionality, however the accumulated information in user profiles allows us to infer potentially more valuable knowledge about a user, which pushes the personalisation boundary much further than what is possible using only the information about the user.
In other words, using the collective intelligence concealed in user profiles, the user will receive more accurate, yet personalised, recommendations. And this is how my story began with Innometrics.
[1] Maurice D. Mulvenna, Sarabjot S. Anand, and Alex G. Büchner. 2000. Personalization on the Net using Web mining: introduction. Commun. ACM 43, 8 (August 2000), 122-125. [2] Steffen Rendle, Christoph Freudenthaler, and Lars Schmidt-Thieme. 2010. Factorizing personalized Markov chains for next-basket recommendation. In Proceedings of the 19th international conference on World wide web (WWW ’10). ACM, New York, NY, USA, 811-820.
This post was written by Shahab Mokarizadeh