21st Century Search and Recommendation: Exploiting Personalisation and Social Media

Using the Internet to find information and interesting content is now one of the most common tasks performed on a computer. Up until recently, search algorithms returned only one-size-fits-all rankings, resulting in very poor performance for ambiguous search queries. Recent work has demonstrated that contextual information - such as the interests of the searcher - can be utilised to provide more accurate results which have been “personalised” and adapted to the user’s current information need and situation. Likewise, information about the user can be brought to bear to mitigate the problem of information overload and filter content so that users are only shown items they are likely to be interested in. In this book chapter we explore new methods for assisting users to find the information they want by reducing the complexity of the search task through personalisation. We explore this problem from the perspective of web search and then by considering a very common form of new socially-generated data - microblogs. We first tackle the problem of search result personalisation in the face of extremely sparse and noisy data from a query log. We describe a novel approach which uses query logs to build personalised ranking models in which user profiles are constructed based on the representation of clicked documents over a topic space. Our experiments show that this model can provide personalised ranked lists of documents which improve significantly over a non-personalised baseline. Further examination shows that the performance of the personalised system is particularly good in cases where prior knowledge of the search query is limited. We then turn our attention to the related problem of recommendation (where the user profile is itself the query) and, more specifically, discuss the possibility of learning user interests from social media data (specifically micro blog posts). We present a short introduction to early work focussing on the difficult task of making use of this vast array of ever-changing data. We demonstrate via experiment that our methods are able to predict, with a high level of precision, which posts will be of interest to users and comment on possibilities for future work.