Schedule and course plan

Lecture 7: Probabilistic information retrieval, language models
Hedvig Kjellström

Lecture 9: Guest lectures

Anders Friberg, KTH                                                                                                Music Information Retrieval                                                                               The recent paradigm shift in music distribution has created a need for new methods of browsing, searching, and recommending music on the Internet. Given the size of current music databases, typically around 10 million songs or more, automatic methods are particularly useful. An overview of methods and challenges in the field with some snapshots from KTH research will be presented.

Hercules Dalianis, SU                                                                                             Clinical text retrieval - some methods and some applications                       Electronic patient records contain a waste source of information, both in form of structured information as diagnosis codes, drug codes, lab values, time stamps, etc and unstructured in form of free text. Methods - both rule based and machine learning based for retrieving this information is presented. Applications as diagnosis codes assignment, hospital acquired infection detection and adverse drug event detection will be discussed.

Lecture 10: Guest lectures

Simon Stenström and Martin Nycander, Findwise                                               Search solutions from the Trenches                                                                    The presentation will describe the difference between a search index and a search solution and the process of building a search solution. We will show real world examples that explains some of the problems that we encounter when working with different parts of search solutions. We will discuss source data, what you should index, how you create a real search query and what you can create from a backing search engine.

Magnus Rosell, FOI                                                                                                Text Clustering Exploration                                                                                Text clustering can be used to explore the contents of an unknown text set. Presentation of text clusters so that humans can grasp them is very important. If the texts are associated with further information which clusters are potentially more interesting can be decided automatically.

Lecture 11: Guest lecture                                                          

Filip Radlinski, Microsoft Research Cambridge                                               Evaluating Search Engines Without Human Judgments
How can you tell how well a search engine is performing?
Traditional search engine evaluation takes a sample of search queries, fetches results, and manually assesses them. This approach often works well, but also poses a number of challenges. In particular, it can be difficult to be sure that the experts assessing queries really know what the users were looking for with a particular query, in the context of a particular time and place. In the competitive space on commercial search, these challenges can become huge.
This lecture will consider some of the most difficult aspects of manually assessing search engine results, and contrast it with an alternative: observing the behavior of actual users. Search engine logs can record a great deal of user interaction data, and we will see some of the many ways that these interactions can be interpreted as feedback on search quality. The talk will consider the challenges and assumptions that different forms of online evaluation make, as well as methods that can be used to design online metrics that reflect user satisfaction most accurately and most efficiently.

Lecture 12: Guest lecture

CANCELLED