M.Sc. Joel Pyykkö defends his doctoral thesis Online Personalization in Exploratory Search on Friday the 15th of June 2018 at 12 o'clock noon in the University of Helsinki Exactum Building, Auditorium B123 (Gustaf Hällströmin katu 2b, 1st floor). His opponent is Professor Moncef Gabbouj (Tampere University of Technology, Finland) and custos Professor Petri Myllymäki (University of Helsinki). The defence will be held in English.
Online Personalization in Exploratory Search
Modern society produces vast amounts of digital data related to multiple domains of our lives. We produce data in our free time when browsing the net or taking photos with various personal devices, such as phones or ipads. Businesses and governments also gather a lot of information related to our interests, habits or otherwise personal information (legal status, health data, etc.). The amount of data produced is growning too large for us to be handled manually, and so to assist the user, specialized information retrieval systems have been developed to allow efficient perusal of different types of data. Unfortunately, as using such systems often requires expert understanding of the domain in question, many users get lost in their attempt to navigate the search space. This problem will only be exacerbated in the future, as the amount of data keeps growing, giving us less time to learn about the domains involved.
Exploratory search is a field of research that studies user behaviour in situations, where users have little familiarity with the search domain, or have not yet decided exactly what their search goal is. Situations such as these arise when the user wishes to explore what is available, or is otherwise synthesizing or investigating the data. To assist the user in exploratory search and in finding relevant information, various methodologies may be employed, such as user modeling techniques or novel interfaces and data visualization techniques.
This thesis presents exploratory search techniques for online personalization and feature representations that allow efficient perusal of unknown datasets. These methods are showcased in two different search environments. First, we present a search engine for scientific document retrieval, which takes the user's knowledge level into account in order to provide the user with more or less diverse search results. The second search environment aims at supporting the user when browsing through a dataset of unannotated images. Overall, the research presented here describes a number of techniques based on reinforcement learning and neural networks that, compared to traditional search engines, can provide better support for users who are unsure of the final goal of their search or who cannot easily formulate their search needs.
Availability of the dissertation
An electronic version of the doctoral dissertation is available on the e-thesis site of the University of Helsinki at http://urn.fi/URN:ISBN:978-951-51-4304-4.
Printed copies will be available on request from Joel Pyykkö: email@example.com.