Improving document retrieval using special characteristics of lecture recording documents

Abstract

With the increasing use of lecture recordings, content providers are facing the challenge to make electronic lecture materials both easily accessible and searchable. Therefore powerful search engines need to be implemented that allow users to easily retrieve documents fulfilling their information needs. While there has been a lot of research in the domain of text search, special characteristics of lecture recording documents have not yet gotten much attention. Lecture recording documents differ from text documents such as papers, scripts or web pages because they usually do not contain running texts but rather listings and enumerations. Additionally, lecture recording documents contain time-based data such as an audio or video stream of the lecturer as well as handwritten annotations. Analyzing these additional data streams leads to an improvement of the search process. Our novel approach to analyze annotations of lecture recording documents improves document relevance estimation during the search in lecture materials. Hence, technology had been developed to make the contents and special properties of lecture recordings accessible and searchable. We describe key issues encountered during these developments and present experimental results of our search engine which takes into account the special characteristics of lecture recording documents during the indexing process. Searching our archive of over 15,000 files only takes a few milliseconds and enables us to offer a search-as-you-type user interface, query auto-completion and visual browsing.