NAVIGATION AND SEARCHING IN RECORDED LECTURES

Recorded lectures are playing an increasingly pronounced role in modern academic education, as is shown in YouTube EDU and iTunesU. At Delft University of Technology (The Netherlands) several lectures are recorded by a sophisticated recording system (Collegerama) to facilitate students with “the ultimate form of lecture notes”. These recorded lectures play a dominant role in courses on Watermanagement within the Delft OpenCourseWare program.

In an MSc research project at University of Twente (The Netherlands) these recording were studied to find ways of enlarging the usability of these lectures. Improving navigation and searching was part of this research.
It was concluded that a table of contents (TOC) of a recorded lecture is important for navigation. Such a table of contents should be drafted by the lecturer or the assisting staff. It is useful to present such a TOC in an interactive way based on screenshots from the HD movie, the timeline and the available text data. This could be generated automatically into a Flash movie by accessing the database content and the related HD lecture movie. The accessibility of a recorded lecture could be further enlarged by creating tag clouds per lecture (limited to 15 words). It was concluded that the best source for these tag clouds are the subtitles of lectures. However even these tag clouds could be improved by removing bad words chosen by the lecturer (in our examples, 25%-40% of the words were removed).

All meta data of recorded lectures such as lecture titles, chapter titles, slide data, transcripts, subtitles and automated speech recognition (ASR) output were collected into a database as the prototype of the Collegerama Lecture Search engine. Evaluation of usability and reliability of this search engine resulted into the following conclusions:
• the best source of information for searching are human-made subtitles followed by ASR output
• chapter titles and slide content has a low importance for searching
• chapter titles and slide titles are only relevant for the generation of table of contents
• by clustering subtitles or ASR per slide, multiple-keyword searching is largely improved because of shorter viewing times in the search results (in our example lecture, it was reduced from 5.3 hours to 21 minutes)

A text-based database per course can also form as a basic container for a course discussion board, using time stamped remarks (“questions and answers”, discussions). A further extension to the database and search engine could be the adding of the other course materials, such as readings (books, lecture notes), activities (assignments, tests, lab tests) and practice exams.