Purchase this article with an account.
Russell L. Woods, Peter J. Bex, Daniel R. Saunders; Free Recall as an Objective Measure of Information Acquisition from Video. Invest. Ophthalmol. Vis. Sci. 2012;53(14):4831.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Information must be acquired from a video to understand its content. Visual acquisition of information from video is affected by degradation both of the video and of the observer’s vision. Currently, there is no widely-accepted method of assessing this information acquisition, which may be related to perceived video quality. We developed a new approach in which free recall responses to a video are used to measure the amount of information that was acquired.
After selecting 200 clips of 30s duration from Hollywood movies and nature documentaries, we prompted (1) normally-sighted viewers in the lab to verbally report; and (2) Amazon.com Mechanical Turk workers to write what they could recall immediately after watching each video clip. We collected responses from up to 32 different viewers per clip, which became the normally-sighted baseline. Information acquisition of a new viewer can then be estimated by comparing their response with the baseline responses to the same clip. Several methods for performing this comparison were evaluated, including latent semantic analysis and related models and a simple metric based on the number of words shared between the new response and the baseline responses. These metrics were evaluated by how well they could match responses to the video clips they described.
The shared-word similarity metric had the best performance, with 95% accuracy at matching a response to its video clip, and so was chosen as the primary means of scoring responses. Participants had a wide range of ages and education levels, making the measure applicable to many different populations. They produced a substantial body of text for each clip, median length 33 words after removing common words. There were no significant effects of age, gender or education level once the shared-word metric was adjusted for the number of words in the response.
We present a novel, objective measure of information acquisition from video and validate its results. The method could be used to evaluate both the impact of low vision on viewing video and the effects of video enhancement techniques for people with low vision.
This PDF is available to Subscribers Only