User-oriented document summarization through vision-based eye-tracking
Comments:
Cindy Skach
Luke Roberts
Reference:
Xu, S., Jiang, H., Lau, F. User-oriented document summarization through vision-based eye-tracking. IUI '10. ACM, New York. http://delivery.acm.org/10.1145/1510000/1502656/p7-xu.pdf?key1=1502656&key2=6714922031&coll=DL&dl=ACM&ip=165.91.4.135&CFID=17152168&CFTOKEN=47763320
Summary:
The researchers seek to create an algorithm that allows eye-tracking to aid in summarizing documents for users. They go about this by estimating the average time spent on a single word in the documents the user is reading and extrapolate that data into the likely-hood the user will find a whole sentence interesting in summing up the probability the user will be interested in the single words of that sentence (as well as similar words). A regular web-cam is used and it is calibrated by their own document viewer they made. Once this is done the user is off to reading.
The researchers compare their results of summarization to two popular methods of summarization - Microsoft Word AutoSummarize and the MEAD summerizer system. The experiment involved using sets of literature from science and leisure. The results showed that this algorithm tailors better to the users preferences as compared to the other two.
Discussion:
After reading the abstract I was very excited to read this article. I thought it interesting that they were doing the algorithm based on single word time rather than something else such as time over a sentence in order to summarize the context of something. I also was unsure of how the summary would go: would it be in whole sentences or just keywords? I then found that it was because of the personal aspect of the algorithm, it is implemented in order to list in order the sentences that would be most preferable for the user to read.
Not only is the idea behind this paper really interesting, but the way in which they approached the problem was really cool. Also the fact that users rated their algorithm above the other two popular ones is a bolster to the credit of this report. The paper itself was not extremely thorough in explaining all of the details of it all, but I am satisfied with that if I can go over the article and not get bored to tears over every inch moved by their finger.
Future projects that I could envision for this research would be possible linking this algorithm across different avenues of software in order to gain a greater understanding of the user, similar to how Google knows everything about us by our searches. Through our web-cams, maybe it could also know all about what we are looking at.
This research is very cool. Currently Google studies eye tracking to understand how people evaluate Google results. They hope to use it to change how they order result hits.
ReplyDeleteI was also really excited about this paper - especially with all of the reading we've been doing this semester. :) I think one of the most interesting areas of future work that they mention in the paper is how they hope to make it capable of creating user-oriented summaries on articles the user has not even read.
ReplyDeleteGoogle already has enough of my personal information, I don't want them tracking what my eyes are looking at too! Haha, but Joe is right, Google does use eye-tracking as a research tool to understand a user's interest. Something interesting they found is that most people will skip the top result from a query.
ReplyDeleteYeah I'm not sure I really want anything tracking what my eyes are looking at. While I can see how this would be useful in a test environment, I think it would be too invasive for the real world. I could see this pushing people over the edge regarding how invasive the internet is.
ReplyDelete