#machine reading

[gauloz]

Transfer learning for machine reading comprehension - Microsoft Research

Microsoft Research is making progress ensuring a scalable solution to extend machine reading comprehension to a wider range of domains.

[autolovecraft]

Apologies for interrupting your feed with non-automatically generated content that's not directly related to the administration of this blog. However. It occurred to me that some of you might be interested not just in automatically generated Lovecraft-like stories, but in automated readings of Lovecraft's stories. In this post on my research blog, I describe some experiments with do so using a technique called latent Dirichlet allocation.

That's all. Back to your lives. Sorry for interrupting if this isn't interesting to you. I'll always keep this kind of post to a minimum.

[filmvis]

Film Classification by Trailer Features

Write up of another student project from Stanford, this one using "Mean Frames," "Scene Variation," and "Face Recognition" to attempt to machine read and classify movie trailers.

Description:

The Matrix, a sci-fi film released in 1999, was famous for telling viewers in its trailer that: “Unfortunately, no one can be told what the Matrix is. You have to see it for yourself.” And film-goers did, resulting in over $460 Million worldwide box office gross, four Academy Award wins, and an 87% critic approval rate (according to Rotten Tomatoes). The decision to see any film is based on many factors: cast, critical opinion, recommendations of our friends; but often it’s simply because we like the trailer. So we ask the question, what information can a machine extract from trailers?

For this paper, we have collected a set of trailers (312 with video, 100 with subtitles), extracted features from the video feeds and the subtitle texts, and attempted to classify the genre and MPAA rating of each film.

pdf at http://cs229.stanford.edu/proj2012/HelmerJi-FilmClassificationByTrailerFeatures.pdf