本文为德国中部黑森工业大学（作者：Christian Schulze）的学士论文，共57页。

需要经验学习来解决的非确定性问题，这些算法已经在邮政服务中的字符识别或自动电话系统中的文本语音转换等问题上使用了多年。最近的发展使这些算法能够以更好的性能解决更复杂的问题，比如图像中物体的识别。因此，机器学习领域允许我们通过处理近似值来自动化没有算法的过程。主观感知的电影氛围总是依赖于人类对电影的经验、记忆等复杂因素的评判，通过人与人的互动来学习。然而在某种程度上，主体间的标准可以用机器学习算法来近似，因此描述电影氛围的过程是可以学习的。

Algorithms that learn with experience solving non-deterministic problems where there is no classical algorithm because the problem is too complex or we have no algorithm for it yet have been used for years now on problems like character recognition in postal services or text-to-speech in automated phone systems. Recent developments allow it to tackle even more complex problems with better performance like recognition of objects in images. Hence the field of Machine Learning allows us to automate processes where no algorithm exists solely by working with approximations. Subjectively perceived atmospheres of movies always rely on experiences, memories and other complex factors of the human judging the movie, learned by human interaction. However to some degree an inter-subjective standard might be approximated by Machine Learning algorithms and therefore the process of describing atmospheres of a movie can possibly be learnt.

1.  引言

2.  相关工作与概念

3.  理论背景

4.  具体实现

5.  结论

附录A实验结果

附录B有关实施中使用的第三方工具信息

更多精彩文章请关注公众号：