Fast and Effective Approximations for Summarization and Categorization of Very Large Text Corpora
نام عام مواد
[Thesis]
نام نخستين پديدآور
Godbehere, Andrew B.
نام ساير پديدآوران
El Ghaoui, Laurent
وضعیت نشر و پخش و غیره
تاریخ نشرو بخش و غیره
2015
یادداشتهای مربوط به پایان نامه ها
کسي که مدرک را اعطا کرده
El Ghaoui, Laurent
امتياز متن
2015
یادداشتهای مربوط به خلاصه یا چکیده
متن يادداشت
Given the overwhelming quantities of data generated every day, there is a pressing need for tools that can extract valuable and timely information. Vast reams of text data are now published daily, containing information of interest to those in social science, marketing, finance, and public policy, to name a few. Consider the case of the micro-blogging website Twitter, which in May 2013 was estimated to contain 58 million messages per day: in a single day, Twitter generates a greater volume of words than the Encyclopedia Brittanica. The magnitude of the data being analyzed, even over short time-spans, is out of reach of unassisted human comprehension. This thesis explores scalable computational methodologies that can assist human analysts and researchers in understanding very large text corpora. Existing methods for sparse and interpretable text classification, regression, and topic modeling, such as the Lasso, Sparse PCA, and probabilistic Latent Semantic Indexing, provide the foundation for this work. While these methods are either linear algebraic or probabilistic in nature, this thesis contributes a hybrid approach wherein simple probability models provide dramatic dimensionality reduction to linear algebraic problems, resulting in computationally efficient solutions suitable for real-time human interaction. Specifically, minimizing the probability of large deviations of a linear regression model while assuming a
نام شخص به منزله سر شناسه - (مسئولیت معنوی درجه اول )