DSpace Repository

Statistical Models for Linguistic Variation in Online Media

Show simple item record

dc.contributor.advisor Skiena, Steven en_US
dc.contributor.author Kulkarni, Vivek V. en_US
dc.contributor.other Department of Computer Science en_US
dc.date.accessioned 2017-09-20T16:52:15Z
dc.date.available 2017-09-20T16:52:15Z
dc.date.issued 2017-05-01 en_US
dc.identifier.uri http://hdl.handle.net/11401/77239 en_US
dc.description 115 pg. en_US
dc.description.abstract Language on the Internet and social media varies due to time, geography, and social factors. For example, consider an online chat forum where people from different regions across the world interact. In such scenarios, it is important to track and detect regional variation in language. A person from the UK, who is in conversation with someone from the USA could say ``he is stuck in the lift'' to mean ''he is stuck in an elevator'', since the word "lift" means an "elevator" in the UK. Note that in the US, "lift" does not refer to an "elevator". Modeling such variation can allow for applications to prompt or suggest the intended meaning to the other participants of the conversation. In this thesis, we conduct two related lines of inquiry focusing on (a) language itself and the variation it manifests and (b) the user and what we can infer about them based on their language use on social media. First, we develop computational methods to track and detect changes in word usage, including semantic and syntactic variation. We examine three modalities: time, geography and domains. Specifically, we outline methods to use distributional word representations (word embeddings) to detect semantic variation in word usage. Our methods are scalable to large datasets, making them particularly suited for social media. Second, we turn our attention towards users. In particular, we model latent traits of users based on their everyday language use on social media. We develop latent factor models, that explicitly seek to build representations of each user based on their inferred latent traits. These models capture latent traits that serve as useful co-variates for a wide variety of tasks like predicting what topics users like on social media and the number of friends in their social circle. This work has broad applications in several fields like information retrieval, semantic web applications, socio-variational linguistics, and computational social science including digital health care and ad-targeting. en_US
dc.description.sponsorship This work is sponsored by the Stony Brook University Graduate School in compliance with the requirements for completion of degree. en_US
dc.format Monograph en_US
dc.format.medium Electronic Resource en_US
dc.language.iso en_US en_US
dc.publisher The Graduate School, Stony Brook University: Stony Brook, NY. en_US
dc.subject.lcsh Computer science en_US
dc.subject.other Computational Social Science, Language change, Variational Linguistics, Word Embeddings en_US
dc.title Statistical Models for Linguistic Variation in Online Media en_US
dc.type Dissertation en_US
dc.mimetype Application/PDF en_US
dc.contributor.committeemember Balasubramanian, Niranjan en_US
dc.contributor.committeemember Schwartz, H Andrew en_US
dc.contributor.committeemember Bamman, David. en_US

Files in this item

This item appears in the following Collection(s)

Show simple item record

Search DSpace

Advanced Search


My Account