University of Twente Student Theses

Login

Author context extraction for interpreting the external validity of opinion mining results

Bloemen, O. (2013) Author context extraction for interpreting the external validity of opinion mining results.

[img]
Preview
PDF
4MB
Abstract:Nurtured by the seemingly ever-growing amount of user-generated content publicly available on the Internet, opinion mining is a growing �eld of interest. Applications show promising results in brand monitoring, where the public sentiment towards product features is observed, the monitoring of sentiment towards current issues in politics, or the prediction of voting outcomes. Despite the fact that the existence of opinion mining can be traced back to span at least a decade, the validity can be questioned. The majority of publications in this research area provide new techniques for text-analysis and/or incremental innovations for sentiment classi�ers. By using labeled corpora, these innovations are typically compared to a baseline method to indicate their superior accuracy in extracting features and determining the corresponding sentiment. For generalizability of the results, however, it is important to possess information about the context of the sample; i.e. information about the authors of the user-generated content. When this information is missing, the relation between the sample and the target population is unknown and conclusions cannot be drawn, possibly rendering opinion mining reports useless in practice. The purpose of this thesis is to investigate whether this problem related to the external validity of opinion mining, or more speci�c, the generalizability of opinion mining results with respect to the review authors, indeed plays a role. Three types of threats to the external validity are identi�ed from a literature review and examined: (1) a mismatch in demographic characteristics of the sample, (2) manipulation of online reviews, and (3) bias due to irrelevant experiences. Di�erent methods are proposed and tested to analyze the in uence of these biases on the sentiment report. Theoretical sampling con�rmed existence of both a demographic and an experience bias.
Item Type:Essay (Master)
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:85 business administration, organizational science
Programme:Business Administration MSc (60644)
Link to this item:http://purl.utwente.nl/essays/63600
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page