Accession Number:

AD1009417

Title:

Individual Profiling Using Text Analysis

Descriptive Note:

Technical Report,15 Sep 2014,14 Sep 2015

Corporate Author:

University of Sheffield Sheffield United Kingdom

Personal Author(s):

Report Date:

2016-04-15

Pagination or Media Count:

17.0

Abstract:

Author profiling is the task of determining the attributes for a set of authors. This report presents the design, approach, and results of our approach to using data from the PAN 2015 Author Profiling Shared Task to predict personal attributes, as per the project brief. Four corpora, each in a different language, were provided. Each corpus consisted of collections of tweets for a number of Twitter users whose gender, age and personality scores are known. The task was to construct some system capable of inferring the same attributes on as yet unseen authors. Our system utilizes two sets of text based features, n-grams and topic models, in conjunction with Support Vector Machines to predict gender, age and personality scores. We ran our system on each dataset and received results indicating that n-grams and topic models are effective features across a number of languages.

Subject Categories:

Distribution Statement:

APPROVED FOR PUBLIC RELEASE