25 Tweets to Know You: A New Model to Predict Personality with Social Media

Promises and Perils of Inferring Personality on GitHub

“…Likewise, by removing the markdown text indicative of images, which otherwise can be linked with personality, we reduce the chances of misclassification. Removing URLs also improved our inference, a factor identified in prior studies [3,5,11,49]. We believe that the improvement is attributed to the words in the URL which are otherwise misclassified as a part of the communication.…”

“…Another factor identified in the existing studies is the removal of numbers [3,5,49,51]. Generally, our analyses show that removing numbers improves the inferred personality scores, except for Yarkoni's extraversion and agreeableness.…”

“…We further incentivized participation by offering an Amazon gift card of 25 USD as prize. To mitigate privacy concerns, we informed our prospective survey respondents on our objective, implementation details, and their right to withdraw, in compliance with GDPR 5 and as approved by the University Ethics Board via Data Privacy Impact Assessment.…”

“…To systematically identify factors that can potentially influence the inferred personality score, we looked into four directions. First, we identified preprocessing steps used by the previous studies in software engineering [3,5,11,49,51]. Second, we searched for terms in LIWC dictionary that can be incorrectly represented in software engineering text.…”

“…We identified twelve preprocessing steps indicating (a) the influence of software engineering on the written text, (b) platformspecific features (e.g., Markdown), and (c) identifying/personal information. We identified removing (1) numbers [3,5,49,51], (2) hashtag [5,11], (3) URLs [3,5,11,49], and (4) @-references [3] from previous studies. Relating to the terms incorrectly represented in software engineering, we identified removing (5) quotes, (6) code blocks, and (7) images as part of Markdown characteristics.…”