For gathering users' emotional reactions to web pages via a survey, a researcher could use either a Likert scale (e.g., Agree Strongly - Agree Somewhat - Indifferent - Disagree Somewhat - Disagree Strongly) or a semantic differential, (e.g., Love - Like - Indifferent - Dislike - Hate). Please note: I'm not talking about doing a usability test; this is about attitude surveys.

Why would a user researcher choose Likert instead of Semantic Differential or visa-versa?