伯特在处理短字符串时有多大用处?

假设我们处理的是一组短字符串,其中每个字符串由2~7个单词组成。

例如

 Assistant to VP    is much more similar to "Assistant" than "VP"
 Python Engineer    is much more similar to "Python Developer" than "Mechanical Engineer"
 Taking Care of Lost Cat    is much more similar to "Pet Care" than "Losing a Cat"

我想知道,对于这种类型的问题,使用什么样的预训练嵌入是最好的。我不太了解BERT,但我知道它是目前NLP的最新算法,所以我想知道它是否与这类问题相关。