The Bidirectional Encoder Representations from Transformers (BERT) algorithm has become a valuable tool for learning common languages and has generated innovative outcomes on numerous projects. Social media stages offer a resource of user-generated information that can be analyzed to pick up bits of knowledge about human behavior and assumption. In this survey paper, we offer an outline of the application of the BERT calculation to a social media investigation. This paper is about the BERT algorithm and how its work is also included in preprocessing and tokenizing social media information. We also discuss the methods used to fine-tune BERT models for social media-specific datasets. We also investigated the diverse estimation approach and named substance acknowledgment utilizing BERT. At last, this paper presents the confinements of the BERT algorithm for social media investigation and long-standing time bearings for investigation in this field. By and large, this audit highlights the potential of the BERT calculation as an important device for social media examination and its capacity to supply unused experiences into human behavior and estimation.