Many analyses in the ATLAS physics program at the LHC are dependent on the identification of jets containing $b$-hadrons ($b$-tagging). The corresponding algorithms are referred to as $b$-taggers. The baseline $b$-taggers are optimized for jets containing one $b$-hadron. A new double $b$-tagging algorithm, the $X\rightarrow b\bar{b}$ tagger, provides better identification efficiency to reconstruct boosted resonant particles decaying into a pair of $b$-quarks. In the boosted regime, it is a challenging task because of high collimation of the two $b$-hadrons. This neural network based $X\rightarrow b\bar{b}$ tagger uses the kinematic information of the large radius ($R=1.0$) jet and the flavour information of associated track-jets. The performance of this tagger was evaluated using Monte Carlo simulation, therefore it could vary in collision data. Thus this poster presents the in situ tagging efficiency calibration using $Z\rightarrow b\bar{b}$ events with a recoiling photon or jet for this boosted $X\rightarrow b\bar{b}$ tagger. The efficiency data to simulation scale factor is derived using the Run 2 $\textit{pp}$ collision data collected by the ATLAS experiment at $\sqrt{s}=13$ TeV, with the integrated luminosity of $139\,\text{fb}^{-1}$.