Digital platforms are being utilized a growing amount frequently every day. Youngsters have a strong devotion to digital media, notably to social media platforms. These platforms exhibit a combination of messages and content, some of which could be abusive and affect young people’s learning environments. These days, youngsters predominantly comprehend slang phrases and utterances from these internet channels. This has an impact on both their moral evolution and cognitive blossoming. We have assembled a vernacular sentence identifi-cation approach in this study, especially for the transliterated Bengali language. Sentences possessing filthy phrases or terms are pinpointed operating the Bi-directional LSTM method. For the first time, We have introduced a remarkable Repulsive Levy random walk established Particle Swarm Optimization (PSO) algorithm to obtain the optimum scales of the hyper-parameters within Bi-directional LSTM. To assist the swarms update their location and attain the goal, Repulsive Levy is utilized for the first time in this study. As a result of the fitness score incentive distribution, which the swarms have never experienced before, they can now improve their velocity integrating the Repulsive Levy method. The proposed Repulsive Levy- PSO algorithm is compared to a few cutting-edge techniques, and it is evident from the comparison that the proposed method exhibits the optimal fitness score for the rastrigin objective function of 1 × +10-59 and thus outperforms the other methods.