This paper presents a novel Dynamic Task Assessment (DTA) module to improve knowledge transfer for multi-task learning (MTL). The DTA module provides local and global task similarities by considering various metrics, such as data distribution and task complexity features, and more domain-specific measures. Afterward, the tasks are grouped in a hierarchy using an algorithm and can be monitored in real time to minimize any negative transfer. Validation on some diversified datasets: The model performances were validated across three diverse data sets are MNIST & Fashion-MNIST, Multi-Domain Sentiment Analysis and Omniglot. We demonstrated that our DTA module brings statistically significant improvements in task performance, with improved accuracy of 7.5% on MNIST & Fashion-MNIST, 6.8% on Multi-Domain Sentiment Analysis, and 5.3% on Omniglot, for instance. The transfer efficiency was much higher, showing up to 15% enhancements compared to conventional methods. Another factor to note is that the DTA module also decreases negative transfer rates by 18% on MNIST & Fashion-MNIST and 22% on Multi-Domain Sentiment Analysis and Omniglot. Higher Silhouette Scores indicated that the clustering quality also improved for all datasets. These results imply that a DTA module could be an efficient and scalable method to facilitate knowledge transfer in MTL, which can be applicable across tasks requiring dynamically adaptive learning strategies.