In recent years, the availability of large amounts of data, advancements in technology, and the emergence of deep learning (DL) frameworks have significantly improved prediction accuracy in DL applications. DL applications can predict accurate results by use of these frameworks. Unfortunately, these frameworks are prone to inherent faults. These faults have been submitted in form of reports in the issues section of these framework repository websites (like TensorFlow). Since these reports are thousands in number, an automatic report extraction is preferred in place of manual fetching of issues (reports). In the current study, an Automatic Deep Learning Frameworks’ Fault Report Collection (DF-FRC) system with text pre-processing has been implemented. The implementation DF-FRC system extracts the following attributes like title, issue type, CUDA/cuDNN version, GPU model and memory, current behavior, standalone code to reproduce the issue, user comments, and many more attributes. Since user comments have been in natural textual format, text pre-processing has also been applied to them. This system upon collecting the reports observed that number of bugs present in open issues are 558 and closed issues are 7015 (from time duration 2020–2024) for TensorFlow DL framework. Further, upon application of text pre-processing, reduction of the collected report’s storage file size (by 20%) has also been observed which reduces the size of dataset and also increases the processing time of the dataset.