Playing multimedia documents in ubiquitous systems may require content adaptation based on gathered context information and accumulated historical data. Several approaches have already been proposed, in which adaptation actions are performed to provide adapted documents. Nevertheless, these approaches focus mainly on efficient use of context information without involving historical users data to improve the adaptation process. Thus, this paper allows for consideration of historical users data during the execution of the adaptation process. To do so, the context elements and the adaptation actions are first modeled using the oriented-object approach and then converted into relational and NoSQL databases schemes. Finally, algorithms for storing, retrieving and analysing data are designed. The proposal is validated by implementing scenarios through a real prototype. At a first step, the performances are measured to estimate the cost of data processing. The experiments show that NoSQL databases excel in data storage and ease of implementation, while relational databases perform well in data retrieve. At a second step, the proposal usefulness is highlighted by showing how historical data contribute to adaptation rules personalization using datadriven rule learning mechanisms rather than defining them explicitly. The analysis algorithm could retain personalized adaptation rules with confidence degree greater than 90%. Overall, the results are satisfactory.