<p>With the widespread use of smart connected devices that continuously evolve in terms of capabilities, today’s computing becomes increasingly pervasive. This gives birth to what’s called ubiquitous computing. The latter enables access to the underlying computing environment anytime and anywhere. One of the key components of such computing environments is context-aware devices and applications that adapt their behavior to entities’ contexts with minimal to no human intervention. This proactive and implicit adaptation provides accurate results concerning entities’ situations, which represent their actual context state. However, treating all contextual information or dimensions as relevant can become a bottleneck, potentially backfiring by adding noise to the results or excessively expanding the search space. This, in turn, makes it vast and combinatorial, leading to computationally intensive and time-consuming reasoning. This literature review examines and classifies approaches and methods for selecting relevant context. We examine context models used with these approaches to determine whether a specific or generic model is required. Identified approaches fall into three classes: retrieval, selection, and matching. Identified methods include semantic, heuristic, machine learning, and others. We synthesize evaluation practices and find retrieval is systems-driven, matching is decision-driven, and selection is evaluated by downstream gains. The review identifies gaps around relevancy drift and privacy and outlines research directions including drift-aware and cross-domain reuse. According to the defined inclusion and exclusion criteria, out of 277 screened studies, 41 relevant studies published between 2003 and 2024 were retained.</p>