This study aimed to examine the semantic distinctions and differences in usage tendencies between the Korean temporal adverb near-synonyms jamkkan and jamsi through a corpus-based analysis and, based on these findings, to propose vocabulary teaching methods employing Data-Driven Learning (DDL). To this end, collocates of the two lexical items were extracted from a large-scale corpus, and their combinational patterns, usage contexts, and semantic characteristics were investigated through an analysis of authentic examples. The results showed that, although both jamkkan and jamsi denoted a short period of time, they displayed different semantic tendencies in actual usage. Jamkkan tended to occur in discourse-oriented and interactional contexts, frequently collocating with lexical items related to conversation, communication, and brief actions, whereas jamsi more often co-occured with words associated with emotional, cognitive, and psychological states, thereby expressing a temporary pause or suspension. On the basis of these analytical findings, this study proposed a semantic differentiation framework and a five-stage instructional design that enables learners to observe and analyze authentic language data and to infer the semantic differences between near-synonyms independently. This study was significant in that it strengthens the empirical foundation of near-synonym instruction by applying corpus-based collocational analysis to Korean vocabulary education and presented a learner-centered and sustainable approach to vocabulary learning.