This paper proposes a novel crossing-workflow fragment discovery mechanism, where an activity knowledge graph (AKG) is constructed to capture partial-ordering relations between activities in scientific workflows, and parent-child relations specified upon sub-workflows and their corresponding activities. The biterm topic model is adopted to generate topics and quantify the semantic relevance of activities and sub-workflows. Given a requirement specified in terms of a workflow template, individual candidate activities or sub-workflows are discovered leveraging their semantic relevance and text description in short documents. Candidate fragments are generated through exploring the relations in AKG specified upon candidate activities or sub-workflows, and these fragments are evaluated through balancing their structural and semantic similarities. Evaluation results demonstrate that this technique is accurate and efficient on discovering and recommending appropriate crossing-workflow fragments in comparison with the state of art’s techniques.