This study aims to compare and analyze the differences in the use of nominalization metaphors between DeepSeek-generated abstracts and human-written abstracts. Since its release, DeepSeek has demonstrated broad application potential in academic writing, yet its generated texts exhibit linguistic differences from human texts. Nominalization, as a metaphorical expression in systemic functional linguistics, significantly enhances the informational density of texts and serves as an effective means to present research findings in academic abstracts. Based on a self-constructed comparable corpus, this study explores the similarities and differences in nominalization metaphor usage between DeepSeek-generated and human-written abstracts from the perspectives of conceptual and interpersonal metaphors. The results reveal that DeepSeek-generated abstracts employ four nominalization strategies (derived nouns, gerunds, infinitives, and nominal clauses) more frequently than human abstracts, favoring more complex and diverse lexical and grammatical structures to convey information. Analysis of factors influencing nominalization shows that while DeepSeek excels at converting verbs and adverbs into corresponding nouns, it underperforms in adjective nominalization and uses significantly fewer personal pronouns than human authors. In conclusion, this study elucidates DeepSeek’s linguistic performance in academic writing tasks and its differences from human authors, providing insights for refining and optimizing generative AI technologies.