Generate modern style Chinese poems based on common sense and evolutionary computation

Writing a poem is considered to be one of intelligent tasks that can be conducted by humans and that demand high competence on creativity, literary knowledge and common sense. This paper proposes an automated framework that can generate modern Chinese style poems under arbitrary topics. It is based on common sense knowledgebase ConceptNet and an evolutionary algorithm to search for a “best scored” poem. We design lexicon patterns and word segmentation sentence templates to compose lexicons into a modern style Chinese five-character verse. It turns out that after cleaning up ConceptNet noisy relations, the evolutionary algorithm can generate poems within 200 generations that make sense both syntactically and semantically under many different topics.