Elitist strategy of genetic algorithms for writing tang poetry

Abstract EN

Automatic Chinese Tang poetry composition arouses researchers' attention these years and faces a lot of challenges.

Most existing poetry generation systems can only generate poems without human interaction; thus, these poems cannot always express the human mind accurately.

To improve this disadvantage, this paper proposes a modified elitist genetic algorithm to generate poetry with arbitrary interaction from the user, which means that the user can specify the poem’s emotion and input words or verses to be used in the poem.

The modified algorithm comprises an improved elitist strategy to retain keywords or verses provided by the users, and a new concrete fitness function for more accurate and effective quality evaluation of poems.

The Turing test and fitness function contrast experiments show that the proposed algorithm could generate poems using given keywords or verse and the poems generated by the algorithm receive higher ratings and recognition than the original poems written by a human.

The experimental results demonstrate