A Multi-Modal Chinese Poetry Generation Model

TL;DR

This paper introduces a multi-modal Chinese poetry generation system that creates poems from images, titles, and initial lines using a hierarchical attention model and LDA for improved relevance and coherence.

Contribution

It presents a novel three-stage multi-modal generation framework with a hierarchy-attention seq2seq model and LDA for title relevance, advancing Chinese poetry generation capabilities.

Findings

Outperforms baseline models in coherence and relevance

Effective in generating complete poems from images and prompts

Validated by both machine and human evaluations

Abstract

Recent studies in sequence-to-sequence learning demonstrate that RNN encoder-decoder structure can successfully generate Chinese poetry. However, existing methods can only generate poetry with a given first line or user's intent theme. In this paper, we proposed a three-stage multi-modal Chinese poetry generation approach. Given a picture, the first line, the title and the other lines of the poem are successively generated in three stages. According to the characteristics of Chinese poems, we propose a hierarchy-attention seq2seq model which can effectively capture character, phrase, and sentence information between contexts and improve the symmetry delivered in poems. In addition, the Latent Dirichlet allocation (LDA) model is utilized for title generation and improve the relevance of the whole poem and the title. Compared with strong baseline, the experimental results demonstrate the effectiveness of our approach, using machine evaluations as well as human judgments.