Towards improving coherence and diversity of slogan generation

Abstract

Previouswork in slogan generation focused on utilising slogan skeletons mined from existing slogans. While some generated slogans can be catchy, they are often not coherent with the company’s focus or style across their marketing communications because the skeletons are mined from other companies’ slogans. We propose a sequence-to-sequence (seq2seq) Transformer model to generate slogans from a brief company description. A naïve seq2seq model fine-tuned for slogan generation is prone to introducing false information. We use company name delexicalisation and entity masking to alleviate this problem and improve the generated slogans’ quality and truthfulness. Furthermore, we apply conditional training based on the first words’ part-of-speech tag to generate syntactically diverse slogans. Our best model achieved a ROUGE-1/-2/-L $\mathrm{F}_1$ score of 35.58/18.47/33.32. Besides, automatic and human evaluations indicate that our method generates significantly more factual, diverse and catchy slogans than strong long short-term memory and Transformer seq2seq baselines.

Effective Slogan Generation with Noise Perturbation

“…Following the previous studies [4,8,13], our transformer-based model uses the same data format which includes descriptions about the firm or target brands as input and slogan as output. The authors of [8] complied such descriptions and slogans by web crawling based on a dataset comprising company details(i.e. domain, industry, and linkedin url).…”

“…We adopt [8,13] as our baselines, representing non-PLM and summarization PLM respectively. The first baseline [13] introduces a reconstruction model based on compressed representation to generate slogans with distinctiveness.…”

“…Traditionally, to generate a slogan, human experts go through the process of brainstorming, generating and evaluating candidate slogans, requiring a lot of resources and manpower. Automation in generating candidate slogans can alleviate a significant amount of labor and resources [2,8,13,19]. Recognizing the importance of automatic slogan generation, prior researchers compiled different datasets and developed algorithms applicable to the datasets.…”

“…In particular, [2,14,19] explored slogan generation in syntactic perspective; they focused on filling in a well-informed sentence skeleton extracted from a slogan database with suitable words. Another line of research utilized slogan descriptions [3,25], while recently, [4,8,13] adopted seq2seq(sequence-to-sequence) models for automatic slogan generation, using descriptions of firms as inputs.…”

“…Several prior studies highlight the importance of creativity and diversity as key factors for creating memorable and likeable slogan. For instance, [8] proposed generating syntactically diverse slogan while maintaining the truthfulness to the source text. To achieve these two characteristics, the method finetunes an abstractive summarization model initialized on CNN/DailyMail dataset [7] with syntactic control.…”