Slogans attract people’s attention by expressing the uniqueness and merits of an object briefly and effectively through short sentences. Because there are many slogan applications, the analysis and generation of slogans is considered a demanding task.
We presented a study of automatic slogan generation. The goal of our research was to apply statistical methods to NLP in order to automatically generate slogans that attract a user’s attention. To achieve this goal, we employed (1) the Japanese slogan corpus, (2) a large-scale N-gram corpus, and (3) preference information. Slogan generation demands creativity, which people involved in artificial intelligence are eager to provide. This study requires both a background in engineering (such as NLP) and cognitive science.
We investigated slogan features and proposed a generating system based on these acquired features. We constructed the largest Japanese slogan corpus at that time and we statistically analyzed it using NLP tools in terms of sentence structure and word appearance. Previous analyses reveal that slogans have distinct expressions and contain specific words in each theme. Using these results as given knowledge, we proposed a slogan generating system mimicking these features.
We focused on the indirect expressions that are unique to slogans. This was difficult in previous systems. Some slogans employ indirect expressions such as “An elephant step cannot break it!” rather than “This is very strong!” To produce this kind of expression, the proposed system acquires knowledge from the web and expands the scale of the slogan corpus.
We applied the indirect slogan generation system to real world problems. The proposed system generates candidate slogans using social networking sites. Preferable candidates can be selected using the bag-of-words model with linear regression. Subjective experiments have shown that the system-generated Slogans are comparable to human-selected slogans.
For more details, please click following link below.
Hiroaki Yamane and Masafumi Hagiwara, “Advertising slogan generation system reflecting user preference on the web,” Proceedings of the 2015 IEEE 9th International Conference on Semantic Computing (IEEE ICSC 2015)