Exploring public perceptions on alternative meat in China from social media data using transfer learning method

Abstract

The emerging social media serves as a complementary source for consumer behavior analysis with spontaneous data it generates. However, most studies employ time-consuming content analysis or lexical sentiment analysis. Considering the richness of data and progress of data science, in this paper, we propose a transfer learning based method to explore public attitudes towards alternative meat (AM) using data from social media in China to provide an alternative perspective. We compare traditional machine learning models: Naive Bayes and Support Vector Machine with our BERT-based Alternative Meat (BAM) model on the annotated sample. BAM model outperforms others in terms of macro F1 score and accuracy and is employed on the whole dataset later. The sentiment analysis result shows that among 41782 related posts we accumulated, about 42.10% of posts are personal posts consisting of negative, neutral, and positive feelings towards AM with a proportion of 28.77%, 22.91%, and 48.32% respectively. It is less promising compared with the consensus previous studies reach that over half of the Chinese people are positive and few Chinese are negative towards AM. Our findings add to the blooming body of studies suggesting the relationship of people’s willingness to try or purchase AM and factors including gender, geography, price, veganism, and food safety. Conspiracy theory is identified for the first time as the main reason for opposition to AM among Chinese consumers. Instead of the booster, traditional vegetarian substitutes especially tofu turn out to be an obstacle for accepting AM with much resemblances.