关键词: 农产品地理标志, 自然语言处理, 政策注意力, 信息抽取, 文本分类

Abstract: [Purpose/Significance] Geographical indications (GIs) are an important tool for local governments in China to carry out brand building of agricultural products. Brand management is a continuous systematic project involving multiple subjects. Among them, the problem of government policy attention in the field of brand management of agricultural products with GIs deserves in-depth study. This paper aims to construct a policy attention analysis model in the field of brand management of GI agricultural products based on natural language processing technology. This model provides technical support for local governments to explore the status quo of local GI agricultural products' brand management, analyze the distribution of policy attention, and assist in optimizing strategies to promote their products' brand development. [Method/Process] The study is focused on the distribution of attention paths of GI agricultural products' brand management policies from the perspective of the local government: an analysis model of brand management policies of GI agricultural products was constructed in order to support the local government to carry out the analysis of the status quo of local agricultural products' brand management and policy optimization, and provide decision-making support for the optimization of brand management measures of the local agricultural products. First, this paper built a basic corpus based on the Python crawler technology, collected authoritative public information on the Internet, utilized the domain dictionary and UIE general information extraction framework to extract the text of management measures published in local government policies, and built a database of brand management measures of GI agricultural product. Second, a classification model of brand management measures of GI agricultural products based on the Transformer model was constructed. Third, this paper built a classification model based on the Transformer model, which can classify the extracted brand management measures of agricultural products and construct the policy attention distribution map. Finally, based on the policy attention distri-bution given by the model, we can find the brand management bottlenecks and recommend countermeasures to solve the bottlenecks. [Results/Conclusions] This paper takes Yantai apples as an example for model validation. After extracting and categorizing Yantai apples' brand management data, it is found that the policy attention of Yantai apples is more con-centrated, and the measures are highly similar, with 41.1% of the text of the measures focused on the part of brand positioning and planning, 31.7% on the part of brand competitiveness enhancement, and less than 10% on the part of brand marketing and protection. It can be seen that the brand effect of Yantai apples with GIs has not been well utilized.

Key words: geographical indications of agricultural products, natural language processing, policy attention, information extraction, text classification