基于多层级指标清洗与聚合的科技创新能力评价研究

Research on Evaluation of Scientific and Technological Innovation Capability Based on Multi-level Indicators Cleaning and Aggregation

摘要:

在科技评价和科技管理须融合应用多源异构数据的发展趋势下，中国目前仍未形成普适的综合评价方法体系框架，且现有评估有关创新活动主体科技创新能力的方法无法分析指标间复杂的关联性，有必要构建能够提高数据可信度、实现可迁移性、适应多种算法的多级综合评价算法框架。鉴于此，提出基于多层级指标清洗与聚合的综合评价框架，包括数据处理层、指标聚合层和综合评价层三层算法，其中双流指标清洗算法根据指标相关性与指标数据间距离关系识别并修正数据中的异常点、极端值，可以提供高可信数据；而结合优劣解距离法的灰色关联法通过构建自适应评价算法，可以根据应用场景特点实现智能的指标聚合，从而克服现有方法在应用场景方面的局限性。基于此，依托深圳市科学技术创新委员会平台和载体专项项目（国际科技信息中心），通过政府官方渠道、调研访谈和次级数据形成研究数据资料，对2016－2021年珠三角地区214家主要科研事业单位，主要通过科技创新基础环境、科技创新产出能力、科技创新投入程度和科技项目承接能力4个一级指标及其二级指标进行科技创新能力综合评价。结果显示： 214家单位的科技创新能力近5年稳步增长，于2021年达到峰值，但总体存在较大的差异，其中科技创新投入程度提升显著，科技创新产出能力和科技项目承接能力也明显上升，但整体科技创新基础环境仍有待改善；此外，科技创新头部事业单位格局比较稳定，新兴头部事业单位的成长路径不同，宜根据自身特点强化优势补足劣势。可见，运用所提出的综合评价框架得到的结果具有较高的可比性、精确度和稳健性，可有效揭示珠三角地区不同创新主体的主要优势、发展态势、创新潜力以及薄弱之处。

Abstract:

In the emerging trend of integrating multi-source and heterogeneous data in scientific and technological evaluation and management, China lacks a universal framework for comprehensive evaluation methods. The current methods for evaluating the scientific and technological innovation capabilities of key players in innovation activities cannot analyze the complex relationships among indicators. Thus, it is essential to establish a multi-level comprehensive evaluation framework that enhances data reliability, supports transferability, and adapts to various algorithms. To this end, we propose an integrated evaluation framework based on multi-level indicator cleaning and aggregation. This framework comprises three algorithmic layers: data processing, indicator aggregation, and comprehensive evaluation. The dual-flow indicator cleaning algorithm identifies and corrects outliers and extreme values in the data by analyzing correlations and distance relationships between indicators, thereby providing highly reliable data. Additionally, the grey relational analysis (GRA) combined with the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS), constructs an adaptive evaluation algorithm that enables intelligent indicator aggregation according to the characteristics of the application scenario, overcoming limitations of existing methods in various contexts. Supported by the Project from Science and Technology Innovation Committee of Shenzhen-Platform and Carrier (International Science and Technology Information Center), research data were collected through official government channels, interviews, and secondary data sources. This dataset consists of 214 scientific public institutions in the Pearl River Delta from 2016 to 2021. The comprehensive quantitative evaluation of their scientific and technological innovation capabilities was based on four primary indicators: the foundational environment for scientific and technological innovation, innovation output capacity, innovation investment, and technology project execution capacity, along with their secondary indicators. The results show that the innovation capabilities of these 214 institutions have steadily increased over the past five years, reaching a peak in 2021, with substantial differences across institutions. There was a significant rise in innovation investment, innovation output capacity, and project execution capacity, but the overall foundational environment for scientific and technological innovation still requires improvement. Furthermore, the hierarchy among leading innovation institutions is relatively stable, with differing growth trajectories for emerging leaders. These differences suggest that strengthening advantages and addressing weaknesses according to specific characteristics would be beneficial. The outcomes derived from the proposed comprehensive evaluation framework offer high comparability, accuracy, and robustness, effectively revealing various innovation entities' main strengths, development trends, innovation potential, and weaknesses in the Pearl River Delta region.