关键词: 决策智能, 政策信息学, 机器学习, 智库问题, 主题建模

Abstract:    Recent years have seen a rise in the field of policy informatics. Think tank problem analytics is a typical approach in policy informatics, which often involves multi-discipline and multi-field knowledge computing. Large-scale, cross-disciplinary opinion gathering and literature information processing serve as the initial building blocks of policy analytics. The policy simulations, case studies, empirical analysis and scientific discoveries in journal papers can provide professional analysis for decision intelligence. This paper seeks to apply a topic modeling tool on parsing the journal abstracts of the think tank challenge "green innovation". By using a Bayesian hierarchical topic model called LDA, it is possible to find the latent themes that underlie huge collections of documents. We use the LDA to analyze 6891 journal paper abstracts on the topic of "green innovation" as an example, which produces a subset of problems. The English abstracts of 6098 papers were retrieved from the Web of Science core database, while the Chinese abstracts of 793 papers were retrieved from the CNKI core collection, and both sets of abstracts were downloaded.
    At various sizes, decision-making processes depend on expert opinions. Over time, scholars have created and employed a variety of techniques for gathering, analyzing, and aggregating expert opinions. Because language is the primary characteristic that sets humans apart from other species, and language is the primary means by which human culture, ideas, and wisdom are transmitted, natural language processing serves the most fundamental component of artificial intelligence. This makes it possible to use NLP to provide pathways for decision intelligence. 
    The exact procedures of this ML assisted decision intelligence method are: 1) The stakeholders identify the research area and the researcher creates a list of search phrases accordingly. 2) The researcher downloads academic papers (or abstracts). 3) Calculate the degree of confusion of the number of topics, and find the most appropriate LDA topic number. 4) The LDA model analyzes the text to create multiple word bags. 5) The professionals read each topic and deduce the topic labels. Each topic represents a subset of the think tank challenge, which can be parsed repeatedly using language processing models to ultimately create a framework for an understandable and workable solution. 
   Themes including "economic determinants and high-quality development", "spatial spillover of green innovation and regional development" and "corporate relationships and supply chain networks" are the main results of the parsing of Chinese literature. In contrast, "ecological and environmental governance" and "carbon dioxide emissions and renewable energy" are given substantially more weight in the English LDA output than they are in the Chinese LDA output. The topics "technology costs of enterprises and government subsidies", "innovation policies and sustainable development", "efficiency of science and technology innovation", "green financial markets and enterprise financing", "corporate competition and social responsibility", and "economic development and environmental regulation" can be found in both the English and Chinese literature parsing results. The amount of Chinese literature on green innovation is generally significantly less than the amount of English literature. Overall, the LDA topics derived from the abstracts of the English and Chinese literature are appropriate for future research.
    One of the contributions of this study is that it analyses the academic literature collection database to provide knowledge extraction and human-computer collaborative decision making for think tank challenges, which bridges the gap between the academia and other policy stakeholders (e.g., politicians, business practitioners, news reporters, etc.). Natural language processing methods, such as the LDA model can handle rich textual information and retain more information in the process of downscaling textual data. A number of additional studies using LDA in literature analysis have emerged in recent years, with some researchers using LDA as an intelligent literature review approach and others using LDA to identify research hotspots and follow new trends. This study presents a novel approach for improving evidence-based decisions and expanding the field of policy informatics research.
    Instead of being the outcome of some individual′s careful deliberation to arrive at a decision, policy making is rather a process of mutual communication and coordination among numerous interested organizations, groups, individuals, including the policy makers themselves. Text analysis offers a think-tank problem analytical approach with machine learning models, which converts a significant amount of complex texts into simple, understandable, and meaningful topics. It is a clever approach for policy analysis, because it can help the stakeholders to communicate and work together to enable decision-making that is based on expert wisdom.

Key words: decision-making intelligence, policy informatics;machine learning, think tank problem, topic modeling