A variety of hot topics lists released by social media platforms serve as a convergence and showcase for hot topic information, which provides significant insights toward our understanding of current popular discussions. However, due to vocabulary sparsity and short text length in hot list texts, traditional LDA and neural network-based topic mining models face poor performance in topic aggregation. To address these challenges, the paper proposes a topic modeling framework enhanced by a large language model—STAB, which combines the generative capabilities of large language models for text data with the excellent performance of document embeddings in topic modeling, enabling the extraction of meaningful topics from short text datasets. Experimental results on multiple datasets show that our framework outperforms existing topic modeling methods in terms of general objective evaluation metrics and applications in downstream tasks.

To address the challenges of passenger difficulty in hailing rides and insufficient vehicle supply during peak periods, this paper is based on an actual scenario in which ride-hailing platforms implement both surge pricing and subsidy strategies during peak periods, taking into account the heterogeneity of driver supply prices. By incorporating the reference point dependence theory of the driver decision-making and aiming to maximize platform expected profits, a Stackelberg game model is established to analyze the interaction between the platform and drivers. Kuhn-Tucker theorem is applied to determine the optimal service price and subsidy level for the platform during peak periods. The results reveal that the expected profit of platforms during the peak period showed an "inverted U-shape" trajectory as subsidies increased. When the subsidy level is within a critical range, the expected profits of the platform initially increase and then decrease with increasing service prices. The maximum expected profits of the platform are achieved when the critical value (peak point) is reached. By adopting appropriate pricing and subsidy strategies, ride-hailing platforms during peak period can effectively reduce the risk of driver attrition, adjust supply-demand relationships, and enhance the benefits for both the platform and drivers.

Recognizing the primary factors that influence information diffusion on social media platforms holds significant importance in the containment of harmful information spread. Previous research has primarily utilized regression analysis to identify variables that have a significant impact on retweets. However, these approaches have been limited in terms of interpretability. Using statistical modeling and causal inference, this study analyzes the variables that affect retweets from user and text features. Subsequently, the dose-response function is generated to elucidate the causal relationship of the text sentiment to retweets. Additionally, considering the potential collection bias in observed social media datasets, this study uses topical clustering for data filtration. In the experimental analysis of Twitter dataset related to the Vaccine discussion and presidential election, we have identified the variables that impact the retweets, and investigated the causal impact of text sentiment to retweets.

The overwhelming selection of movies on the market nowadays makes it difficult for users to make a decision. An efficient movie recommendation system plays a significant role in improving user experience and the market competitiveness of movie service providers. The challenge lies in how to integrate multiple data sources for personalized recommendations while balancing algorithm accuracy and diversity. Research on this issue is of great theoretical and practical significance. User portraits can depict rich user characteristics from multiple dimensions, helping us to better understand user interests and behaviors. Meanwhile, link prediction offers special benefits when modeling from a network topology standpoint. The integration of them provides a possibility to solve the above issues. Therefore, this study proposes a novel user portrait and link prediction-based personalized recommendation algorithm called UPLPR. The algorithm is designed under the background of movie recommendation. It distinguishes between the interest similarity that reflects in user behavior and genre domain. By abstracting user portraits from multiple data sources and integrating them into the link prediction process as external information of the network, the accuracy of the algorithm can be improved. Furthermore, from the perspective of scarcity, the algorithm improves the calculation of interest similarity between users in bipartite graph projection and evaluates the promoting or inhibiting effect of links in the recommendation process. Such consideration improves the novelty and personalization of the recommendation and mitigates the popularity bias problem to some extent. Finally, experiments were conducted on two MovieLens datasets to verify the proposed recommendation algorithm. Results show that compared with representative algorithms, the algorithm proposed in this paper not only achieved significant performance in accuracy but also demonstrated obvious advantages in diversity-related indicators. Additionally, the abstracted user portraits can help recommendation platforms understand their user base, thereby formulating more scientific marketing and management strategies.

In order to solve the problem of large group classification decision-making in social network environment, firstly, we construct a personalized similarity threshold learning model, integrate the similarity with social network, and get the modified social network; then, we use sub-network segmentation algorithm to group decision-makers, and compute the group preferences of subgroups through DeGroot model; next, we integrate the degree of consistency of the preference order, cohesion of subgroups and the number of their members in the process of aggregation of group preferences, and compute the optimal weight assignment of three indicators through parameter learning, and then compute the weights of subgroups. Secondly, in the process of group preference aggregation, we combine the degree of preference order consistency, subgroup cohesion and its number of members, and calculate the optimal weight assignment of the three indexes through parameter learning to compute the subgroup weights. Finally, the validity and feasibility of the method are verified by an example.

To reveal the direct and indirect impact of the business environment on the innovation efficiency of small and medium-sized enterprises (SMEs) in China, this paper takes 115 SMEs listed on the GEM from 2015 to 2019 as data samples, uses the entropy method to measure the indicators of the business environment of enterprises, and uses the SBM method to measure the overall innovation efficiency and stage innovation efficiency of SMEs. The Tobit regression model is constructed to analyze the impact of business environment on enterprise innovation efficiency and its sub-stage efficiency, two intermediary variables of digitization degree and innovate willingness are introduced and the step-up regression method and Bootstrap method are applied to investigate the transmission mechanism of business environment on enterprise innovation efficiency and its sub-stage, so as to explain the internal logic of business environment affecting enterprise innovation efficiency. The results show that: 1) The business environment has significant positive effect on the innovation efficiency of overall technological innovation and its knowledge transformation and achievement transformation stage of SMEs. 2) The indirect effects of business environment on the innovation efficiency of overall and its two sub-stage of SMEs are significantly positive through the mediating variables of digitization degree or innovation willingness. 3) The heterogeneity analysis by region, whether they are specialized and innovative and industry shows that, the business environment has a significant positive effect on the innovation efficiency and overall and its sub-stage of SMEs belonging to the eastern region and specialized new enterprises, and the business environment has a significant positive effect on the overall innovation efficiency of some manufacturing SMEs, and a significant negative effect on some manufacturing and service industries SMEs.

This study focuses on the prediction of credit risk for small and medium-sized enterprises (SMEs) by leveraging a comprehensive indicator system that incorporates unstructured textual information such asannual report texts and news reports. The Recursive Feature Elimination (RFE) method is utilized to select original indicators and indicators such as annual report text complexity, annual report sentiment tendency and news sentiment polarity for SMEs are incorporated. By utilizing Bayesian optimization-based XGBoost (BO-XGBoost) and other methodologies, the predictive performance of various machine learning models is compared across different sets of feature attributes. Furthermore, the SHAP (SHapley Additive exPlanations) interpretability method is employed to provide visual and comprehensive explanations of the model at both the local and global levels. The research demonstrates that the inclusion of unstructured textual feature indicators significantly enhances the predictive performance of the models, thereby highlighting the valuable predictive role of these features in assessing credit risk for SMEs. BO-XGBoost outperforms the baseline prediction performance, and the unstructured textual features rank highly in terms of importance. The SHAP waterfall plot, scatter plot, and dependence plot are used to explain the reasons for misjudgment cases, the polarity and degree of features impact on model's output, the evolutionary trends between unstructured textual features and credit risk. The empirical conclusions are further theoretically supported by principal-agent theory and other theories.

In the current volatile and uncertain international situation, the pharmaceutical supply chain faces many challenges. Pharmaceutical products are crucial for patient health and life, and any supply disruption will seriously affect the continuity of medical services and threaten patient safety, especially during the epidemic outbreaks. Therefore, pharmaceutical companies urgently need to build a resilient supply chain with the ability to avoid risks and quickly recover. Based on this, this paper defines the definition and structure of the pharmaceutical supply chain, and provides the definition and measurement indicator of pharmaceutical supply chain resilience. Then, this paper takes the single-product, multi-cycle pharmaceutical supply chain composed of the self-built factory/back-up supplier, pharmaceutical companies, and distributors as the research object. By introducing the back-up supplier as the resilient measure in the pharmaceutical supply chain, this paper constructs a pharmaceutical supply chain resilience model with the goal of maximizing profit under different disruption scenarios. The resilience level is characterized by the demand fulfillment rate of the distributors, and the shortage penalty cost is reflected in the total profits. Subsequently, this paper uses the supply chain of Pharmaceutical Company A as an example, and applies the fast non-dominated genetic algorithm to solve and analyze the model. Numerical experiments show that the introduction of the back-up supplier can not only improve the resilience level of the pharmaceutical supply chain, but also stabilize the total profits of the supply chain, helping the Pharmaceutical Company A respond quickly and recover rapidly to the initial state or even a more ideal state.

2023年12月2--3日第22届知识与系统科学国际研讨会(The 22nd International Symposium on Knowledge and Systems Sciences - KSS2023)在广州华南理工大学召开, 这是KSS会议在新冠肺炎疫情结束后首次线下活动, 会议的开、闭幕式, 主旨报告和"生成式人工智能和知识管理产业论坛"同时做了线上直播, 访问量累积超过7万人次. 遵从自2016年以来形成的范式, 会议全文投稿经评审录用由Springer出版了英文论文集, 并在会前上线(https://link.springer.com/book/10.1007/978-981-99-8318-6). 鉴于会议继续由中国的学术机构承办, 为扩展学术交流, 尤其是面向青年学生的研究成果的交流, 本次会议继续KSS2022的模式, 在第一轮全文投稿结束后, 在开放英文摘要提交的同时, 分别开通了英文期刊(Journal of Systems Science and Information, JSSI)及中文期刊《系统科学与数学》全文投稿, 即有英文摘要, 如能同期提交全文, 就相当于投稿期刊, 由KSS2023程序委员会评审. 第一轮评审结果反馈作者, 并注册会议作报告, 同时完成论文修改, 会上的交流更有助于作者进一步修改论文.
本次专题收录了投稿中文期刊并在KSS2023报告后最后录用的9篇文章, 主题涉及复杂系统建模、决策分析、在线媒体、意见挖掘、知识技术、知识管理与区块链等, 遍及会议10个平行分组中的7个. 这些工作来自多类项目资助, 包括国家重点研发计划项目"面向中小微企业的综合质量服务技术研发与应用". 第一作者主要是青年学者, 特别是研究生. 有的文章从开始投稿到最终录取经历了质的飞跃, 是作者的努力, 亦会议和期刊提供的平台的支持. 会议与期刊的直接联动增加了程序委员会的工作量, 但此工作模式上的创新有助于青年学者最新研究工作的公平和快速的发表, 让他们经历了规范的学术训练, 亦是KSS会议组织理念的一种体现.

Consumer satisfaction is essential for the production and operation of enterprises. In this regard, a framework SAFCS is proposed using word vectors derived from online reviews. Important nouns were selected from online reviews using contrastive attention based on BERT word vectors. The UMAP-PCA method was used for dimension reduction, and consumer satisfaction dimensions corresponding to the domain were obtained after clustering. Attribute-opinion phrases from online reviews were acquired via dependency parsing, and a pre-trained language model was utilized to achieve sentiment classification of the attribute-opinion phrases. Empirical analysis was conducted using reviews from four clothing brands: AT, GRN, LN, and TB. The results indicate that consumers have obvious characteristics in their attention to various dimensions, and at the same time, compared to negative reviews, consumers tend to conduct comprehensive evaluations when posting positive reviews. Finally, the study results provided guidance on the preferred production and operational approaches for the brands.