Generating natural-language explanations is an important task in explainable recommendation, as it helps users understand why an item is recommended. However, existing generative methods typically encode user, item, and recommendation signals as entangled conditioning inputs, leaving explanation-relevant structure to be inferred implicitly by the language model. This leads to suboptimal explanation generation performance.In this paper, we propose a structured explanation generation framework for explainable recommendation, which explicitly models explanation-relevant information instead of leaving it fully entangled in recommendation signals. Specifically, the framework decomposes explanatory preferences into two complementary components: general explanatory preferences, which capture domain-shared explanatory aspects and their importance weights, and personalized explanatory preferences, which capture residual user- and item-specific explanation signals beyond the shared aspect space. Based on this decomposition, we organize the learned explanatory representations into a structured prompt for explanation generation, and further introduce a regularization mechanism to improve explanation generation performance. Experiments on three benchmark datasets with different language-model backbones demonstrate that the proposed framework consistently improves recommendation explanation generation. The results highlight the benefit of explicitly modeling explanation-relevant structure for generative explainable recommendation.

The analytic hierarchy process (AHP) is a structured technique used to analyze complex decision-making situations such as resource allocation, benchmarking, and quality management. In the weight valuation step of using AHP to select the best design, pairwise comparison matrices are used to calculate the local priorities for designs that have contentious and unresolved criticisms. In this study, we propose a Bayesian approach using a Dirichlet-multinomial model to estimate local priorities during weight valuation. Experts are only asked to select the best design with respect to predetermined criterion. Subsequently, local priorities are estimated without pairwise comparison matrices. To improve the efficiency of the AHP, we propose two expert allocation policies (AHP-KG and AHP-AKG) based on the ranking and selection procedures. Our numerical results show that the proposed AHP-KG and AHP-AKG policies outperform pure exploration and proportional allocation policies.

User interests are highly individualized and evolve over time. Nevertheless, the exploration of these personalized traits alongside the dynamics of such interests is less studied. Based on the observation that the sequences of a user's reviews of verified purchases on a product can reflect the user's dynamic interests toward the product type, we propose a time-series-based framework that elucidates both the evolutions and characteristics within user interests, learns the personalized patterns in interest series, and suggests three strategies to predict users' prospective interests. The experiments used 1.29 million Amazon's reviews of verified purchases on electronic products to validate the proposed framework. The results showed that a part of the users' interest series did have unique life-cycles, and the proposed prediction strategies outperformed both traditional static method and recent deep sequential models in terms of precision, recall rate, and F1, which indicated that "personalization" and "time" indeed played critical roles in user interest analysis. The proposed approach can be used to enhance predictions regarding users' prospective interests while mitigating information overload. This work holds potential for a variety of applications such as user profile construction, precision marketing, and personalized recommendation.

In credit evaluation, feature selection and grouping effect analysis are used to identify the most relevant credit risk features. Most feature selection and grouping effect analysis are implemented via regularizing linear models. Nevertheless, substantial evidence shows that credit data are linearly inseparable due to heterogeneous credit customers and various risk sources. Although many nonlinear models have been proposed in the last two decades, the majority of them required recombination of the original features, which made it difficult to interpret the results of the models. To cope with this dilemma, we propose a diagonal distance metric learning model that improves distance metrics by rescaling the features. Meanwhile, feature selection and grouping effect analysis are realized by adding regularizations to the model. The main merit of the proposed model is that it avoids the limitation of the linear models by not pursuing linear separability, yet guaranteeing the interpretability. We also prove and explain why feature selection and grouping effect can be achieved and decompose the optimization problem into parallel linear programming problems, plus a small quadratic consensus-reaching problem, such that the optimization can be efficiently solved. Experiments using a real credit data set of 96,000 instances show that the proposed model improves the area under the receiver operating characteristic curve (AUC) of the distance-based classifier k-nearest neighbors by 14% in two-class credit evaluation and surpasses linear models in terms of accuracy, true positive rate, and AUC. The proposed regularized diagonal distance metric learning approach also has the potential to be applied to other fields where data are linearly inseparable.

Irregular loan transaction behaviors and poor market situation indicate potential risk for wholesale and retail small and medium-sized enterprises (SMEs) in loan use and operation, but the value of these risk information is rarely explored and utilized. This paper innovatively proposed to mine such information to solve the difficulty of default prediction for wholesale and retail SMEs and enrich the credit risk research of special groups. The experimental results showed that tracking loan transaction and market evaluation data can improve the accuracy and recall of the models by an average of 1.38% and 7.66%, respectively, and identify default samples 233.22 days in advance on average, which can provide a new perspective for financial institutions to predict the default risk.

Nowadays, the malicious MS-Office document has already become one of the most effective attacking vectors in APT attacks. Though many protection mechanisms are provided, they have been proved easy to bypass, and the existed detection methods show poor performance when facing malicious documents with unknown vulnerabilities or with few malicious behaviors. In this paper, we first introduce the definition of im-documents, to describe those vulnerable documents which show implicitly malicious behaviors and escape most of public antivirus engines. Then we present GLDOC-a GCN based framework that is aimed at effectively detecting im-documents with dynamic analysis, and improving the possible blind spots of past detection methods. Besides the system call which is the only focus in most researches, we capture all dynamic behaviors in sandbox, take the process tree into consideration and reconstruct both of them into graphs. Using each line to learn each graph, GLDOC trains a 2-channel network as well as a classifier to formulate the malicious document detection problem into a graph learning and classification problem. Experiments show that GLDOC has a comprehensive balance of accuracy rate and false alarm rate - 95.33% and 4.33% respectively, outperforming other detection methods. When further testing in a simulated 5-day attacking scenario, our proposed framework still maintains a stable and high detection accuracy on the unknown vulnerabilities.

Profit-driven artificial intelligence (AI) systems and profit-based performance measures are widely used in credit scoring. When assessing the performance of an AI system for credit scoring, previous research typically assumes that the cost and benefit parameters and their distributional information are available. In reality, however, these parameters and their distributions are often not precisely known. This study considers parameter uncertainty in the development of credit-scoring models and the estimation of profits and risks generated by those models. We propose a novel profit-based metric-the worst-case expected minimum cost (WEMC)-to estimate the profit of credit-scoring models with uncertain parameters. Furthermore, we introduce the worst-case conditional value-at-risk (WCVaR) metric to measure the loss incurred from employing a classification model in credit scoring under the deterioration of cost parameters. A multiobjective feature -selection framework based on WEMC (or minimum cost) and WCVaR is then presented for model development. Using a comprehensive bankruptcy database, we compare the proposed methods with wrapper methods that use traditional metrics as selection criteria, as well as filter and embedding methods. We conduct extensive experiments to evaluate the economic benefits of the proposed methods under different scenarios that simulate dynamic changes in macroeconomic conditions. The results suggest that the proposed methods outperform other feature-selection methods in the aspects of profit and risk performance metrics in most cases.

Real-world information is often characterized by uncertainty and partial reliability, which led Zadeh to introduce the concept of Z-numbers as a more appropriate formal structure for describing such information. However, the computation of Z-numbers requires solving highly complex optimization problems, limiting their practical application. Although linguistic Z-numbers have been explored for their computational straightforwardness, they lack theoretical support from Z-number theory and exhibit certain limitations. To address these issues and provide theoretical support from Z-numbers, we propose a Z-number linguistic term set to facilitate more efficient processing of Z-number-based information. Specifically, we redefine linguistic Z-numbers as Z-number linguistic terms. By analyzing the hidden probability density functions of these terms, we identify patterns for ranking them. These patterns are used to define the Z-number linguistic term set, which includes all Z-number linguistic terms sorted in order. We also discuss the basic operators between these terms. Furthermore, we develop a multi-criteria group decision-making (MCGDM) model based on the Z-number linguistic term set. Applying our method to predict the acceptance of academic papers, we demonstrate its effectiveness and superiority. We compare the performance of our MCGDM method with five existing Z-number-based MCGDM methods and eight traditional machine learning clustering algorithms. Our results show that the proposed method outperforms others in terms of accuracy and time consumption, highlighting the potential of Z-number linguistic terms for enhancing Z-number computation and extending the application of Z-number-based information to real-world problems.

This repository provides an implementation of the regularized diagonal Distance Metric Learning (DML) model, which improves distance metrics, selects features, and conducts grouping effect analysis by rescaling the features. One characteristic of the proposed model is that it does not pursue linear separability, which is highly unrealistic in financial data. Another characteristic of the proposed model is that it considers correlated features when conducting feature selection, and thus, does not neglect important credit risk sources when used for credit evaluation. The implementation of the solver based on the Alternating Direction Method of Multipliers(ADMM) makes it suitable for large-scale financial applications. The repository also provides the scripts, data, and experimental results reported in the paper. This repository includes four folders, src, scripts, data, and results.

In this rapidly evolving era of multimodal generation, diffusion models exhibit impressive generative capabilities, significantly enhancing the realm of creative image synthesis by intricately textual prompts. Yet, their effectiveness is limited in certain niche sectors, like depicting Chinese ancient architecture. This limitation is primarily due to the insufficient data that fails to encompass the unique architectural features and corresponding text information. Hence, we build an extensive multimodal dataset capturing the essence of Chinese architectures mostly from the Tang to the Yuan Dynasties. The dataset is categorized on the types, including image&text, video, and style models. In details, images and videos are methodically categorized based on locations. All images are annotated at two levels: initial annotations and descriptive terms based on distinctive characteristics and official information. Moreover, seven artistic styles fine-tuning models are provided in our dataset for further innovations. Significantly, this is the first Chinese ancient architecture dataset and the instance of using the Pinyin system to annotate unique terms related to Chinese architectural styles.

Group decision making (GDM) is normally resource consuming and requires a moderator to lead a group of experts to achieve consensus. A moderator’s preference on consensus level affects the cost of GDM and has important impacts on the consensus results. However, no previous research has considered the moderator’s preference in consensus. The objective of this article is to analyze the effect of the moderator’s preference on consensus result in GDM. We develop a set of optimization models and propose a bilateral compromise consensus-reaching framework, which provides a flexible way to reach an agreement. We apply the models to a loan consensus problem under a peer-to-peer lending environment. The results show that a moderator with a radical preference on consensus levels presents a lower initial group consensus level and requires more resources than the other two preference types. A moderator with a mixed preference presents the highest initial group consensus level and requires the lowest resources among the three types of preferences.

Objectives: Online medical crowdfunding has gained popularity in recent years in China. The objective of this study was to identify unmet medical needs in the public healthcare system through analysis of Chinese medical crowdfunding data. Study design: Text information extraction and statistical analysis based on large-scale data. Methods: From 19 June 2011 to 15 March 2020, data from 30,704 medical crowdfunding projects were collected from Tencent GongYi, which is one of the largest Chinese medical crowdfunding platforms. Text mining methods were used to extract data on the medical conditions and locations of the applicants of medical crowdfunding. In addition, 125 medical crowdfunding projects initiated by leukaemia patients in Chongqing and Nanyang were further investigated through manual data extraction, and the factors impacting the fundraising goals were explored using a generalised linear model. Results: The most common conditions using medical crowdfunding to raise funds were as follows: cancer (31.87%), chronic conditions (18.14%), accidental injury (7.80%) and blood system-related conditions (7.75%). Treatments for cancer and blood system-related conditions are expensive and have serious long-term impacts on the lives of patients. Results showed that the cities of Nanyang and Chongqing had the largest number of crowdfunding projects. Conclusions: This study found that the medical conditions that prompted individuals to apply for crowdfunding were those with long treatment cycles, complexities and expensive medical or nonmedical costs. Furthermore, discrepancies in health insurance policies between different regions and residents seeking treatments outside their insurance locations were also important factors that triggered medical crowdfunding applications. Adjusting health insurance policies accordingly may improve the efficiency of utilising health insurance resources and reduce the financial burden on patients. (c) 2023 The Royal Society for Public Health. Published by Elsevier Ltd. All rights reserved.

Technology-oriented micro and small enterprises (TMSEs) play an important role in technology innovation, employment increase, and economic growth. Due to their high risk and resource consuming nature, the Chinese government has set up special funding, and encouraged banks to provide credit support for TMSEs. However, traditional enterprises' credit evaluation methods are not suitable for TMSEs due to their characteristics. How to assess the credit risk of TMSEs is a relatively new and challenging topic. This study is to identify novel indicators that can capture the creditrelated characteristics of TMSEs. Specifically, we proposed that innovation capability and business models can be used to identify some special aspects of TMSEs' credit risk. To validate the new indicators, we compared the performance of traditional financial indicators and the new indicators. The results showed that innovation and business model indicators can effectively improve the performance of classifiers in the credit risk assessment of TMSEs.

Representation learning has an important impact on the performance of machine learning methods and has been used to solve many distribution problems for numerous graphical and sequential mining tasks. While the distributions of credit data are very complex, the represen-tations of such data are less studied. This study proposes a new representation learning approach based on a neural network called Nystro center dot mNet, which represents the credit data to benefit credit evaluation and sub-pattern analysis. The Nystro center dot mNet is developed to utilize the advantages of the Nystro center dot m method - a kernel approximation method in credit evaluation, yet overcomes its two limitations: distance distortions in kernel functions, and parameter tuning. The two main modules contained in Nystro center dot mNet, i.e., the Distance Metric Learning module and the Nystro center dot m module, can benefit each other and yield an overall optimum. Experiments using six real-life large-scale credit data showed that the AUC of the distance-based classifiers and the linear classifiers were improved by 2-11% and 2-14% with the newly generated distributions. The proposed approach also has certain practical advantages over traditional approaches because it is free from complex parameter tuning, consumes fewer memories, and is easy to utilize automatic differential frameworks such as PyTorch. The proposed approach is highly suitable for large-scale credit evaluation.

BackgroundDifferences in bronchial microbiota composition have been found to be associated with asthma; however, it is still unclear whether these findings can be applied to recurrent wheezing in infants especially with aeroallergen sensitization.ObjectivesTo determine the pathogenesis of atopic wheezing in infants and to identify diagnostic biomarkers, we analyzed the bronchial bacterial microbiota of infants with recurrent wheezing and with or without atopic diseases using a systems biology approach.MethodsBacterial communities in bronchoalveolar lavage samples from 15 atopic wheezing infants, 15 non-atopic wheezing infants, and 18 foreign body aspiration control infants were characterized using 16S rRNA gene sequencing. The bacterial composition and community-level functions inferred from between-group differences from sequence profiles were analyzed.ResultsBoth α- and β-diversity differed significantly between the groups. Compared to non-atopic wheezing infants, atopic wheezing infants showed a significantly higher abundance in two phyla (Deinococcota and unidentified bacteria) and one genus (Haemophilus) and a significantly lower abundance in one phylum (Actinobacteria). The random forest predictive model of 10 genera based on OTU-based features suggested that airway microbiota has diagnostic value for distinguishing atopic wheezing infants from non-atopic wheezing infants. PICRUSt2 based on KEGG hierarchy (level 3) revealed that atopic wheezing-associated differences in predicted bacterial functions included cytoskeleton proteins, glutamatergic synapses, and porphyrin and chlorophyll metabolism pathways.ConclusionThe differential candidate biomarkers identified by microbiome analysis in our work may have reference value for the diagnosis of wheezing in infants with atopy. To confirm that, airway microbiome combined with metabolomics analysis should be further investigated in the future.

Since the performances of suppliers usually fluctuate over time and are affected by environmental changes, enterprises need to evaluate suppliers in the past several periods, rather than a single period. As the environment changes, decision makers will exchange their views and influence each other. In addition, multiple decision makers are not always knowledgeable and sometimes express their preferences about suppliers using fuzzy numbers. To meet the different evaluation requirements of decision makers and analyze the influence of time factors and opinion interaction between decision makers, this paper develops a group decision making approach considering multi-period fuzzy information and opinion interaction of decision makers for supplier selection. In this approach, decision makers provide their preferences in multiple periods using generalised fuzzy numbers and the weights of different periods are determined by a mathematical programming method. The effects of opinion interaction are considered by assigning various weights to different decision makers. Fuzzy Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) method is used to rank potential suppliers. The results of a supplier selection example show the proposed approach can select suitable suppliers by considering multi-period fuzzy information and opinion interaction.

Pairwise comparison matrix (PCM) has been widely employed in the multi-criteria decision-making (MCDM) problems to rank the criteria and alternatives according to the considered criteria in Analytic Hierarchy Process (AHP). The PCM should have the acceptable consistency before deriving a priority vector from it. Approximate thresholds of geometric consistency index (GCI) and consistency ratio (CR) have been proposed to test whether the PCM has the acceptable consistency. However, approximate thresholds of GCI and CR always suffer from some criticisms and disagreements in existing literature. In this paper, we try to induce dynamic thresholds of GCI by combining hypothesis testing and random index (RI), which vary with the order of the PCM, significance level and assessment level of decision maker. The induced dynamic thresholds of GCI may explain different (or conflicting) results obtained by approximate thresholds of GCI and CR and avoid the unnecessary revisions of some judgments of the PCM for the desired consistency. Finally, several numerical examples and real-world decision-making problems are examined and compared with existing decision-making methods to illustrate the performance of dynamic thresholds of GCI.