Domain Expert Knowledge Research Articles

A Self-Adaptive Mutation Neural Architecture Search Algorithm Based on Blocks

Recently, convolutional neural networks (CNNs) have achieved great success in the field of artificial intelligence, including speech recognition, image recognition, and natural language processing. CNN architecture plays a key role in CNNs' performance. Most previous CNN architectures are hand-crafted, which requires designers to have rich expert domain knowledge. The trial-and-error process consumes a lot of time and computing resources. To solve this problem, researchers proposed the neural architecture search, which searches CNN architecture automatically, to satisfy different requirements. However, the blindness of the search strategy causes a 'loss of experience' in the early stage of the search process, and ultimately affects the results of the later stage. In this paper, we propose a self-adaptive mutation neural architecture search algorithm based on ResNet blocks and DenseNet blocks. The self-adaptive mutation strategy makes the algorithm adaptively adjust the mutation strategies during the evolution process to achieve better exploration. In addition, the whole search process is fully automatic, and users do not need expert knowledge about CNNs architecture design. In this paper, the proposed algorithm is compared with 17 state-of-the-art algorithms, including manually designed CNN and automatic search algorithms on CIFAR10 and CIFAR100. The results indicate that the proposed algorithm outperforms the competitors in terms of classification performance and consumes fewer computing resources.

Human and Machine Intelligence Together Drive Drug Repurposing in Rare Diseases.

Repurposing is an increasingly attractive method within the field of drug development for its efficiency at identifying new therapeutic opportunities among approved drugs at greatly reduced cost and time of more traditional methods. Repurposing has generated significant interest in the realm of rare disease treatment as an innovative strategy for finding ways to manage these complex conditions. The selection of which agents should be tested in which conditions is currently informed by both human and machine discovery, yet the appropriate balance between these approaches, including the role of artificial intelligence (AI), remains a significant topic of discussion in drug discovery for rare diseases and other conditions. Our drug repurposing team at Vanderbilt University Medical Center synergizes machine learning techniques like phenome-wide association study—a powerful regression method for generating hypotheses about new indications for an approved drug—with the knowledge and creativity of scientific, legal, and clinical domain experts. While our computational approaches generate drug repurposing hits with a high probability of success in a clinical trial, human knowledge remains essential for the hypothesis creation, interpretation, “go-no go” decisions with which machines continue to struggle. Here, we reflect on our experience synergizing AI and human knowledge toward realizable patient outcomes, providing case studies from our portfolio that inform how we balance human knowledge and machine intelligence for drug repurposing in rare disease.

A machine learning framework with dataset-knowledgeability pre-assessment and a local decision-boundary crispness score: An industry 4.0-based case study on composite autoclave manufacturing

Manufacturing advanced fiber-reinforced polymer composite structures e.g. in aerospace sector is known to be a highly complex task, subjected to a stringent framework of process qualifications and structural substantiation. Conversely, Industry 4.0 is an emerging set of sensing, network and predictive technologies and tools which can enable better decision-making towards risk reduction, supported by data-driven models, representing a novel and fruitful intersection between information technologies and manufacturing. This study aims to investigate and develop innovative applications of artificial intelligence, machine learning and data-driven approaches in support of manufacturing advanced composites, including the model building process of analyses, optimization, inverse problems and experimental investigations. This article presents an Industry 4.0-based composite process modeling pipeline, including both dataset- and model-based knowledge engineering activities pertinent to industrial considerations, with the novel approach of integrating domain expertise directly into the dataset and model building activities. This is investigated through a case study on the autoclave cure process of the Hexcel AS4/8552 carbon prepreg material. In particular, surrogate and highly interpretable models such as Logistic Rule Regression are combined with domain expert knowledge under a fuzzy scoring to provide a tool for assessing the quality (knowledgeability) of complex manufacturing datasets, prior to training black-box models. Further, specificity as a global indicator is combined with the novel Decision Boundary Crispness Score (DBSC) as a local indicator, to better quantify the overall confidence of the users in the model’s prediction in the context of small perturbations in manufacturing conditions. Results demonstrate that the DBSC yields a more granular and conservative assessment of both dataset and model quality. Such explainability and local assessment features is deemed vital for production engineers in practice, especially when deploying black-box models for high-risk, in-situ decision-making.

Towards Legal Ontology for Education

Semantic Web using ontology-based knowledge model has been largely applied to e-Learning systems. The use of domain ontologies leverages the transformation of the learning contents of serving human to machine understandable formats. The ontology-based approach could further support e-Learning activities related to legal compliances, such as legally validate learners’ profiles or protecting intellectual rights in e-Learning courses. This study proposes a novel ontological e-Learning framework including the legal knowledge model, the e-Learning knowledge model and the common knowledge model. A customized ontology engineering method is conducted to construct the preliminary version of these specified knowledge models, while a Semantic Web Rule Language (SWRL) rule base is constructed to capture knowledge of both laws and domain experts. FOCA based validation is introduced to confirm the feasibility of this proposed framework. Improving the FOCA score of the legal core ontology and enriching both the ontological knowledge model and its SWRL rule base are topics that can be explored in future studies.

Optimizing the energy efficiency of chiller systems in the semiconductor industry through big data analytics and an empirical study

Big data analytics is playing a more and more prominent role in the manufacturing industry as corporations attempt to utilize vast amounts of data to optimize the operation of plants and factories to gain a competitive advantage. Since the advent of Industry 4.0, also known as smart manufacturing, big data analytics, combined with expert domain knowledge, is facilitating ever-greater levels of speed and automaticity in manufacturing processes. The semiconductor industry is a fundamental driver of this transformation; moreover, due to the highly complex and energy-consuming nature of the semiconductor manufacturing process, semiconductor fabrication facilities (fabs) can also benefit greatly from incorporating big data analytics to improve production and energy efficiency. This paper developed a big data analytics framework, along with an empirical study conducted in collaboration with a semiconductor manufacturer in Taiwan, to optimize the energy efficiency of chiller systems in semiconductor fabs. Chiller systems are one of the most energy-consuming systems within a typical modern fab. The developed big data analytics framework allows production managers to ensure that chiller systems operate at an optimized level of energy efficiency under dynamically changing conditions, while fulfilling the chilling demands. Compared to the commonly-used heuristics previously employed at the fab to tune chiller system parameters, by the utilization of big data analytics, it is shown that fabs can achieve substantial energy savings, greater than 12%. The developed framework and the lessons learned from the empirical study are not only generalizable but also useful for practitioners who are interested in applying big data analytics to optimize the performance of other equipment systems in fabs.

Automated activity and progress analysis based on non-monotonic reasoning of construction operations

PurposeReal-time location sensing (RTLS) systems offer a significant potential to advance the management of construction processes by potentially providing real-time access to the locations of workers and equipment. Many location-sensing technologies tend to perform poorly for indoor work environments and generate large data sets that are somewhat difficult to process in a meaningful way. Unfortunately, little is still known regarding the practical benefits of converting raw worker tracking data into meaningful information about construction project progress, effectively impeding widespread adoption in construction.Design/methodology/approachThe presented framework is designed to automate as many steps as possible, aiming to avoid manual procedures that significantly increase the time between progress estimation updates. The authors apply simple location tracking sensor data that does not require personal handling, to ensure continuous data acquisition. They use a generic and non-site-specific knowledge base (KB) created through domain expert interviews. The sensor data and KB are analyzed in an abductive reasoning framework implemented in Answer Set Programming (extended to support spatial and temporal reasoning), a logic programming paradigm developed within the artificial intelligence domain.FindingsThis work demonstrates how abductive reasoning can be applied to automatically generate rich and qualitative information about activities that have been carried out on a construction site. These activities are subsequently used for reasoning about the progress of the construction project. Our framework delivers an upper bound on project progress (“optimistic estimates”) within a practical amount of time, in the order of seconds. The target user group is construction management by providing project planning decision support.Research limitations/implicationsThe KB developed for this early-stage research does not encapsulate an exhaustive body of domain expert knowledge. Instead, it consists of excerpts of activities in the analyzed construction site. The KB is developed to be non-site-specific, but it is not validated as the performed experiments were carried out on one single construction site.Practical implicationsThe presented work enables automated processing of simple location tracking sensor data, which provides construction management with detailed insight into construction site progress without performing labor-intensive procedures common nowadays.Originality/valueWhile automated progress estimation and activity recognition in construction have been studied for some time, the authors approach it differently. Instead of expensive equipment, manually acquired, information-rich sensor data, the authors apply simple data, domain knowledge and a logical reasoning system for which the results are promising.

Feature fusion using deep learning for smartphone based human activity recognition.

Identification of human physical activities is an active research area since long due to its application in personalized health and fitness monitoring. The performance accuracy of human activity recognition (HAR) models mainly depend on the features which are extracted from domain knowledge. The features are the input of the classification algorithm to efficiently identify human physical activities. Manually extracted features (handcrafted) need expert domain knowledge. Thus these features have significant importance to identify different human activities. Recently deep learning methods are utilized to extract the features automatically from raw sensory data for HAR models. However, state-of-the-art HAR literature established that the importance of handcrafted features can’t be ignored as it is extracted from expert domain knowledge. Thus, in this paper we use the fusion of both the handcrafted features and automatically extracted features using deep learning (DL) for HAR model to enhance the performance of HAR. Extensive experimental results demonstrate that our proposed feature fusion based HAR model gives higher accuracy compared with state-of-the-art HAR literature for both the self collected and public dataset.

Evidential classification and feature selection for cyber-threat hunting

In recent years, there has been an immense research interest in applying Machine Learning for defending networked systems from cyber threats. A particular challenge in this domain is the identification and selection of appropriate features that ensure prompt and correct cyber threat detection. This work proposes a novel approach that leverages recent advances in evidence theory to provide a deep insight on the effect of each feature’s uncertainty on the overall classification decision. As a result, a network security analyst may rank the features in a dataset from the most to the least ambiguous, without requiring expert domain knowledge in cyber threats. Ultimately, this enables the creation of cyber threat phenotypes, which may be used to detect and differentiate between similarly manifested cyber threats. The proposed approach is evaluated on a recent, challenging scenario of network security attacks and compared against multiple feature selection techniques. Based on the selected features, cyber threat classification analysis is performed using seven state-of-the-art ML classification algorithms. The results indicate the proposed evidence-based feature selection method performs better, or, at least as good, to the state-of-the-art. Against the best performing state-of-the-art technique, Decision Tree, the proposed technique’s features enabled the classification process to take place in 93.25% of the time, whilst maintaining a high F1 Score of 0.99. Furthermore, the proposed technique’s features enable a faster classification process requiring, on average, just 29.25% of the time compared to the average across other evaluated techniques.

Defect detection in CT scans of cast aluminum parts: A machine vision perspective

One of the many applications of X-ray computed tomography (CT) in industry is the detection of pores, cavities and other flaws in cast metal parts. Because of its improvement on part safety and saving of expenses, CT inspection is moving from a random sample inspection towards a full in-line inspection. With the increasing amount of produced data, however, comes the need for an automated processing. Due to tight time constraints the resulting CT scans are very artifact afflicted, which impedes automated inspection. In recent years, deep learning methods—convolutional neural networks in particular—have been used with great success to tackle even complex segmentation tasks in cluttered scenes. As we show, these methods are also applicable to the domain of industrial CT data: they are able to cope with noise, beam hardening, scatter and other artifacts which we encounter here. However, these methods need a vast amount of precisely labeled training data to work properly. Gathering the necessary data is not only cumbersome due to the need of annotating three-dimensional data but also expensive as it requires the knowledge of domain experts. Therefore, we present a new approach: We train our models on realistically simulated CT data only. Here, a precise per-voxel ground truth can simply be computed. In order to show that the simulated data is sufficient to train a segmentation network, we turn to its prediction performance on real CT data. We compare the prediction performance of traditional algorithms as well as the trained segmentation network on simulated and real validation data and demonstrate that they behave similarly. The ground truth for the real validation data is hand-labeled using high-quality CT scans, while the actual validation set consists of CT scans of lower quality of the exact same parts. For a comprehensive evaluation, we evaluate the probability of detection as well as the intersection over union. The first tells us how likely a flaw of given size can be found with a given confidence, which is of special interest to domain experts. The latter gives us a per-voxel information of how precise the overall segmentation is. Moreover, our synthetic data enables us to examine the influence of different artifact types on the detection rate. Besides these quantitative analyses we show some qualitative results of real-world applications. To the best of our knowledge, we describe the first approach for defect detection in three-dimensional CT data, which is solely trained with simulated data.

Visual interpretability in 3D brain tumor segmentation network

Medical image segmentation is a complex yet one of the most essential tasks for diagnostic procedures such as brain tumor detection. Several 3D Convolutional Neural Network (CNN) architectures have achieved remarkable results in brain tumor segmentation. However, due to the black-box nature of CNNs, the integration of such models to make decisions about diagnosis and treatment is high-risk in the domain of healthcare. It is difficult to explain the rationale behind the model's predictions due to the lack of interpretability. Hence, the successful deployment of deep learning models in the medical domain requires accurate as well as transparent predictions. In this paper, we generate 3D visual explanations to analyze the 3D brain tumor segmentation model by extending a post-hoc interpretability technique. We explore the advantages of a gradient-free interpretability approach over gradient-based approaches. Moreover, we interpret the behavior of the segmentation model with respect to the input Magnetic Resonance Imaging (MRI) images and investigate the prediction strategy of the model. We also evaluate the interpretability methodology quantitatively for medical image segmentation tasks. To deduce that our visual explanations do not represent false information, we validate the extended methodology quantitatively. We learn that the information captured by the model is coherent with the domain knowledge of human experts, making it more trustworthy. We use the BraTS-2018 dataset to train the 3D brain tumor segmentation network and perform interpretability experiments to generate visual explanations.

Intelligent Health Care: Applications of Deep Learning in Computational Medicine.

With the progress of medical technology, biomedical field ushered in the era of big data, based on which and driven by artificial intelligence technology, computational medicine has emerged. People need to extract the effective information contained in these big biomedical data to promote the development of precision medicine. Traditionally, the machine learning methods are used to dig out biomedical data to find the features from data, which generally rely on feature engineering and domain knowledge of experts, requiring tremendous time and human resources. Different from traditional approaches, deep learning, as a cutting-edge machine learning branch, can automatically learn complex and robust feature from raw data without the need for feature engineering. The applications of deep learning in medical image, electronic health record, genomics, and drug development are studied, where the suggestion is that deep learning has obvious advantage in making full use of biomedical data and improving medical health level. Deep learning plays an increasingly important role in the field of medical health and has a broad prospect of application. However, the problems and challenges of deep learning in computational medical health still exist, including insufficient data, interpretability, data privacy, and heterogeneity. Analysis and discussion on these problems provide a reference to improve the application of deep learning in medical health.

Dataset2Vec: learning dataset meta-features

Meta-learning, or learning to learn, is a machine learning approach that utilizes prior learning experiences to expedite the learning process on unseen tasks. As a data-driven approach, meta-learning requires meta-features that represent the primary learning tasks or datasets, and are estimated traditonally as engineered dataset statistics that require expert domain knowledge tailored for every meta-task. In this paper, first, we propose a meta-feature extractor called Dataset2Vec that combines the versatility of engineered dataset meta-features with the expressivity of meta-features learned by deep neural networks. Primary learning tasks or datasets are represented as hierarchical sets, i.e., as a set of sets, esp. as a set of predictor/target pairs, and then a DeepSet architecture is employed to regress meta-features on them. Second, we propose a novel auxiliary meta-learning task with abundant data called dataset similarity learning that aims to predict if two batches stem from the same dataset or different ones. In an experiment on a large-scale hyperparameter optimization task for 120 UCI datasets with varying schemas as a meta-learning task, we show that the meta-features of Dataset2Vec outperform the expert engineered meta-features and thus demonstrate the usefulness of learned meta-features for datasets with varying schemas for the first time.

Using Natural Language Processing and Artificial Intelligence to Explore the Nutrition and Sustainability of Recipes and Food.

In this paper, we discuss the use of natural language processing and artificial intelligence to analyze nutritional and sustainability aspects of recipes and food. We present the state-of-the-art and some use cases, followed by a discussion of challenges. Our perspective on addressing these is that while they typically have a technical nature, they nevertheless require an interdisciplinary approach combining natural language processing and artificial intelligence with expert domain knowledge to create practical tools and comprehensive analysis for the food domain.

Condition based maintenance of the two-beam laser welding in high volume manufacturing of piezoelectric pressure sensor

Two-beam laser welding (TBLW) is an advanced process for precise, low distortion joining of cylindrical miniature parts. The process is composed of a laser source, optics and various actuators, which form a sophisticated system for control and maintenance in high volume manufacturing. A well-established method for identifying welding defects and ensuring welding quality is the monitoring of plasma light emission in TBLW. Although such monitoring systems can detect a change in process status, they are not able to diagnose the nature of the fault. The main challenge in this research was to extend the use of quality-based monitoring systems to measure additional deterioration-related parameters and to estimate system deterioration from them by using expert knowledge.This paper shows a novel condition-based maintenance (CBM) for the TBLW system, which performs condition identification using online monitoring of plasma light emission in combination with offline inspection of the seam macrographs. A combination of quality parameters derived from seam macrographs of defective parts is used to identify process deterioration, such as contamination of the optics, misalignment of the optomechanical system, or reduced laser power. The information obtained is used to make predefined process adjustments based on expert domain knowledge. The implementation of the developed CBM in high volume manufacturing of piezoelectric pressure sensors resulted in more predictable TBLW by reducing system failures as well as shorter diagnosis times.

INTERVAL TYPE II FUZZY EXPERT SYSTEM FOR DIAGNOSIS VISCERAL LEISHMANIASIS

Leishmaniasis is neglected tropical a protozoan infectious disease caused by Leishmania parasites. VL (also known as Kala-Azar) is the most severe chronic form of Leishmaniasis, almost always fatal if untreated. The annual burden of VL in Ethiopia is estimated to be between 4,500 and 5,000 cases. In a medical environment, incomplete information and imprecise are driven to making an incorrect medical decision and maximize the rate of morbidity and mortality. Fuzzy logic technology enables to provide a simple way to attain to a certain conclusion from vague, imprecise and ambiguous medical data. Clinical suspicion of VL is complex because of its clinical manifestation are shared by other commonly occurring tropical disease like malaria, typhoid and tuberculosis. Several studies were conducted on VL, nevertheless, no one addresses the VL diagnosis using learning capacity. This thesis, we developed and investigated the application of an intelligent interval Type-2 Fuzzy Logic Expert System for diagnosis Visceral Leishmaniasis (VLDES) to assist health workers easy diagnosis a patient and decision-makers to prevent the expansion epidemic. The Type-2 Fuzzy Logic System would be clear to present a highly interpretable and transparent model that is very suitable for the handling uncertainties in the input factors and converting the accumulated data to linguistic formats. First we acquired well organized domain expert knowledge via interview questions and relevant document from the university of Gondar Kala-Azar Treatment and Research center. Secondly, we defined the parameters of fuzzy membership functions of Type-1 Fuzzy Logic System mentioned based on domain knowledge with intensive discussion. Finally, we employed to define uncertainty of the Footprint of Uncertainty (FOU) percentage of Interval Type-2 Fuzzy Logic System. We obtained 98.97% classification accuracy respectively from fifteen patients testing data. This show that the proposed Interval Type-2 Fuzzy Logic Classification System provides a more interpretable model that Diagnosis VL.

Pal-GAN: Palette-conditioned Generative Adversarial Networks

Recent advances in Generative Adversarial Networks (GANs) have shown great progress on a large variety of tasks. A common technique used to yield greater diversity of samples is conditioning on class labels. Conditioning on high-dimensional structured or unstructured information has also been shown to improve generation results, e.g. Image-to-Image translation. The conditioning information is provided in the form of human annotations, which can be expensive and difficult to obtain in cases where domain knowledge experts are needed. In this paper, we present an alternative: conditioning on low-dimensional structured information that can be automatically extracted from the input without the need for human annotators. Specifically, we propose a Palette-conditioned Generative Adversarial Network (Pal-GAN), an architecture-agnostic model that conditions on both a colour palette and a segmentation mask for high quality image synthesis. We show improvements on conditional consistency, intersection-over-union, and Fréchet inception distance scores. Additionally, we show that sampling colour palettes significantly changes the style of the generated images.

Big data analytics energy-saving strategies for air compressors in the semiconductor industry – an empirical study

Industry 4.0, smart manufacturing and its related technologies are now becoming the leading trend in the development of the manufacturing industry. One of the key drivers of Industry 4.0 is big data analytics, which can transform large amounts of data into useful information, enabling astute and rapid decision-making strategies when combined with expert domain knowledge. The semiconductor industry is the most important high-tech industry in Taiwan, but it is also one of the most energy-consuming industries in the country. Therefore, it is critical to improve the efficiency of the manufacturing process and reduce the overall energy consumption of facility systems. This research demonstrates how to apply big data analytics in the semiconductor industry to explore the relationships of various machine parameters, develop predictive models for machine energy efficiency and apply optimisation tools to minimise energy consumption, while meeting the production demands. An empirical study is conducted in conjunction with a semiconductor corporation in Taiwan, targeting the air compressor system in its factory. The research framework is shown to be capable of assisting semiconductor fabrication plant decision-makers in optimising machine configurations, resulting in more than 10% savings on energy consumption and significantly decreased manufacturing costs.

Exploiting structured high-level knowledge for domain-specific visual classification

In the last decade, deep learning models have yielded impressive performance on visual object recognition and image classification. However these methods still rely on learning visual data distributions and show difficulties in dealing with complex scenarios where visual appearance only is not enough to effectively tackle them. This is the case, for instance, of fine-grained image classification in domain-specific applications for which it is very complex to employ data-driven models because of the lack of large amounts of samples and that, instead, can be solved by resorting to specialized human knowledge. However, encoding this specialized knowledge and injecting it into deep models is not trivial.In this paper, we address this problem by: a) employing computational ontologies to model specialized knowledge in a structured representation and, b) building a hybrid visual-semantic classification framework. The classification method performs inference over a Bayesian Network graph, whose structure depends on the knowledge encoded in an ontology and evidences are built using the outputs of deep networks.We test our approach on a fine-grained classification task, employing an extremely complex dataset containing images from several fruit varieties as well as visual and semantic annotations. Since the classification is done at the variety level (e.g., discriminating between different cherry varieties), appearance changes slightly and expert domain knowledge — making using of contextual information — is required to perform classification accurately.Experimental results show that our approach significantly outperforms standard deep learning–based classification methods over the considered scenario as well as existing methods leveraging semantic information for classification. These results demonstrate, on one hand, the difficulty of purely-visual deep methods in tackling small and highly-specialized datasets and, on the other hard, the capabilities of our approach to effectively encode and use semantic knowledge for enhanced accuracy.

Choosing between Gaussian and MPS simulation: the role of data information content\u2014a case study using uncertain interpretation data points

Integration of geophysical data with information such as from boreholes and expert domain knowledge is often performed as cognitive or explicit geological modeling when generating deterministic geological models of the subsurface. However, such cognitive geological models lack the ability to express the uncertainty of layer boundaries. To remedy the shortcomings of this strategy we propose a novel stochastic methodology combining the efforts of probabilistic data integration and cognitive modeling. We treat geological interpretation points from the cognitive model as uncertain “soft” data. These data are then combined with analogous geology in a probabilistic model. We test two ways of combining and sampling from such a probabilistic model. Firstly, a high-entropy setup based on Gaussian distributions simulation. Secondly, lower entropy (and conceivable more realistic) geological solutions are obtained from multiple-point geostatistics (MPS). We apply both ways of solving the problem at a study site near Horsens, Denmark, where airborne transient electromagnetic measurements, seismic data, and borehole information are available and interpreted in cognitive modeling. We explain the complete framework for integrating the uncertain interpretations in geostatistical simulation. Results show that both the Gaussian simulation and multiple-point geostatistical approach allows satisfactory simulations of uncertain geological interpretations and are consistent with prior geological knowledge. Our results suggest that the number of uncertain data points and their information content play a pivotal role in selecting the most appropriate simulation method for the given framework. MPS simulations allow connectivity in scenarios with few data points due to the low entropy of the model. When the number of soft data increases, Gaussian simulation is less prone to produce simulation artifacts, faster to apply, and could be considered preferential to using MPS simulation.

UserRBPM: User Retweet Behavior Prediction with Graph Representation Learning

Social and information networks such as Facebook, Twitter, and Weibo have become the main social platforms for the public to share and exchange information, where we can easily access friends’ activities and in turn be influenced by them. Consequently, the analysis and modeling of user retweet behavior prediction have an important application value, such as information dissemination, public opinion monitoring, and product recommendation. Most of the existing solutions for user retweeting behavior prediction are usually based on network topology maps of information dissemination or designing various handcrafted rules to extract user‐specific and network‐specific features. However, these methods are very complex or heavily dependent on the knowledge of domain experts. Inspired by the successful use of neural networks in representation learning, we design a framework, UserRBPM, to explore potential driving factors and predictable signals in user retweet behavior. We use the graph embedding technology to extract the structural attributes of the ego network, consider the drivers of social influence from the spatial and temporal levels, and use graph convolutional networks and the graph attention mechanism to learn its potential social representation and predictive signals. Experimental results show that our proposed UserRBPM framework can significantly improve prediction performance and express social influence better than traditional feature engineering‐based approaches.