To leverage the advantages of LLM in addressing challenges in the Text-to-SQL task, we present XiYan-SQL, an innovative framework effectively generating and utilizing multiple SQL candidates. It consists of three components: 1) a Schema Filter module filtering and obtaining multiple relevant schemas; 2) a multi-generator ensemble approach generating multiple high-quality and diverse SQL queries; 3) a selection model with a candidate reorganization strategy implemented to obtain the optimal SQL query. Specifically, for the multi-generator ensemble, we employ a multi-task fine-tuning strategy to enhance the capabilities of SQL generation models for the intrinsic alignment between SQL and text, and construct multiple generation models with distinct generation styles by fine-tuning across different SQL formats. The experimental results and comprehensive analysis demonstrate the effectiveness and robustness of our framework. Overall, XiYan-SQL achieves a new SOTA performance of 75.63% on the notable BIRD benchmark, surpassing all previous methods. It also attains SOTA performance on the Spider test set with an accuracy of 89.65%.

Biomedical multimodal large language models (MLLMs) outperform general-domain models on complex biomedical images, yet they still face significant modality disparities. Current methods mainly fine-tune general MLLMs with specialized data, lacking efficient architectural innovation. Tailored frameworks often underperform across multiple modalities, reducing visual comprehension and relying on unsustainable data scaling. To address this, we propose Multi-level Multi-branch Collaborative LLaVA (MMC-LLaVA), a new framework that enhances multi-modal visual information capture in biomedical images. Based on LLaVA-Med, MMC-LLaVA adds two visual branches: one for global structure and another for text-guided fine-grained details, complemented by a gated dynamic fusion module. Experiments on Med-VQA show that MMC-LLaVA outperforms previous state-of-the-art models across multiple metrics, with notable gains in detailed description scores.

Electroencephalography (EEG) is a non-invasive technique to measure and record brain electrical activity, widely used in various BCI and healthcare applications. Early EEG decoding methods rely on supervised learning, limited by specific tasks and datasets, hindering model performance and generalizability. With the success of large language models, there is a growing body of studies focusing on EEG foundation models. However, these studies still leave challenges: Firstly, most of existing EEG foundation models employ full EEG modeling strategy. It models the spatial and temporal dependencies between all EEG patches together, but ignores that the spatial and temporal dependencies are heterogeneous due to the unique structural characteristics of EEG signals. Secondly, existing EEG foundation models have limited generalizability on a wide range of downstream BCI tasks due to varying formats of EEG data, making it challenging to adapt to. To address these challenges, we propose a novel foundation model called CBraMod. Specifically, we devise a criss-cross transformer as the backbone to thoroughly leverage the structural characteristics of EEG signals, which can model spatial and temporal dependencies separately through two parallel attention mechanisms. And we utilize an asymmetric conditional positional encoding scheme which can encode positional information of EEG patches and be easily adapted to the EEG with diverse formats. CBraMod is pre-trained on a very large corpus of EEG through patch-based masked EEG reconstruction. We evaluate CBraMod on up to 10 downstream BCI tasks (12 public datasets). CBraMod achieves the state-of-the-art performance across the wide range of tasks, proving its strong capability and generalizability. The source code is publicly available at https://github.com/wjq-learning/CBraMod.

Generating life-like whole-body human grasping has garnered significant attention in the field of computer graphics. Existing works have demonstrated the effectiveness of keyframe-guided motion generation framework, witch focus on modeling the grasping motions of humans in temporal sequence when the target objects are placed in front of them. However, the generated grasping poses of the human body in the key-frames are limited, failing to capture the full range of grasping poses that humans are capable of.To address this issue, we propose a novel framework called COOP (DeCOupling and COupling of Whole-Body GrasPing Pose Generation) to synthesize life-like wholebody poses that cover the widest range of human grasping capabilities. In this framework, we first decouple the wholebody pose into body pose and hand pose and model them separately, which allows us to pre-train the body model with out-of-domain data easily. Then, we couple these two generated body parts through a unified optimization algorithm.Furthermore, we design a simple evaluation method to evaluate the generalization ability of models in generating grasping poses for objects placed at different positions. The experimental results demonstrate the efficacy and superiority of our method. And COOP holds great potential as a plug-and-play component for other domains in whole-body pose generation. Our models and code are available at https://github.com/zhengyanzhao1997/COOP.

In this paper, we consider a novel task, Video Corpus Spatio-Temporal Grounding (VCSTG) for material selection and spatio-temporal adaption in intelligent video editing. Given a text query depicting an object and a corpus of untrimmed and unsegmented videos, VCSTG aims to localize a sequence of spatio-temporal object tubes from the video corpus. Existing methods tackle the VCSTG task in a multi-stage approach, which encodes the query and video representation independently for each task, leading to local optimum. In this paper, we propose a novel one-stage multi-task learning based framework named MTSTG for the VCSTG task. MTSTG learns unified query and video representation for video retrieval, temporal grounding and spatial grounding tasks. Video-level, frame-level and object-level contrastive learning are introduced to measure the mutual information between query and video at different granularity. Comprehensive experiments demonstrate our newly proposed framework outperforms the state-of-the-art multi-stage methods on VidSTG dataset.

Earthquake magnitude estimation is one of the critical parts of earthquake early warning systems. It uses the first few seconds of a waveform recorded by an earthquake detection station, which is required to be rapid and accurate. In this paper, we propose a novel framework to estimate magnitude integrated with deep learning, consisting of feature stage and regression stage. In the feature stage, we extract temporal & spatial features by deep learning methods, and combine them with hand-crafted features embedded expert domain knowledge. Then, each earthquake can be represented by a hybrid feature. Therefore, magnitude estimation can be modeled as a regression problem to solve. Our framework is evaluated on 5,503 earthquake records collected in Sichuan province, China. It is found that, learning the temporal & spatial features by deep neural networks is critical for magnitude estimation. The results demonstrate the state-of-the-art performance, compared with other approaches.

Smartphones are changing humans' lifestyles. Mobile applications (apps) on smartphones serve as entries for users to access a wide range of services in our daily lives. The apps installed on one's smartphone convey lots of personal information, such as demographics, interests, and needs. This provides a new lens to understand smartphone users. However, it is difficult to compactly characterize a user with his/her installed app list. In this article, a user representation framework is proposed, where we model the underlying relations between apps and users with Boolean matrix factorization (BMF). It builds a compact user subspace by discovering basic components from installed app lists. Each basic component encapsulates a semantic interpretation of a series of special-purpose apps, which is a reflection of user needs and interests. Each user is represented by a linear combination of the semantic basic components. With this user representation framework, we use supervised and unsupervised learning methods to understand users, including mining user attributes, discovering user groups, and labeling semantic tags to users. Extensive experiments were conducted on three data subsets from a large-scale real-world dataset for evaluation, each consisting of installed app lists from over 10 000 users. The results demonstrated the effectiveness of our user representation framework.

Graph classification problem is becoming one of research hotspots in the realm of graph mining, which has been widely used in cheminformatics, bioinformatics and social network analytics. Existing approaches, such as graph kernel methods and graph Convolutional Neural Network, are facing the challenges of non-interpretability and high dimensionality. To address the problems, we propose a novel recurrent attention model, called g-Inspector, which applies the attention mechanism to investigate the significance of each region to make the results interpretable. It also takes a shift operation to guide the inspector agent to discover the next relevant region, so that the model sequentially loads small regions instead of the entire large graph, to solve the high dimensionality problem. The experiments conducted on standard graph datasets show the effectiveness of our g-Inspector in graph classification problems.

Due to the significant cognition reduction, multi-media content has become an increasingly important information type nowadays. More and more descriptions are coupled with images to make them more attractive and persuasive. Currently, several text-image retrieval methods have been developed to improve the efficiency of the time-consuming and professional process. However, in practical retrieval applications, it is the vivid and terse descriptions that are widely used, instead of the shallow captions that describe what is contained. Therefore, the most existing methods designed for the caption-style text can not achieve this purpose. To eliminate the mismatch, we introduce a novel problem about description-image retrieval and propose the specially designed method, named Adapted Graph Reasoning and Filtration (AGRF). In AGRF, we firstly leverage an adapted graph reasoning network to discover the combination of visual objects in the image. Then, a cross-modal gate mechanism is proposed to cast aside those description-independent combinations. Experiment results on the real-world dataset demonstrate the advantages of the AGRF over the state-of-the-art methods.

Role-playing games (RPGs) are one of the most exciting and most rapidly expanding genres of online games. Virtual characters that are not controlled by players, have become an integral part, which helps to advance narratives of RPGs. Believable characters can enhance game engagement and further improve player retention. However, game players easily find that most characters' behaviors are limited and improbable, resulting in a less meaningful game experience. In this work, we propose a framework to model game behaviors to learn behavior patterns of human players. Based on the learned behavior patterns, it generates human-like action sequences that can be used for the design of believable virtual characters in RPGs, so as to enhance game engagement. Specifically, considering the influence of game context in behavior patterns, we integrate game context (players' levels and game classes) with actions together to model behaviors. We propose a long-term memory cell on actions and game context to learn the hidden representations. We also introduce an attention mechanism to measure the contribution of the actions previously performed to the next action. Given only one action, our model can generate action sequences by predicting the succeeding action based on the previously generated actions. The model was evaluated on a real-world data set of over 22 000 players and more than 51 million action logs of an RPG game in 21 days. The results demonstrate the state-of-the-art performance.

Generating natural language descriptions for knowledge graph (KG) is an important category for intelligent writing. Recent models on this task substitute the sequence encoder in a commonly used encoder-decoder framework with a graph encoder. However, these models suffer from entity missing and repetition. In this paper, we propose a novel end-to-end generation model named G2T, which integrates a novel Graph Structure Enhanced Mechanism (GSEM) and a Copy Coverage Loss (CCL). Instead of just considering graph structure in the encoding phase in most existing methods, our GSEM fully utilizes graph structure in the decoding phase and helps to mitigate entity missing problem. Moreover, our CCL can further improve performance by avoiding generating repeated entities. With their help, our model is capable of generating fluent description for KG. The results of automatic and human evaluations show that our model outperforms the state-of-the-art models.

The integration of the fifth generation (5G) networks and artificial intelligence (AI) benefits to create a more holistic and better connected ecosystem for industries. User profiling has become an important issue for industries to improve company profit. In the 5G era, smartphone applications have become an indispensable part in our everyday lives. Users determine what apps to install based on their personal needs, interests, and tastes, which is likely shaped by their genders-the behavioral, cultural, or psychological traits typically associated with their sex. It is possible to profile users' gender based simply on a single snapshot of apps installed on their smartphones. With this inference based on easy to access data, we can make smartphone systems more user-friendly, and provide better personalized products and services. In this article, we explore such possibilities through an empirical study on a large-scale dataset of installed app lists from 15 000 Android users. More specifically, we investigate the following research questions: 1) What differences between females and males can be explored from installed app lists? 2) Can user gender be reliably inferred from a snapshot of apps installed? Which snapshot feature(s) are the most predictive? What is the best combination of features for building the gender prediction model? 3) What are the limitations of a gender prediction model based solely on a snapshot of apps installed on a smartphone? We find significant gender differences in app type, function, and icon design. We then extract the corresponding features from a snapshot of apps installed to infer the gender of each user. We assess the gender predictive ability of individual features and combinations of different features. We achieve an accuracy of 76.62% and area under the curve of 84.23% with the best set of features, outperforming the existing work by around 5% and 10%, respectively. Finally, we perform an error analysis on misclassified users and discussed the implications and limitations of this article.

Parkinson’s disease is a progressive nervous system disorder afflicting millions of patients. Among its motor symptoms, bradykinesia is one of the cardinal manifestations. Experienced doctors are required for the clinical diagnosis of bradykinesia, but sometimes they also miss subtle changes, especially in early stages of such disease. Therefore, developing auxiliary diagnostic methods that can automatically detect bradykinesia has received more and more attention. In this article, we employ a two-stage framework for bradykinesia recognition based on the video of patient movement. First, convolution neural networks are trained to localize keypoints in each video frame. These time-varying coordinates form motion trajectories that represent the whole movement. From the trajectory, we then propose novel measurements, namely stability , completeness , and self-similarity , to quantify different motor behaviors. We also propose a periodic motion model called PMNet . An encoder--decoder structure is applied to learn a low dimensional representation of a motion process. The compressed motion process and quantified motor behaviors are combined as inputs to a fully-connected neural network. Different from the traditional means, our solution extends the application scenario outside the hospital and can be easily transplanted to conduct similar tasks. A commonly used clinical assessment is served as a case study. Experimental results based on real-world data validate the effectiveness of our approach for bradykinesia recognition.

App usage prediction, i.e. which apps will be used next, is very useful for smartphone system optimization, such as operating system resource management, battery energy consumption optimization, and user experience improvement as well. However, it is still challenging to achieve usage prediction of high accuracy. In this paper, we propose a novel framework for app usage prediction, called AppUsage2Vec, inspired by Doc2Vec. It models app usage records by considering the contribution of different apps, user personalized characteristics, and temporal context. We measure the contribution of each app to the target app by introducing an app-attention mechanism. The user personalized characteristics in app usage are learned by a module of dual-DNN. Furthermore, we encode the top-k supervised information in loss function for training the model to predict the app most likely to be used next. The AppUsage2Vec was evaluated on a dataset of 10,360 users and 46,434,380 records in three months. The results demonstrate the state-of-the-art performance.

The direct air condensers present unique control problems because they are more sensitive to ambient disturbances than water-cooled condensers. Use ambient air as a cooling medium, the direct air condenser achieves significant water saving at the expense of high fan power consumption, severely affected by ambient wind and coupling effects. To solve these problems, this paper proposed a divisional regulation scheme on an air-cooled array based on model predictive control (MPC) with feedforward. Air flow characteristics of the air-cooled condenser are investigated based on an experimental platform and propose a partition scheme, which divides the air-cooled array into two parts: the outer circle and inner part. Then the dynamic models of the air-cooled array are established based on an integral identification method on step tests, including the wind disturbance response model. Finally, MPC with feed-forward is introduced as the control strategy, with optimized flow set-point for each part. It is concluded that the proposed method effectively reduces the energy consumption of the air-cooled array and suppresses the ambient wind disturbance.

As a rapidly developing renewable energy source, effective active power dispatching for wind power has become common. For offshore wind farms with high maintenance costs, balanced fatigue loads among wind turbines need to be considered in order to reduce the number of maintenances. Based on this target, this paper proposes an optimized power distribution function to balance the fatigue difference among wind turbines. At the same time, this paper adopts the structure of the unsupervised multi-agent system (MAS) in the offshore wind farm, the communication between wind turbines is used to complete the allocation of power commands. The simulation results show that the wind farm with MAS structure can effectively deliver the power commands. And the optimized power distribution function can reduce the fatigue difference of the wind turbines after longterm operation.

As the population ages, eldercare and healthcare have become major issues in recent years. Crossover healthcare services, instead of individual ones, have become the main form of service provision. In this work, a scenario-based requirement model (SBRM) is proposed for crossover healthcare service. A DSL and a prototype system are designed based on the model as well. Our model defines the requirements as: WHO, in what SCENARIOs, what PROCESSes need to be performed, and what RULEs need to be satisfied. We verify our model in the real case of the MEH (medical, eldercare, healthcare) crossover service. SBRM supports the service better in our cases and shows satisfactory efficiency, effectiveness, and reusability.

As more business workflow systems are being deployed in modern enterprises and organizations, more employee-activity log data are being collected and analyzed. In this paper, we develop a latent ability model (LAM) as a generative probabilistic learning framework for workforce analytics over employee-activity logs. The LAM development is novel in three aspects. First, we introduce the concept of latent ability variables to model hidden relations between employees and activities in terms of job performance, such as the set of skills provided by an employee and the set of skills required by an activity, and how well they matchup in employee-activity assignment. Second, we construct the latent ability model by learning latent ability parameters from the employee-activity log data using expectation-maximization and gradient descent. Finally, we leverage LAM to build inference and prediction models for employee performance prediction, employee ability comparison, and employee-activity matchup quality estimation. We evaluate the accuracy and efficiency of our approach using real log datasets collected from a workflow system deployed in the government of the city of Hangzhou, China, which consists of 5,287,621 log records over two years involving 744 activities and 1,725 employees. We show that LAM approach outperforms existing representative methods in both accuracy and efficiency.

The sentiment behind financial text has been observed to have correlations with stock market trend. Though widely discussed, the study on this topic faces the challenge coming from the lack of open dataset and labeled financial text. In this work, we collected a large amount of Chinese financial text from financial news, research report, stock BBS and corporate announcements. It contains 3 million articles about 128 stocks from 2010 to 2018. And then we proposed a model mapping from the text and latent sentiment to the abnormal market trend. It combines the posting amount, daily market index with the RBM-embedded document vector, and extracts the abnormal features via LSTM. After that a neural net is employed to identify the abnormal trend. The experimental results on our dataset show the effectiveness of our approach comparing to baseline methods.

Previous work has demonstrated that typography design has a great influence on users' reading experience. However, current typography design guidelines are mainly for general purpose, while the individual needs are nearly ignored. To achieve personalized typography designs, an important and necessary step is accurately evaluating user satisfaction with the typography designs. Current evaluation approaches, e.g., asking for users' opinions directly, however, interrupt the reading and affect users' judgments. In this paper, we propose a novel method to address this challenge by mining users' implicit feedbacks, e.g., touch interaction data. We conduct two mobile reading studies in Chinese to collect the touch interaction data from 91 participants. We propose various features based on our three hypotheses to capture meaningful patterns in the touch behaviors. The experiment results show the effectiveness of our evaluation models with higher accuracy on comparing with the baseline under three text difficulty levels, respectively.

Crossover service is a kind of services, which can provide multi-dimension service, great user experience and high values, through deeply converging services from different industries, different organizations and different value chains. Convergence is the key challenges for crossover service application. Using Alibaba's crossover service case, this paper illustrates five challenges of the service convergence process: pattern convergence, ecosystem convergence, environment convergence, quality convergence and value convergence. In addition, we propose a technical framework addressing these technical challenges, which includes all the major theories and models, techniques and methods, and tools and platforms supporting enterprises' crossover service convergence in the modeling phase, the design phase, the running phase and the management phase.

NgramCNN is a deep convolutional neural network developed for classification of graphs based on common substructure patterns and their latent relationships in the collection of graphs. Our NgramCNN deep learning framework consists of three novel components: (1) The concept of n-gram graph block to transform each raw graph object into a sequence of n-gram blocks connected through overlapping regions. (2) The diagonal convolution layer to extract local patterns and connectivity features hidden in the n-gram blocks by performing n-gram normalization before conducting deep learning through the network of convolution layers. (3) The extraction of deeper global patterns based on the local patterns and the ways that they respond to overlapping regions by building a n-gram deep convolutional neural network. Extensive evaluation of NgramCNN using five real graph repositories from bioinformatics and social networks domains show the effectiveness of NgramCNN over the existing state of art methods with high accuracy and comparable performance.

In the modern service industry, both service processes and data structures are becoming increasingly diverse and complex. In addition, interdependences exist among data, such that the use of "shoe size" data must be based on the "type of goods" data returning "shoe". This is also observed for the functions and interfaces in a system, as one can use the function "order payment" only after the function "order generation". This kind of phenomenon is rather common in service systems nowadays, especially when the service is a transboundary service such as the new retail proposed by Jack Ma. Traditional modeling methods have difficulties in handling such scenarios. There have been studies on service modeling over the past several years, and they have focused mainly on the service processes and interactions among services. In this work, we construct MeCo-TSM based on three sub-models to handle multi-entity complex service process. We verify our model in the real processes of our cooperation company and compare it with related works. MeCo-TSM supports the service better in our cases and shows satisfactory efficiency, effectiveness and reusability.

Convolutional Neural Network (CNN) has gained attractions in image analytics and speech recognition in recent years. However, employing CNN for classification of graphs remains to be challenging. This paper presents the Ngram graph-block based convolutional neural network model for classification of graphs. Our Ngram deep learning framework consists of three novel components. First, we introduce the concept of n-gram block to transform each raw graph object into a sequence of n-gram blocks connected through overlapping regions. Second, we introduce a diagonal convolution step to extract local patterns and connectivity features hidden in these n-gram blocks by performing n-gram normalization. Finally, we develop deeper global patterns based on the local patterns and the ways that they respond to overlapping regions by building a n-gram deep learning model using convolutional neural network. We evaluate the effectiveness of our approach by comparing it with the existing state of art methods using five real graph repositories from bioinformatics and social networks domains. Our results show that the Ngram approach outperforms existing methods with high accuracy and comparable performance.

Modern service industry (MSI) is becoming a leading and pillar industry in recent years. Its theory construction, however, has not kept up with the industry development. Specifically, the business process designing and reconstructing, the critical part in business transformation and competition of modern service enterprise, are still handled manually. The existing models, e.g., BPMN and EPC, mainly adopt the business activities and events without considering the resource and data generated and consumed in interaction with business collaborators, which is ubiquitous in MSI business process. Hence, the deficiency of resource and data in these models hindered them from popularization in MSI. In this paper, with a systematic analysis of the above issues, we introduce the service pattern of MSI business process from the point view of resource and data with formalized description and some concrete basic patterns. To support the process designing and reconstructing better, we propose a pattern-centered formalization language, called Service Pattern Description Language (SPDL). Furthermore, a service pattern matching approach, in the Constructing-As-Identifying style, is studied and a service process designing tool, called SPDL-Editor, is developed. Additionally, a case of the famous video-on-demand service (Youku) in China is given, though out of this paper, for a better understanding of our theories.

The procedure of picking services bound to abstract tasks is usually called service selection in Service Oriented Architecture. In recent years, most studies focus on improving the Quality of Service (QoS) of the composed service. These techniques, however, are facing a new challenge, brought by the big data era, namely, the time and money wasted in data transmission, called transmission cost, cannot be optimized locally like QoS. To address this challenge, in this paper, we study and formalize the problem of transmission cost aware service selection, named TcSS. Owing to the insufficient service transfer rates, we propose a framework on a relaxation problem by making use of the service network ontology structure. The entire framework comprises two stages, an off-line stage to arrange the service network information from logs and an online stage to satisfy the service selection requirement efficiently. The solution of the relaxation problem is an approximation of the original TcSS with the approximate ratio guarantees. Finally, extensive experiments on real data establish the effectiveness and efficiency of our approach.

The research on the business models is a hot topic in recent years. It is an interesting problem to study the business model of the service. There are three kinds of models related: classical service models, Business process (BP) models and the enterprise business (EB) models in management. However, none of them covers all the properties of the service business model. In this paper, we define the business model of the service as the combination of four kinds of strategies and name it as the service pattern. We also propose a language named Service Pattern Description Language (SPDL) covering all the elements involved in these strategies. We formulate the language syntax and two basic extraction rules assisting economic analysis. Furthermore, we extend Business Process Model Notation (BPMN) to support SPDL, which is named BPMN for Service Pattern (BPMN4SP). The example of Mobile Application Platform is studied in detail for a better understanding of SPDL.

This paper introduces a service selection model with the service location considered. The location of a service represents its position in the network, which determines the transmission cost of calling this service in the composite service. The more concentrated the invoking services are, the less transmission time the composite service costs. On the other hand, the more and more popular big data processing services, which need to transfer mass data as input, make the effect much more obvious than ever before. Therefore, it is necessary to introduce service location as a basic feature in service selection. The definition and membership functions of service location are presented in this paper. After that, the optimal service selection problem is represented as an optimization problem under some reasonable assumptions. A shortest-path based algorithm is proposed to solve this optimization problem. At last, the case of railway detection is studied for better understanding of our model.