Qian Zhang, Xinye Li, Xiaokai Wu, Junhao Xu, Zhanyue Qin, Qingbin Liu, Junxian Cai, Xi Chen, Bolin Zhang, Zhiying Tu, Dianhui Chu, Xiaoyan Yu, Dianbo Sui. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

ABSTRACT Microservice bad smells (MBSs) affect system quality. However, the lack of comprehensive and detailed explanations of MBSs and current classification methods does not comprehensively encompass the microservice characteristics, making it challenging to study and address MBSs. Existing studies have focused on MBS detection methods, but identifying and quantifying the effects of different smells and properly assessing system health remain challenging. This review aims to provide an exhaustive list of the MBSs and formulate a reasonable classification based on microservice characteristics. By reasonably assessing the harmfulness of smells, we would comprehensively explain different MBSs. We would explore microservice system evaluation methods and propose a microservice system health assessment model (MSHAM) based on MBSs. We conducted a systematic literature review (SLR) of the catalog and evaluation literature in the field of MBSs. We also combined the open questionnaire form to answer the research questions qualitatively and quantitatively. This paper presents a comprehensive list of 69 types of MBSs, incorporating a highly scalable classification. Following the quality characteristics in ISO/IEC 25010:2023, we determined the sets of characteristics affected by different smells, and provided detailed explanations of smells in the form of MBSs knowledge base. Furthermore, we introduced the MSHAM, revealing the quantification process of harmfulness. The formulated smells list and classification method capture MBS characteristics comprehensively, ensuring scalability. We have exposed the impact of MBSs, enhancing researcher and practitioner understanding. MSHAM provides a logical model to quantify MBSs harm and system health, supporting the development of real‐time health monitoring.

As a structured representation of real-world facts, knowledge graphs (KGs) play a vital role in IoT applications, due to their strong reasoning capabilities and interpretability. However, private user IoT KG data often needs to be centrally collected for embedding training, which poses significant privacy risks and limits the scalability of knowledge-driven downstream applications in distributed IoT environments. Federated learning (FL) has emerged as a promising solution for decentralized model training, eliminating the need for direct data collection. However, existing federated knowledge graph embedding (KGE) methods often struggle to preserve the inherent graph structure of entities and relations, leading to fragmented and incomplete representations. Additionally, they struggle to effectively capture diverse relational dependencies within personal KGs. To address these challenges, this article proposes an enhanced federated KG embedding method for personal knowledge sharing (FPKS) to enable privacy-preserving KGE training. The FPKS framework consists of a central server and multiple federated clients. To enhance entity and relationship alignment across clients, FPKS maintains separate embedding tables for entities and relationships on the server. Moreover, to capture the structural and contextual information of personal KGs, we introduce a local encoder-decoder architecture that employs a graph convolutional network (GCN) variant as an encoder and a KGE scoring function as a decoder. Furthermore, we propose a bidirectional composite operator for GCN (BiDGCN) to enhance multi-relational information aggregation. Extensive experiments on two widely used KG datasets demonstrate that FPKS significantly outperforms existing methods, improving the quality of learned embeddings while ensuring data privacy. Our approach facilitates decentralized personal knowledge sharing, marking an advancement in secure and efficient IoT knowledge-driven services.

In the metaverse, human-agent interactions increasingly take diverse forms, such as natural dialogues, which enhance user experience but pose challenges for precise service delivery, as users often provide incomplete or ambiguous information—a phenomenon we term fuzzy requirements. This paper addresses how to clarify fuzzy requirements through multi-turn interactions. We formally define the task, construct a benchmark across five domains with user profiles, and propose Tower, a user experience-oriented decision-making algorithm for fuzzy requirement clarification. Tower organizes service parameters into a feature tree as a search space for inquiries and refines it using user profiles and diverse interview questioning strategies. To enable realistic evaluation, we develop a user simulator based on action selection, modeling user personality traits using prospect theory and Fitts's law from cognitive psychology. The simulator proactively presents fuzzy requirements and responds adaptively according to personality traits. Using dialogue histories between Tower and the simulator, we train TowerLM, an automatic clarification model with strong domain generalizatio. Experimental results show that both Tower and TowerLM achieve state-of-the-art performance across three user personality types and can be seamlessly integrated as plugins into existing service delivery methods, improving the delivery success rate under fuzzy requirements on four public datasets.

Software engineering and service computing education play a vital role in service economy. To address the increasing demand for innovative and application-oriented pro-fessionals, it is crucial to overcome challenges such as compressed course hours, an overabundance of theoretical courses with limited practical training, and insufficient cultivation of scientific thinking. This paper explores effective approaches to talent cultivation by integrating curriculum design, project-based practice, competitions, and laboratory research. It highlights key contributions, including the optimization of the curriculum system, the combination of project-based learning with practical training, the balance between teaching and competitions, and the nromotion of active learning.

Parameter-Efficient Fine-Tuning (PEFT) adapts large language models (LLMs) to specific domains by updating only a small portion of the parameters. Although fine-tuning on a single task within a specific domain has demonstrated promising results, there remains limited exploration on how to effectively integrate these adapters for optimal performance. In this paper, we propose Adapters Selector (AS): a novel framework for better integrating usage of multiple adapters by training a middleman adapter to select the appropriate adapter for inference. Our approach utilizes PEFT to train a selector that determines which input content corresponds to which task in which domain, and subsequently selects the homologous adapter. By the way, The AS has developed the capability to execute cross-domain multi-tasks effectively through the utilization of a compact model in combination with multiple LoRA modules. Our code is publicly available.

Large Language Models (LLMs) have shown impressive abilities in solving various natural language processing tasks and are now widely offered as services. LLM services enable users to accomplish tasks without requiring specialized knowledge, simply by paying service providers. However, numerous providers offer various LLM services with variations in pricing, latency, and performance. These factors are also affected by different invocation methods, such as the choice of context and the use of cache, which lead to unpredictable and uncontrollable service cost and quality. Consequently, utilizing various LLM services invocation methods to construct an effective (cost-saving, low-latency and high-performance) invocation strategy that best meets task demands becomes a pressing challenge. This paper provides a comprehensive overview of methods help LLM services to be invoked efficiently. Technically, we define the problem of constructing an effective LLM services invocation strategy, and based on this, propose a unified LLM service invocation framework. The framework classifies existing methods into four categories: input abstraction, semantic cache, solution design, and output enhancement, which can be used separately or jointly during the invocation life cycle. We discuss the methods in each category and compare them to provide valuable guidance for researchers. Finally, we emphasize the open challenges in this domain and shed light on future research.

The rapid proliferation of large language models (LLMs), such as GPT-4 and Gemini, underscores the intense demand for resources during their training processes, posing significant challenges due to substantial computational and environmental costs. In this paper, we introduce a novel checkpoint merging strategy aimed at making efficient use of intermediate checkpoints during LLM pretraining. This method utilizes intermediate checkpoints with shared training trajectories, and is rooted in an extensive search space exploration for the best merging weight via Bayesian optimization. Through various experiments, we demonstrate that: (1) Our proposed methodology exhibits the capacity to augment pretraining, presenting an opportunity akin to obtaining substantial benefits at minimal cost; (2) Our proposed methodology, despite requiring a given held-out dataset, still demonstrates robust generalization capabilities across diverse domains, a pivotal aspect in pretraining.

Knowledge graph (KG)-based service recommendation methods address issues such as data sparsity and cold start in real-world service recommendations by integrating external knowledge as auxiliary information. Recently, large language models (LLMs) have gained significant attention due to their powerful comprehension and reasoning capabilities. LLM-based recommendation systems also demonstrate advantages in interpretability and few-shot service reasoning. However, the integration of LLMs and KGs into existing service recommendation methods presents two major challenges: (1) the difficulty of aligning service recommendation tasks with language modeling tasks, and (2) the lack of interpretable quantification of the relationship between knowledge and personalized preferences. To address these challenges, this paper proposes WPKL (Weighted Preference Optimization based on KG and LLM). WPKL leverages external knowledge to assist LLMs in modeling user preferences and employs a hybrid graph neural network (GNN) framework to enhance preference representation. Additionally, a weighted preference optimization (WPO) approach is proposed to fine-tune the LLM, enabling interpretable quantification of user preferences and personalized knowledge. Extensive experimental results demonstrate that WPKL achieves high-quality service recommendations.

Chatbots have advanced from basic conversational agents to versatile tools by integrating external services. However, traditional chatbots are constrained by predefined service boundaries, limiting their ability to handle complex tasks with unintegrated services. While most research focuses on improving service discovery and invocation through data-intensive pretraining, only 13.29% of services are well-documented, hindering practical deployment. This paper proposes a self-improving workflow for chatbots, using a “wide in, strict out” self-supervised learning approach to acquire domain knowledge efficiently and generate high-quality service documents. Compatible with existing methods, it eliminates the need for dataset collection or pre-training. Experiments demonstrate that our workflow significantly improves the pass and success rate of chatbots in utilizing unintegrated services, offering a powerful solution for real-world applications where service integration is limited.

To address the limitations of existing microservice bad smell (MBS) detection and refactoring tools, particularly the lack of fully automated architectural bad smell refactoring solutions, this paper proposes a MAPE-K-based self-adaptive framework for microservice bad smell refactoring (SABER). The framework aims to eliminate architectural smells through closed-loop self-repair, thereby reducing risks related to main-tainability, scalability, and security. SABER employs a cloud-edge collaborative architecture: edge-side components collect real-time metrics from a Kubernetes cluster, while cloud-side components detect architectural smells and dynamically generate refactoring strategies. These strategies include service merging, splitting, adding, and adjustment. By automatically executing these strategies, SABER achieves architectural bad smell elimi-nation. Experimental results show that the framework achieves 95.53 % precision and 84.71 % recall across ten benchmark systems, significantly improving refactoring efficiency compared to semi-automated and manual methods. Its deep integration with DevOps pipelines validates its effectiveness in sustaining microservice health, offering a novel paradigm for autonomous maintenance in distributed systems.

In the era of industrial internet, achieving a dynamic balance between mass production and personalized customization has become a core demand for industrial development. This necessitates that product service systems can accurately capture users' personalized requirements. Although large language models (LLMs) possess powerful dialogue and reasoning capabilities, enabling them to identify implicit requirements, they still exhibit limitations in supply-demand matching, particularly in the precise alignment between personalized requirements and product capabilities. To this end, this study innovatively proposes a personalized product customization service framework (Req2Sol) based on fine-grained supply-demand cognition. This framework formalizes the modeling of supply-demand capabilities and finegrained personalized requirements through a knowledge graph (KG), integrating them into the LLM training process. This significantly enhances the model's understanding of supplydemand relationships, enabling accurate product configuration and customization recommendations. Firstly, a multi-view modeling approach for supply-demand capabilities and personalized fine-grained requirements is proposed, constructing a requirementproduct knowledge graph. Secondly, the knowledge graph is utilized as pre-training data to achieve domain-specific finetuning of LLMs. By introducing conditional scenarios and strategies, a five-level quantitative evaluation system for Req2Sol is established, improving its performance by 7.3 % compared to the baseline model when handling unconventional or inaccurately expressed user requirements. Finally, using the air conditioning domain as a case study, the effectiveness of the framework in achieving precise supply-demand cognition and customized product recommendations is validated through the Req2Sol-QAS system developed by invoking Req2Sol services.

The integration of Mobile Edge Computing (MEC) into the Industrial Internet of Things (IIoT) has markedly improved resource accessibility and propelled digital-intelligent advancements. Nevertheless, the substantial costs of MEC infrastructure also pose critical challenges to incentive mechanism design. Specifically, there is still an absence of a standardized and widely recognized incentive framework for the dynamic and non-cooperative interactions between edge service requesters and providers. Furthermore, the highly complicated characteristics of IIoT necessitate a greater reliance on dependable and trustworthy edge resource provision than other paradigms, which implies that human-centric factors (e.g., credit) are equally crucial as profit-driven metrics (e.g., price) in incentive design. To tackle these challenges, we propose In3Edge, a Stackelberg game-based incentive mechanism that systematically considers the interplay between profit-driven and interest-oriented indicators while accommodating heterogeneous peers, subjective interest divergences, and objective resource disparities. Particularly, leveraging convex optimization theory, we provide rigorous proofs and in-depth analyses of the intrinsic properties of In3Edge, encompassing the concavity/convexity of utility functions, equilibrium solution boundaries, optimal responses under peer/interest heterogeneity, and closed-form solutions for symmetric multipeer scenarios while articulating a series of propositions and theorems to underpin future research. Finally, extensive experiments are constructed under diverse dynamically changing scenarios with distinct characteristics, confirming the strong motivational capabilities of In3Edge in MEC-empowered IIoT.

Large language models (LLMs) cannot effectively collaborate with humans who provide imperfect information at the initial stage of the dialogue, unless they learn to proactively ask questions.

The predominant obstacles confronting contemporary cyber-physical systems (CPSs) are their extensive heterogeneity and stringent constraints, such as diverse applications and real-time service requirements. While the incorporation of mobile edge computing could alleviate some of these constraints, challenges persist in equilibrating services and loads due to the finite computational resources of edge servers. Concurrently, the implementation of associated modules or methodologies has been prompted by escalating apprehensions regarding service security, which may have a detrimental impact on the performance of CPS, particularly in terms of resource occupation and service overhead. To this end, we develop an end-edge-cloud-collaborative CPS framework in which tasks are modeled as workflow applications, followed by constructing a multi-stage service-enhanced method named SerFlow. In SerFlow, a relevance-aware security precaution mechanism is devised, which evaluates the connection hierarchy and correlation metrics across subtasks, subsequently establishing the service anti-conflict mechanism to augment security levels. Particularly, a comprehensive evaluation score for security precaution levels is provided, which enables the security precaution to engage in the following optimization operations and evaluation modules as a quantifiable metric. Leveraging the non-Euclidean geometry framework, we then develop a Pareto frontier modelling method that integrates Newton-Raphson and geodesic while a survival value evaluation strategy with correlation constraints is involved in accomplishing coarse-grained cluster selection. Subsequently, an improved value-based and model-free deep reinforcement learning algorithm is suggested to generate fine-grained service strategies in end-edge-cloud collaborative scenarios. Finally, comprehensive experiments demonstrate the effectiveness of SerFlow in achieving enhanced security precaution levels while maintaining superior service efficiency.

Instruction tuning is a vital step of training large language models (LLMs), so how to enhance the effect of instruction tuning has received increased attention. Existing works indicate that the quality of the dataset is more crucial than the quantity during instruction tuning of LLMs. Therefore, recently a lot of studies focus on exploring the methods of selecting high-quality subset from instruction datasets, aiming to reduce training costs and enhance the instruction-following capabilities of LLMs. This paper presents a comprehensive survey on data selection for LLM instruction tuning. Firstly, we introduce the wildly used instruction datasets. Then, we propose a new taxonomy of the data selection methods and provide a detailed introduction of recent advances, and the evaluation strategies and results of data selection methods are also elaborated in detail. Finally, we emphasize the open challenges and present new frontiers of this task.

ABSTRACT Microservice bad smells, arising from poor design and development practices, can severely degrade system quality if unaddressed. While rule‐based detection methods exist, their applicability is limited by subjective metric thresholds and the difficulty in defining certain bad smells, particularly complex microservice bad smells that are challenging to express through rules or involve high subjectivity. These smells often involve multiple services or manifest across multiple layers within a service, making them particularly challenging to detect using traditional methods. Without efficient and accurate detection mechanisms, the self‐healing capabilities of microservices during operation and continuous evolution will also be compromised. Given the promise of machine learning in code smell detection, this study empirically evaluates its performance in detecting complex microservice bad smells. We employ two sampling techniques and eight classification models on 1180 samples from 55 systems, generating 45 detection models and identifying top classifiers for seven complex microservice bad smell types. We compare machine learning with rule‐based methods for high‐subjectivity smells, analyze performance gaps, and propose a MAPE‐K‐based conceptual framework for runtime detection and refactoring. Finally, we discuss the necessity for future research.

In recent years, the development of online social network platforms has led to increased research efforts in social recommendation systems. Unlike traditional recommendation systems, social recommendation systems utilize both user-item interactions and user-user social relations to recommend relevant items, taking into account social homophily and social influence. Graph neural network (GNN)-based social recommendation methods have been proposed to model these item interactions and social relations effectively. However, existing GNN-based methods rely on centralized training, which raises privacy concerns and faces challenges in data collection due to regulations and privacy restrictions. Federated learning has emerged as a privacy-preserving alternative. Combining federated learning with GNN-based methods for social recommendation can leverage their respective advantages, but it also introduces new challenges: (1) existing federated recommendation systems often lack the capability to process heterogeneous data, such as user-item interactions and social relations; (2) due to the sparsity of data distributed across different clients, capturing the higher-order relationship information among users becomes challenging and is often overlooked by most federated recommendation systems. To overcome these challenges, we propose a federated social recommendation approach with enhanced hypergraph neural network (HGNN). We introduce HGNN to learn user and item embeddings in federated recommendation systems, leveraging the hypergraph structure to address the heterogeneity of data. Based on carefully crafted triangular motifs, we merge user and item nodes to construct hypergraphs on local clients, capturing specific triangular relations. Multiple HGNN channels are used to encode different categories of high-order relations, and an attention mechanism is applied to aggregate the embedded information from these channels. Our experiments on real-world social recommendation datasets demonstrate the effectiveness of the proposed approach. Extensive experiment results on three publicly available datasets validate the effectiveness of the proposed method.

The diversified and personalized service matching provided by the metaverse is hindered by challenges such as sparse interaction data, the long-tail effect, and the fragmentation of user demand profiles. Knowledge graph (KG)-based service recommendation methods alleviate these issues by incorporating external knowledge related to service items. However, their effectiveness is often compromised by irrelevant KG entities and relations. This irrelevant knowledge noise leads to two core chal lenges: (1) incomplete encoding of service knowledge into user and item embeddings, and (2) the absence of reliable true noise labels for distinguishing low-relevance knowledge information. To overcome these challenges, we propose MSKD, a metaverseser vice recommendation framework based on knowledge denoising. MSKD first constructs a user preference KG by fusing service item external knowledge with interaction data, then employs an adaptive noise pruning module to eliminate irrelevant entities. Subsequently, it constructs distinct Interaction and Preference KG views and aligns their embeddings within a shared repre sentation space through contrastive learning, thereby drawing semantically related items closer. To further remove noise without ground-truth labels, a KG bottleneck mechanism optimizes the information flow, retaining only knowledge most relevant to the recommendation task. Finally, a multi-task learning strategy jointly optimizes recommendation loss, contrastive alignment, and denoising objectives. Extensive experiments on three public service recommendation datasets show consistent improvements in top-K metrics (e.g., an average Recall@K increase of nearly 6.55% on Alibaba-iFashion dataset), demonstrating MSKD's effectiveness in denoising and its impact on enhancing metaverse service recommendations.

As information technology has been enabling services, the modern service industry, a cornerstone of the global economy, has driven development of the era of the service economy. According to the World Bank, since the 21stcentury, the service sector's share of global GDP has consistently been between 61% and 65%. “Everything as a service” has become a phenomenon in human society. Education on service science and engineering has been developed to cultivate more young people for the emerging service industry and service society.

Service Science and Engineering play an important role in service economy. In order to meet the demands of professionals for modern service industry, service-oriented manufacturing, and service-oriented society, it is necessary to cultivate many innovative service professionals who understand information technologies, management, and services. Education on service science and engineering has been established and developed for more than twenty years. This paper summarizes the connotations and development process of education on service science and engineering, shows the relevant knowledge-bodies and curricula for service science and engineering, introduces the best practice and experiences of service science and engineering education, and presents the considerations and suggestions on the future development of service science and engineering education.

Parameter-Efficient Fine-Tuning (PEFT) adapts large language mod- els (LLMs) to specific domains by updating only a small portion of the parameters. To easily and efficiently adapt LLMs to custom domains, we present a no-code fine-tuning platform, GongBu, sup- porting 9 PEFT methods and open-source LLMs. GongBu allows LLM fine-tuning through a user-friendly GUI, eliminating the need to write any code. Its features include data selection, accelerated training speed, decoupled deployment, performance monitoring, and error log analysis. The demonstration video is available at https://www.youtube.com/watch?v=QuDR_WNoB9o.

With the development of mobile internet platforms, recommendation systems have become crucial in various fields. However, they face several limitations: (1) Traditional recommendation methods focus solely on basic user profile, overlooking other behavioral characteristics of individuals, such as behavioral inertia; (2) Before making a recommendation, they lack a systematic methods to generate emergencies or events that users are concerned about. (3) They neglect collaboration among different agents, including users, scenes and emergencies. In this situation, when emergencies or other events occur, it will be difficult to actively provide behavioral decision recommendations for the current scene. To address these existing challenges, we propose our multi-collaborative agent recommendation model: (1) We introduce a user profiling definition that takes into account behavioral inertia; (2) We propose the CBEOC (Campus-Based-Element-Object-Consequence) model to systematically generate emergencies; (3) We propose an active recommendation strategy based on multi-collaborative agent. We then collect data from multiple universities based on psychological theories to demonstrate the effectiveness of our strategy. Finally, we propose a technical framework for practical application.

Deyuan Liu, Zhanyue Qin, Hairu Wang, Zhao Yang, Zecheng Wang, Fangying Rong, Qingbin Liu, Yanchao Hao, Bo Li, Xi Chen, Cunhang Fan, Zhao Lv, Dianhui Chu, Zhiying Tu, Dianbo Sui. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Zhanyue Qin, Haochuan Wang, Deyuan Liu, Ziyang Song, Cunhang Fan, Zhao Lv, Jinlin Wu, Zhen Lei, Zhiying Tu, Dianhui Chu, Xiaoyan Yu, Dianbo Sui. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Knowledge graph completion (KGC) tasks have been developed to address the inherent incompleteness of KGs. Recently, knowledge graph embedding (KGE) methods have gained popularity for embedding entities and relations, proving effective in KGC. However, privacy concerns make it challenging to collect privacy KG data from different institutions in the actual application. Federated learning has emerged as a solution for training models with decentralized data, eliminating the need for collecting private data. However, existing federated KGE methods overlook the implicit graph structural information of entities and relations, resulting in fragmented and incomplete representations within federated clients. Moreover, these methods often struggle with capturing multiple relational representations. To address these challenges, we propose a Federated Graph to Embedding (FedGE) approach based on encoder-decoder to capture interactions among entities and relations. Extensive experiments on two common KG datasets demonstrate the superiority of our method. The code is available at https://github.com/s460305450/FedGE.git.

In the context of dynamic service ecosystems, the inability of conventional knowledge graph embedding (KGE) methods to efficiently update incremental knowledge poses a significant challenge for the effectiveness of intelligent web applications. To address the continuous updating challenges of service knowledge, this paper introduces MetaHG, a meta-learning strategy for KGE. Unlike existing meta-learning KGE studies that focus solely on local entity information, MetaHG incorporates both local and potential global structural information from current snapshot’s seen knowledge graphs (KGs) to mitigate issues such as spatial deformation and enhance the representation of unseen entities. Our approach initializes entity embeddings using ‘in’ and ‘out’ relationship matrices and refines them through a hybrid graph neural network (GNN) framework, which includes a GNN layer for local information and a hypergraph neural network (HGNN) layer for potential global information. The meta-learning strategy embedded in MetaHG effectively transfers meta-knowledge for the accurate representation of emerging entities. Extensive experiments are conducted on a self-collected clothing industry service dataset and two publicly available open-source KG datasets. By comparing with several baselines, experiment results demonstrate the superior performance of MetaHG in generating high-quality embeddings for emerging entities and dynamically updating service knowledge.

Traditional methods of interacting with web pages, such as clicking and scrolling, greatly hinder users, especially those with disabilities and the elderly from conveniently accessing web services. By following user instructions, automatic web service navigation agents accomplish complex tasks on the websites, which is a natural interactions with web services. To study this task, previous works constructed simple web pages within simulated environments, but the realistic websites are far more intricate in true environments. Moreover, existing methods for this task rely on manually collecting human demonstrations on the given websites, which is time-consuming and labor-intensive, and reduce the generalization ability of service agents to unseen websites. Thus, we construct the first Chinese multimodal benchmark for web services navigation under the realistic settings: across domains and without human demonstrations. Our benchmark comprises a dataset (RealWeb) and a baseline method (WeServe). RealWeb consists of 40 real-world websites, 110 pages, and 11,739 language instructions. To detect and understand the feasible operations of pages in the visual mode, the screenshot of each page is annotated with 5 critical areas in RealWeb. WeServeis a multi-modal framework for web services navigation that combines visual and textual information, enabling universal navigation on any web pages with a success rate of 68.61%. Our dataset and codes are available at https://gitee.com/plabrolin/real-web.

Multimodal large language models (MLLM) have been observed to exhibit biases originating from their training datasets. Unlike unimodal LLMs, biases in MLLMs may stem from interactions between multiple modalities, which increases the complexity of multimodal debiasing. Conventional approaches like fine-tuning to alleviate biases in models are costly and data-hungry. Model editing methods, which focus on post-hoc modifications of model knowledge, have recently demonstrated significant potential across diverse applications. These methods can effectively and precisely adjust the behavior of models in specific knowledge domains, while minimizing the impact on the overall performance of the model. However, there is currently no comprehensive study to drive the application of model editing methods in debiasing MLLM and to analyze its pros and cons. To facilitate research in this field, we define the debiasing problem of MLLM as an editing problem and propose a novel set of evaluation metrics for MLLM debias editing. Through various experiments, we demonstrate that: (1) Existing model editing methods can effectively alleviate biases in MLLM and can generalize well to semantically equivalent image-text pairs. However, most methods tend to adversely affect the stability of the MLLM. (2) Compared to editing the visual modality of the MLLM, editing the textual modality yields better results in addressing MLLM biases. (3) Model editing based debiasing method can achieve generalization across different types of biases.

Temporal graph neural network (GNN) has recently received significant attention due to its wide application scenarios, such as bioinformatics, knowledge graphs, and social networks. There are some temporal GNNs that achieve remarkable results. However, these works focus on future event prediction and are performed under the assumption that all historical events are observable. In real-world applications, events are not always observable, and estimating event time is as important as predicting future events. In this article, we propose MTGN, a missing event-aware temporal GNN, which uniformly models evolving graph structure and timing of events to support predicting what will happen in the future and when it will happen. MTGN models the dynamic of both observed and missing events as two coupled temporal point processes (TPPs), thereby incorporating the effects of missing events into the network. Experimental results on several real-world temporal graphs demonstrate that MTGN significantly outperforms the existing methods with up to 89% and 112% more accurate time and link prediction. Code can be found on https://github.com/HIT-ICES/TNNLS-MTGN.

Dynamic link prediction has become a trending research subject because of its wide applications in the web, sociology, transportation, and bioinformatics. Currently, the prevailing approach for dynamic link prediction is based on graph neural networks, in which graph representation learning is the key to perform dynamic link prediction tasks. However, there are still great challenges because the structure of graphs evolves over time. A common approach is to represent a dynamic graph as a collection of discrete snapshots, in which information over a period is aggregated through summation or averaging. This way results in some fine-grained time-related information loss, which further leads to a certain degree of performance degradation. We conjecture that such fine-grained information is vital because it implies specific behavior patterns of nodes and edges in a snapshot. To verify this conjecture, we propose a novel fine-grained behavior-aware network (BehaviorNet) for dynamic network link prediction. Specifically, BehaviorNet adapts a transformer-based graph convolution network to capture the latent structural representations of nodes by adding edge behaviors as an additional attribute of edges. GRU is applied to learn the temporal features of given snapshots of a dynamic network by utilizing node behaviors as auxiliary information. Extensive experiments are conducted on several real-world dynamic graph datasets, and the results show significant performance gains for BehaviorNet over several state-of-the-art (SOTA) discrete dynamic link prediction baselines. Ablation study validates the effectiveness of modeling fine-grained edge and node behaviors.

Cloud and edge computing have been widely adopted in many application scenarios. With the increasing demand of fast iteration and complexity of business logic, it is challenging to achieve rapid development and continuous delivery in such highly distributed cloud and edge computing environment. At present, the microservice-based architecture has been the dominant deployment style, and a microservice system has to evolve agilely to offer stable Quality of Service (QoS) in the situation where user requirement changes frequently. A lot of research have been conducted to optimally re-deploy microservices to adapt to changing requirements. Nevertheless, complex dependencies between microservices and the existence of multiple instances of one single microservice in a microservice system together have not been fully considered in existing work. This article defines SPPMS, the Service Placement Problem in Microservice Systems that featurecomplex dependenciesandmultiple instances, as a Fractional Polynomial Problem (FPP). Considering the high computation complexity of FPP, it is then transformed into a Quadratic Sum-of-Ratios Fractional Problem (QSRFP) which is further solved by the our proposed greedy-based algorithms. Experiments demonstrate that our models and algorithms outperform existing approaches in both qualities of the generated solutions and computation speed.

The outbreak of Covid-19 has exposed the lack of medical resources, especially the lack of medical personnel. This results in time and space restrictions for medical services, and patients cannot obtain health information all the time and everywhere. Based on the medical knowledge graph, healthcare bots alleviate this burden effectively by providing patients with diagnosis guidance, pre-diagnosis, and post-diagnosis consultation services in the way of human-machine dialogue. However, the medical utterance is more complicated in language structure, and there are complex intention phenomena in semantics. It is a challenge to detect the single intent, multi-intent, and implicit intent of a patient’s utterance. To this end, we create a high-quality annotated Chinese Medical query (utterance) dataset, CMedQ (about 16.8k queries in medical domain which includes single, multiple, and implicit intents). It is hard to detect intent on such a complex dataset through traditional text classification models. Thus, we propose a novel detect model Conco-ERNIE , using concept co-occurrence patterns to enhance the representation of pre-trained model ERNIE. These patterns are mined using Apriori algorithm and will be embedded via Node2Vec. Their features will be aggregated with semantic features into Conco-ERNIE by using an attention module, which can catch user explicit intents and also predict user implicit intents. Experiments on CMedQ demonstrates that Conco-ERNIE achieves outstanding performance over baseline. Based on Conco-ERNIE, we develop an intelligent healthcare bot, MedicalBot . To provide knowledge support for MedicalBot , we also build a Chinese medical graph, CMedKG (about 45k entities and 283k relationships).

An increasing number and diversity of services are available, which results in significant challenges to effectively reuse service during mashup creation. Many works have modeled the mashup creation problem as a service recommendation task and have achieved remarkable results. However, the performance of these methods can be further improved. The main problems affecting these methods include the constraints among recommended services, the evolution of services, and the semantic gap existing in services and mashups. In this article, we model the mashup creation problem as a service bundle recommendation task that is formally defined to address the constraints among recommended services. And then, a dynamic graph neural network based model called DySR is proposed to tackle the evolution of service and the semantic gap between services and mashups. In order to quantitatively measure how significant the semantic gap between mashups and services is, a measurement method of a semantic gap is given. With it, experiments show that to what extent DySR can reduce the semantic gap in the context of mashup creation. In addition, new evaluation metrics are introduced to overcome the preference for popular services in traditional service recommendations. Extensive experiments conducted on a real-world dataset from ProgrammableWeb, and the experiment results show that DySR outperforms existing state-of-the-art methods.

Classical services computing tasks such as service design and service recommendation need to comprehensively consider objective Quality of Services (QoS) and subjective Quality of Experiences (QoE) of users. There are close relationships between QoS and QoE, and how to construct an accurate QoS/QoE correlation model has been a hot topic in academia for years. Particularly, it is a challenge to construct such a model for complex composite services that are composed of services and their corresponding providers from multiple domains. This is because the number of QoS parameters is huge while the number of QoE parameters is comparatively smaller, and consequently, to reasonably encode the imbalanced QoS and QoE parameters of composite services becomes challenging. In addition, different users have different concerns and personalized experiences on the same service, and the QoS/QoE correlation model should be personalized, too; however, traditional end-to-end models which simply use QoS as input and QoE as output ignore such personalized preferences of different users, thus the model accuracy is not high enough. Based on the transformer pre-trained language model, this paper mines users’ fine-grained concerns and their sentiment polarity from comments. Then, personalized preferences of users are encoded with CNN, QoS of composite services are encoded with multi-layer Bi-LSTM, and the QoS/QoE correlation is established based on the attention mechanism. In the experiments, our model achieves the highest on the accuracy of sentiment polarity prediction of user concerns, and the QoS and QoE encoded by the proposed model can accurately express the differentiated preferences of different users in a concrete composite service scenario. Potential downstream applications of the proposed QoS/QoE correlation model are comprehensively discussed.

With the widespread use of Virtual Personal Assistants (VPAs) in people’s daily lives, research on conversational recommender systems (CRSs) has become a hot topic in services computing. A key challenge is to reduce search space of candidate services for providing satisfying recommendations in fewer dialogue rounds. Firstly, user portraits are used to record long-term user preferences for assisting personalized recommendations; however, traditional label-based user portraits cannot reach at higher precision of quantified preferences and higher timeliness of preference updates along with time. For this challenge, this paper imports Multi-domain Dialogue-based User Portrait (MDUP) model to record multi-domain numerical preferences and continuously accumulates long-term preference updates, so that candidate services can be matched more precisely. Secondly, recommender systems usually compose massive services from multiple domains to meet coarse-grained user needs, and in order to improve the search, filtering and sorting efficiency of services during dialogue, this paper models potential connections among services of different domains by introducing service feature dependency to effectively compress the scale of candidate services. Combined the two perspectives together, this paper proposes a new service pre-sorting algorithm and a novel dialogue policy (intra-domain feedback-based iteration and inter-domain backtracking), then introduces architecture and working process of a novel service recommendation VPA (srVPA). Experiment demonstrates that compared with existing CRSs, the importation of MDUP and service feature dependency into srVPA significantly improve dialogue efficiency and user satisfaction on recommendation results.

Many cross-domain and cross-region services and applications have flooded the Internet, such as API services, IoT services, etc. These services are interconnected and form a new service ecosystem, the Internet of Services (IoS). Through IoS, service providers have a broader platform to deliver their services to users. Users can ask for a composite service solution without finding multiple suitable services themselves. However, IoS requires a standard modeling language and methodology to provide developers with a unified understanding of IoS. This modeling methodology should guide developers through the IoS modeling task. Therefore, we propose an Open System Architecture for the Internet of Services (IoS-OSA). We divide IoS into a three-dimensional cube according to layers, views, and lifecycle dimensions, with 64 different views. It integrates the existing scattered work of IoS into a unified framework and introduces the content description and the recommended modeling specification of each view. Finally, we propose the corresponding modeling methods for several key views in IoS-OSA such as service ecosystem, value/quality/capability (VQC) of IoS, etc., which improve the overall architecture of IoS.

Although many principles have been put forward to guide microservice design, such as Domain-Driven Design, Architectural Bad Smells (ABS) would be inevitably imported during microservice system design and development. There has been a consensus that the existence of ABSs would bring negative effects to microservice systems. Although some approaches use static analysis to detect ABSs in the design phase, we conjecture that the design-phase ABSs would result in some performance deficiencies at runtime. This paper conducts an empirical study on underlying correlations between runtime performance deficiencies and ABSs in microservice system design. An automated experimental stress testing framework called MRSTF is developed for automatic deployment of MSS and collecting/analyzing runtime performance data. A microservice system TrainTicket is used in this empirical study. We manually inject several typical ABSs into the system and use MRSTF to compare the runtime performances before and after the injection and check if the existence of ABSs has significant effects on the runtime performance. Experiment results show that ABSs have significant negative effects on some performance metrics of runtime MSS while doing not on others. This study provides solid evidence on the feasibility of optimizing MSS design by eliminating ABSs based on runtime performance data.

With the development of human-machine dialogue technology, more and more companies have launched their cognitive service products, such as Virtual Personal Assistant (VPA), smart speakers, shopping guide robots, etc. However, in these practical applications, most of the bots passively respond to user's utterances, lacking user preference knowledge and the proactive consciousness to lead the dialogue. Therefore, it is essential that bots proactively and naturally lead the dialogue from chitchat to service recommendation to meet user's requirements. To address this challenge, bots not only needs to detect the user's dialogue goal in real time, but also needs to plan a goal sequence based on user profile. In this paper, we propose DGPF, a Dialogue Goal Planning Framework. DGPF plans a reasonable goal sequence grounded on user's interests and personal KB before the conversation, additionally predicts user's true intent (i.e. dialogue goal) and judges whether the goal is completed based on the utterances during the conversation. DGPF includes a novel joint learning model that can simultaneously fix the two sub-tasks of goal completion estimation as well as current goal prediction, and improve each other's performance interactively. Our experimental results on the open dataset DuRecDial have been significantly improved compared to the baseline, which proves the effectiveness of our framework.

With the rapid development of new technologies such as cloud, edge and mobile computing, the number and diversity of available services are dramatically exploding and services have become increasingly important to people's daily work and life. As a consequence, using service label recommendation techniques to automatically categorize services plays a crucial role in many service computing tasks, such as service discovery, service composition, and service organization. There have been many service label recommendation studies that have achieved remarkable performance. However, these studies mainly focus on using the text information in service profiles to recommend labels for services while overlooking those social relations that widely exist among services. We argue that such social relations can help to obtain more precise recommendation results. In this paper, we propose a novel Social Relation aware Service Label Recommendation model called SRaSLR, which combines text information in service profiles and social network relations among services. A deep learning based model is constructed based on feature fusion of the two perspectives. We conduct extensive experiments on the real-world Programmable Web dataset, and the experiment results show that SRaSLR yields better performance than existing methods. Additionally, we discuss how service social network affects service label recommendation performance based on the experiment results.

In the Internet of Services (IoS), service platforms need to integrate services from multiple domains and various service providers to fulfill user requirements. Considering services and service solutions (composed of services) in IoS are complicated, a big challenge is how to design and implement a system to effectively monitor, measure and evaluate the quality of the services and solutions. For this challenge, we introduce a framework and develop a platform for service providers to establish a VQC (value, quality, capability) index hierarchy and efficiently obtain running indices of services and solutions in IoS. The main idea is to divide the indices into three types: value, quality, and capability. Then, service providers can establish a dynamic hierarchy that defines the calculation relationship between indices as well as binds the indices to corresponding services and solutions by this platform so that the system can monitor services and solutions in run-time. Real-time data would be collected and measured. The design and implementation of the platform supporting the above process are introduced in detail. Finally, a case study is proposed to demonstrate the framework and system.

With the application of more and more new technologies, such as cloud computing, IoT, and big data, the available intelligent service scenarios become complex. For better service identification and recommendation, conversational AI bots need to construct user needs expression models accurately. Most traditional user needs expression models could do better in individual tasks but nearly not support the social requirements. To solve the challenge before, this study expands the Intention Tree Model for Personal Needs(ITM-PN) to the Intention Tree Model for Social Needs based on Negotiation(ITM-SN) to model different users' needs during the process of multi-user negotiation(initial, processing, ending). And finally, this study conducted a case study to prove the ITM-SN is thoroughly and effectively, which could be widely applicable in many fields.

In recent years, soft service robot(e-SBOT) that can exist in any intelligent terminal have received widespread attention. However, currently the services provided by e-SBOT are extremely limited. This defect mostly dues to two problems, one is that the inaccuracy of capturing user’s need caused by incomplete understanding, the other is that the richness, diversity and complexity of service resources make it difficult to select appropriate resources. The existing service resource library is mainly based on the digital world and lacks a unified description and management of service resources both in digital and physical world. In addition, it is mostly useless to recommend service resources that do not match the user’s current scenario, resulting in a worse experience. Therefore, it’s essential to describe the service resources in combination with the scene and provide the service resources that adapted to the scenario to users in a more accurate and rapid way. To achieve this, we propose a scenario-based service resource description model based on ontology modeling. A multi-view retrieval method is simultaneously developed for rapid retrieval of service resources. Experimental results show that our proposed model and method can improve the retrieval speed and accurately return service resources suitable for user scenarios.

The mutual invocation and composition of services conduct the Internet of Services (IoS). Dynamic QoS is a prominent feature of IoS. Existing QoS-aware methods always consider the QoS as a static indicator to differentiate similar services, and attributes of QoS are independent of each other, which may lead to poor performance of service selection. In this paper, we proposed a real-time QoS-aware service selection method to find good service to match the users’ requirements, in which an FM-based dynamic evaluation model was employed to predict the real-time QoS, and then a hash space segment-based service matching algorithm is adopted to select service resources rapidly. Extensive experimental results show that mining the dynamic secondary characteristic relationship among the QoS attributes can improve the service selection effect on the real-world scene.

The distributed services from different domains, organizations, and regions in the real and virtual world are converged together to form the Internet of Services (IoS). It is required to provide a trusted, orderly and efficient platform environment for service collaboration and delivery. Blockchain is currently recognized as the best practice technique to solve the problem of orderly and trusted execution of system. However, the efficiency issue introduced by it is also a problem that every platform needs to take seriously. Therefore, this paper proposes a blockchain-based IoS architecture, which uses the virtual chain and the embedded EVM(Ethereum Virtual Machine) notary technology to ensure the security and credibility of the service collaboration process in the IoS, and also the efficient operation of the entire system. This architecture has been compared with non-blockchain architecture without credible guarantees and also blockchain-only architecture. This comparative experiment shows that the proposed architecture not only ensures the orderly and trusted execution of the system, but also effectively controls the loss of performance.

In recent years, chat-bot has become a new type of intelligent terminal to guide users to consume services. However, it is criticized most that the services it provides are not what users expect or most expect. This defect mostly dues to two problems, one is that the incompleteness and uncertainty of user's requirement expression caused by the information asymmetry, the other is that the diversity of service resources leads to the difficulty of service selection. Conversational bot is a typical mesh device, so the guided multi-rounds Q&A is the most effective way to elicit user requirements. Obviously, complex Q&A with too many rounds is boring and always leads to bad user experience. Therefore, we aim to obtain user requirements as accurately as possible in as few rounds as possible. To achieve this, a user intention recognition method based on Knowledge Graph (KG) was developed for fuzzy requirement inference, and a requirement elicitation method based on Granular Computing was proposed for dialog policy generation. Experimental results show that these two methods can effectively reduce the number of conversation rounds, and can quickly and accurately identify the user intention.

Ever since smart speakers became popular, the functions that can help users complete a series of tasks through voice interaction are called “skills”. The market integrates all “skills” is called “skill market”. There is a serious imbalance in the distribution of hot spots and user concerns in the skill market, and the research on the distribution of user needs satisfied by skills and points of interest(POI) that users pay attention to is insufficient. User needs and POIs are contained in unstructured data, in order to analyze the distribution of user needs and POIs from unstructured data, this paper conducted an empirical study that used the BERT multi-label classification model to extract the user needs that meets the Maslow's hierarchy of needs from the skill description, and used RAKE algorithm to extract user POIs from user reviews and used knowledge graph to extract the relationships between POIs. Using the analysis results of the extracted data, the paper gives suggestions related to the development direction and POIs that should pay attention to in development for skill developers.

Services flourish drastically along with the trend of Everything as a Service (EaaS). Over the years, services have become more and more interconnected to facilitate trans-boundary collaboration in business and technology levels for creating and delivering distinct new values to customers. This phenomenon is called Internet of Services (IoS). IoS arises and evolves in terms of tremendous discrete decisions made by individual service stakeholders, and its structure and evolution show quite similar characteristics of the ones of natural ecosystems. We propose a novel multi-layer network model to describe the static structure and dynamic evolution traces of service ecosystems. Different from traditional models which are usually focused on specific service types and granularity levels, our model comprehensively covers the interconnections between stakeholders, services delivery channels, service functional and non-functional features and business domains. “Event” and “event chains” which are extracted from public news corpus are imported into the model as triggers of service ecosystem evolution and evolutionary relations are introduced into the model delineate evolution traces. Elements, relations, and application scenarios of this model are introduced in detail. Qualitative comparison with state-of-the-art models shows that our model has a higher degree of coverage of fine-grained elements/relations of service ecosystem and higher interpretability. Supplementary materials are provided to help understand how to build the model based on external data. In our opinion, this work lays a solid base for the exploration of structural characteristics and evolution laws of IoS.

“Transboundary” has become a common phenomenon in today's Internet-based service industries. Service providers are not only focused on their original business domains but also extend their attentions to other domains. By aggregating and orchestrating services from different domains to form complex innovative transboundary services, distinct new values are created and delivered to customers. Along with increasing interconnections among services from different domains, many novel transboundary service ecosystems (TSE) emerge. In this paper, we give a six-layered conceptual model for TSE, including atomic service layer, organization layer, region layer, inner-domain layer, cross-domain layer, and customer demand layer. Types of competitions and collaborations in TSE are elaborately refined. A set of ecological metrics for evaluating state of TSE constituents, state of TSE as a whole, and distribution of customer demands, are emphatically discussed. This work gives a high-level inspection on transboundary services from an ecological standpoint, and is of great significance for service providers to identify innovation opportunities in a service ecosystem, and for regulators to issue laws and regulations to guide the healthy development of transboundary service industries.

Nowadays, consumers have a variety of service requirements which are strongly correlated. A single service provider has not enough capabilities to meet various functional and nonfunctional requirements, thus giving rise to Crossover Services. Traditional modeling approaches focus more on common model elements and their relationships, but rarely make recommendations on what modeling methods are appropriate for specific service scenarios. Therefore, this paper summarizes the objective problems faced by Crossover Services modeling, and proposes a collaborative modeling solution, which aims to make full use of each participant's domain expertise and implement complete modeling efficiently and flexibly. Based on the global and local hierarchical dependency of crossover collaboration, the method divides the modeling process into the schema layer and the instance layer, then establishes mappings between global service choreography process and local business execution processes with a series of conversational interfaces exposed by different service participants, and finally models the hierarchical constraints from the abstract decision layer to the concrete implementation layer. At the end, through the analysis of a rural e-commerce service modeling case, it is shown that this method can make full use of the existing domain service specifications and enterprise service standards. Moreover, it has certain advantages in modeling accuracy and comprehensiveness, efficiency and flexibility.

In order to solve the quality issues caused by the status change of the internal elements as well as the external environment while transboundary service system is running, this paper proposes a root-cause analysis and fine tuning method for the run-time quality issues of transboundary services. After receiving feedback on the run-time quality issues, this method first built a causal diagram model for root-cause analysis on the basis of a large amount of quality and capability data generated by the operation of transboundary service system, and then based on the constructed causal diagram model, locate the problematic quality/capability parameters by tracing back to the source of quality issues, and finally aiming at the problematic quality/capability configuration scheme, carry out fine tuning through the data mining method that combines data clustering with decision tree, so as to ensure the continuous and healthy operation of transboundary service system with the minimum tuning cost. Finally, the effectiveness and practicability of the method is verified through the case of the Rural taobao service.

Because of rapid growth in mobile application markets, competition between companies that provide similar applications has become fierce. To improve user satisfaction for keeping existing users and attracting new users, application developers need to quickly respond to customer feedback regarding functionality and performance defects. In software engineering, specifying an accurate evolution plan according to user feedback is useful but quite difficult. Hence, we propose an approach for predicting and recommending evolution plans to application developers that includes: (1) when a new version of an App should be released; (2) which features should be updated in the next version and (3) if a new version is released, to what degree users would like or dislike it. This approach is based on an elaborate text analysis of massive numbers of user reviews and App update histories. A collocation-based mRAKE method is presented to extract requested and updated features from user reviews and update logs, and the intensity and sentiment scores of each feature are calculated to quantitatively represent time-series histories of App updates and user requests. Machine learning algorithms including linear support vector, Gaussian naïve Bayes and logistic regression are employed to discover the underlying correlation between user opinions embedded in their reviews and the App update behaviors of developers, and rich experiments were conducted on real data to validate the effectiveness of the proposed approach. Overall, our approach can achieve an average accuracy of 72.8% and 93.7% in release time recommendation and content updates of successive versions, respectively, and it can predict user reactions to a planned version with an average accuracy of above 89.0%.

In order to improve the user oriented service delivery, we present a soft service robot, e-SBOT. It plays the "bridge" role between a user and massive external services. By sensing explicit and implicit intentions and demands of a user, it helps find appropriate services and construct coarse-grained service solutions. Both global and personal knowledge graph are utilized to for intention reasoning and service solution planning. Sensors are designed for the dynamic changes of external services and resources. By e-SBOT, the cognition of a user could be amplified to a large extent. This paper gives a brief introduction to e-SBOT's technical challenges, architecture design, and key theoretical problems.

The service technology and crowdsourcing movement have spawned a host of successful efforts that promote the rapid development of the human service ecosystem. In this ecosystem, a large number of globally-distributed freelancers are organized to tackle a range of tasks over the web. These crowdsourcing services provide convenience for civilians with lower price and shorter response time. However, the convenience cannot whitewash many unstable factors that are caused by human involvement, such as undefinable reputation, unstable quality, crowdturfing, and etc. In this paper, we present a comprehensive data-driven investigation of one prominent supply-driven human services marketplace-Fiverr-wherein we analyze freelancers' marketing behaviors and their offering services (called "gigs"). As part of this investigation, we identify the key features that can be used to evaluate freelancers' influence and develop a GSRC (Gig service property + Seller Impact + Customer Review + Semantic Content) model to predict gig service quality. As far as we know, this is the first attempt that involves the service semantic info in the prediction model and integrates all these four aspect factors.

In recent years, freelancer economy has been a new normalcy. In the supply-driven freelancer marketplace, people sell their capabilities or labor as service on the internet platform to help others with some particular micro-tasks. As this kind of human service ecosystem is at the fast growth stage, it is inundated with a variety of services whose quality is uneven. Quite often, when facing these services, customers hesitate to make the decision. The root causes of this hesitation are: (1) customers do not know these services well, even the explicit service category and description are provided, (2) customers do not know their own demands well. Most of the time, customers only have a general/fuzzy goal, but have no sense of the requirements in detail. Therefore, this study aims at proposing a human services recommendation method for fuzzy customer requirement. The experimental data of this study is collected from Fiverr.com, which is one prominent supply-driven human services marketplace. By analyzing the transaction data, any details of services, freelancers, customers, and their relations will be extracted to construct a supply-demand relation graph. In this study, customer's fuzzy requirement description will be transferred into a query subgraph, which is the input of an evolved subgraph matching algorithm. This algorithm will help to retrieve the recommendable services (combinations). In addition, a guided Q&A approach is designed to complement customer's fuzzy requirement, so that subgraph matching algorithm can retrieve better results.

A key issue in big service era is how to optimallyconfigure and operate existing service resources to serve variousrequirements of customers, i.e., how to reuse available servicecomponents to develop new service systems to fulfill massiveindividualized requirements effectively and efficiently. In termsof this challenge, in this paper we present a new paradigmfor software service engineering called RE2SEP (Requirement-Engineering Two-Phase Service Engineering Paradigm) which iscomposed of three components: a service-oriented requirementengineering approach, a domain-oriented service engineering approach,and a software service development approach. RE2SEPenables the rapid design and implementation of service solutionsto match the requirements of massive individualized customerswith the help of patterns, and potentially revolutionizes thetraditional software engineering paradigms. A case study on anelderly care service system is given to demonstrate the feasibilityof RE2SEP.