ABSTRACT Software development plays a crucial role in the modern mobile application domain, reflecting its significance through widespread application. With the continuous evolution and vast number of Android APIs, developers need to invest considerable effort in learning how to use various suitable APIs for their projects. Unfortunately, most current recommendation methods, when representing programs as source code sequences, abstract syntax trees, or API call paths, often focus only on contextual relationships while ignoring valuable information in API dependency relationships. Moreover, existing sequence models (such as RNN, LSTM) often fail to make correct predictions for low‐frequency API methods with high‐frequency suffixes, as these models tend to capture the most common API sequence patterns, causing these relatively low‐frequency but potentially more applicable APIs to be overlooked. To address this issue, we propose an API dependency path‐based Android API recommendation method, DPAPIRec. This approach combines program analysis with deep learning, which first extracts API methods and their data flow and control flow dependency relationships from a large number of Android APPs through program analysis techniques and then obtains a comprehensive API dependency paths repository. Finally, a deep learning method is applied to learn and represent these dependency relationships to improve API recommendation accuracy. Furthermore, to better extract dependency relationships, we employ an improved attention‐based LSTM model with a novel loss architecture, enhancing the global dependency relationships between APIs through a weighted mixed loss, thereby strengthening the weight of initial nodes and alleviating the problem of low‐frequency APIs with high‐frequency suffixes. Our experiments on the AndroZoo dataset demonstrate that DPAPIRec significantly outperforms baseline methods in Android API recommendation tasks, showing substantial improvements in both Accuracy and Mean Reciprocal Rank (MRR).

Increasingly low barriers to Internet applications allow a large number of ordinary users to become developers or users of Web services. However, confronted with massive services and complex application scenarios, users often struggle to filter out satisfactory services, in fact, even professional users find it difficult to describe their requirements specifically and accurately in many cases. In order to aggregate more feature information and mitigate the negative impact of low-quality user requirement description, we propose a novel group feature aggregation service recommendation framework (GFASR). Concretely, we first calculate the semantic similarity between users, and create a group for each user according to the similarity ranking. Furthermore, on the basis of learning neural embeddings of users, candidate services, and groups, we employ a dual-attention mechanism to capture effective feature (such as requirement description, service history invoked information, etc.) and preference information of group members for each user, thereby supplementing or enhancing the user’s feature representation. Finally, we aggregate and propagate the information of all embeddings, and a neural and attentional factorization machine model is used to recommend services for users. Comparative experiments on a real dataset demonstrate that our method significantly outperforms the state-of-the-art service recommendation models.

Programming language pre-training models have made significant progress in code representation learning in recent years. Although various methods, such as data flow and Abstract Syntax Tree (AST), have been widely applied to enhance code representation, there has been no research literature, up to date, specifically exploring the use of intermediate code of the source codes for code representation. For example, the intermediate code of Python, namely opcode, not only includes the data input and output stack processes during program execution, but also describes the specific execution order and control flow information. These features are not possessed in source code, data flow, AST and other structures or are difficult to directly reflect. In this paper, we propose OpCodeBERT1approach, which is the first to utilize Python opcode for code representation learning and improves code representation by encoding the underlying execution logic, comments, and source code. To support the training of opcode, we filter the public datasets to exclude unparsable data and innovatively propose an opcode-to-sequence mapping method to convert them into a form suitable for model input. In addition, we pre-train OpCodeBERT using a two-stage masked language modeling (MLM) and a multi-modal contrastive learning. To evaluate the effectiveness of OpCodeBERT, we have done experiment with multiple downstream tasks. The experimental results show that OpCodeBERT performs excellently on these tasks, validating the effectiveness of incorporating opcode and further demonstrating the feasibility of this method in code representation learning.

With the rapid development of service computing technology, how to recommend the desirable Web APIs to developers from the large number of APIs is a challenge. The traditional methods based on collaborative filtering are limited by data sparsity. With the support of multi-dimensional relational feature modeling, graph neural network-based methods are proposed to mitigate data sparsity, but their convolution nature may amplify the noise effect. Therefore, how to simultaneously reduce the impact of sparsity and noisy data has been an open question in the field of API recommendation. To address the problem, this paper proposes a self-supervised learning method based on multi-view hypergraph for Web API recommendation. First, the Mashup-API interaction graph is transformed into a hypergraph, and the hyperedges are used as intermediate hubs to transfer messages between nodes, maintaining the global collaboration effect between Mashup and API nodes. Then, a multi-view strategy is adopted to generate embeddings of Mashups and APIs through information fusion, by which recommendation probability is derived by dot product between the embeddings of Mashups and Web APIs. To train the model parameters effectively, a self-supervised learning method is used to reduce the effect of noisy data to improve the embeddings. Extensive experiments are conducted on a real-world dataset, and the experimental results show that the proposed model outperforms the baselines.

With the wide adoption of SOA (Service Oriented Architecture) in software engineering, a large number of Web services have emerged to meet the Mashup development requirements. Due to the existence of numerous Web services with similar or identical functionalities, it is challenging for users to select the appropriate Web API for Mashup creation, which makes Web service recommendation an effective approach. The performance of current FM-based service recommendation methods is limited by the sparsity of semantic features related to their functionalities. Furthermore, the network structure features of Web services are often overlooked. However, these features are of great importance and should be incorporated into the service recommendation process. Based on the above considerations, this paper proposes a service recommendation model which fuses content information and network information. Firstly, service content information and network structure information are extracted respectively. Then, these two types of information are characterized separately, and their functional semantics are extracted. Finally, the above information is fused and processed by Neural and Attention Factorization Machine to obtain the final recommendation results. Experimental results show that fusing service network representation information can effectively improve the accuracy of service recommendation results.

With the wide adoption of Web APIs released on Internet, users tend to reuse them for business requirements or software development. Mashup is a useful technology for composing Web APIs into a new and value-added application. With the increasing number of Web APIs and Mashups, the API-Mashup ecosystem has emerged based on the invocation relationship between Mashups and Web APIs. In this paper, we take ProgrammableWeb, a typical API-Mashup ecosystem, as an example to investigate its dynamic evolutionary analysis. Although there have been some works on the API-Mashup ecosystem, they mainly focus on static analysis, i.e., the static characteristics of the API- Mashup ecosystem on a fixed time point. This paper conducts a comprehensive study on the dynamic evolutionary analysis of the API-Mashup ecosystem with a long time range from 2005 to 2021. First, we conduct a dynamic statistical analysis based on the API-Mashup ecosystem dataset. Next, we construct two cooperation networks, one between Web APIs, and the other between their categories. And the general characteristics of the two cooperation networks are presented. Finally, we investigate the derived cooperation networks from four perspectives: dynamic characteristics, degree distribution, betweenness centrality, and assortative mixing. Meanwhile, the corresponding insights are uncovered. Our work provides a foundation for visualization and understanding of the API-Mashup ecosystem from the timeline.

Clustering Web services can improve the quality and efficiency of service discovery and management within a service repository. Nowadays, Web services frequently interact (e.g., composition relation and tag sharing relation) with each other to form a complex and heterogeneous service relationship network. The rich network relations inherently reflect either positive or negative clustering association between Web services, which can be a strong supplement to service semantics for characterizing functional affinities between Web services. In this paper, we propose to cluster Web services by utilizing both description documents and the structural information from the service relationship network. We first learn the content semantic information from service description documents based on the widely used Doc2vec model, and meanwhile, learn the structural semantic information from the service relationship network based on a network representation learning algorithm. Then, we propose to pretrain the content and structural semantic information to obtain the most relevant and unified features through training a service classification model with partially labeled data. Finally, a spectral clustering algorithm is utilized for Web services clustering based on the above unified features with preserved content and structural semantics. Therefore, the proposed services clustering approach takes advantage of both service content semantic and service network structure semantic based similarity between services. Extensive experiments are conducted on a real-world dataset from ProgrammableWeb, composed of 12919 Web API services. Experimental results demonstrate that our approach yields an improvement of 4.78% in precision and 5.4% in recall over the state-of-the-art method.

QoS prediction for Web Services is becoming increasingly important for various QoS-aware Web Services management tasks. However, the existing methods for QoS prediction of Web Services have some drawbacks, such as poor performance in dealing with data sparsity, insufficient consideration of latent information in user-service interaction behavior, and no consideration on discriminating the weight of latent information. To address these shortcomings, this paper proposes a QoS Prediction approach via combining multi-component graph convolutional collaborative filtering and deep factorization machine. A user-service bipartite graph is constructed, and the edges of the graph are decomposed into multiple latent spaces with node-level attention to identify latent components. Then, the importances of latent components are determined, and they are aggregated to obtain the corresponding user-service embedding vectors. Finally, the embedding vectors are taken as the input of a deep factorization machines model to obtain the prediction of unknown QoS. Extensive experiments are conducted on a real-world dataset. The experimental results demonstrate that MGCCF-DFM achieves superior prediction accuracy in terms of mean absolute error (MAE) and root mean square error (RMSE) compared with the existing QoS prediction techniques.

Learning network representations of Web services plays a critical role in the service ecosystem and facilitates many downstream tasks, e.g., service composition, service recommendation, service clustering, and service classification, etc. However, the performance of most of the existing approaches is limited by the sparse and non-interaction relationships between services. Considering these shortcomings, by proposing a balance theory based weighted signed graph convolutional network, we explore a dedicated signed service link prediction method to expand accurate links in service relation networks. Concretely, we first define the positive and negative links based on historical prior knowledge concerning services, and then construct a signed service relation network. Furthermore, on the basis of quantifying the influence of different neighbor nodes, we employ balance theory to correctly aggregate and propagate the information across layers through a weighted signed graph convolutional network. Finally, we splice all service embeddings in pairs, and a multi-layer perceptron classifier is used to predict the links between services. Comparative experiments with six baselines demonstrate that our method significantly outperforms the state-of-the-art link prediction models.

Abstract With the rapid growth of the number of Web services on the Internet, how to classify Web services correctly and efficiently become particularly important in service management tasks, such as service discovery, service selection, service ranking, and service recommendation. Existing functionality‐based service classification techniques have some drawbacks: (1) the keyword order and context information are not considered; (2) the embedding features of keywords are taken as equal importance to learn the classification model; (3) the topic number is hard to determine manually. Due to these drawbacks, the accuracy of service classification needs to be improved further. At present, deep learning techniques show the strong power in modeling complex and nonlinear function relationship. Thus, to address the problems above, this paper exploits attention mechanism to combine the local implicit state vector of Bidirectional Long Short‐Term Memory Network (BiLSTM) and the global hierarchical Dirichlet process (HDP) topic vector, and proposes a Web service classification approach with topical attention‐based BiLSTM. Specifically, BiLSTM is used to automatically learn the keyword feature representations of Web services. Then, the topic vectors of Web service documents are obtained with HDP by offline training, and topic attention mechanism is adopted to strengthen the feature representation by discriminating the importance or weight of different keywords in Web service documents. Finally, the enhanced Web service feature representation is used as the input of a softmax neural network layer to perform the classification prediction for Web services. Extensive experiments are conducted to validate the effectiveness of the proposed approach.

The wide adoption of Service Oriented Architecture (SOA) has driven the creation of a massive amount of applications on the Internet, which includes the popular Mashups composed from multiple existing Web APIs. The availability of a large number of Web APIs with diverse functionalities on the Web makes it difficult for users to find APIs meeting their needs for Mashup development. To relieve this difficulty, recommending Web APIs for Mashup development has become an effective solution. A dozen of service recommendation approaches were proposed based on multi-dimensional features extracted from the service repository over the last couple of years, e.g., similarity based matching methods, matrix factorization based models, and factorization machine based models. Among these existing works, Factorization Machine (FM) based models, in particular the deep learning based FM models, have shown better performance compared with other conventional collaborative filtering techniques. Despite their superiority, the deep learning based FMs still have some strong model assumptions that can harm the recommendation accuracy. For example, it models factorized interactions with the same weight and ignores the non-linear and complex inherent structure in data. In a real-world service recommendation scenario, different predictor variables usually have different predictive power and not all features are predictable for estimating the target. Also, higher-order feature interactions are usually underlain in complex user-service environments. To address these deficiencies, this paper proposes a hybrid factorization machine model with a novel neural network architecture, named NAFM, which integrates a deep neural network to capture the non-linear and complex feature interactions and uses an attention mechanism to capture the varying importance of feature interactions. Comprehensive experiments are conducted on a real-world dataset from ProgrammableWeb. The experimental results show that the proposed approach outperforms the existing state-of-the-art models for service recommendation.

Classifying Web services plays a critical role in several fundamental service management tasks, such as service discovery, selection, ranking, and recommendation. However, traditional Web service classification approaches usually difficult to dispose unstructured sparse documents and underutilize the rich network relations. The consideration of multiple document representation schemes can ameliorate the former problem, whereas an appropriate network representation method could be a positive solution to the latter problem. In this paper, we propose a general Web service classification framework via deep fusion of structured and unstructured features, named LDNM. Firstly, we transform each service document into feature vectors by using two document representation methods: topic distribution based on LDA, and neural-network-based document embedding model known as Doc2vec. Then we obtain structured representation vectors which stem from service invoking and tagging graphs by applying Node2vec. Finally, we fuse these features and train a service classifier by using an MLP neural network. Comprehensive experiments are conducted on real-world datasets to demonstrate the effectiveness of the proposed approach.

With the explosive growth of Web APIs on the Internet, it is a challenge to recommend desirable Web APIs from multiple ecosystems to develop a Mashup. Most existing API service recommendation methods focus on functional semantic similarity, but underutilize the rich network relations which inherently reflect either positive or negative relevance between services. Moreover, in the recommendation process, they usually pay too much attention to the interactions between Mashups and APIs, but ignore the cooperation between APIs. In this paper, we propose a novel method named SRAWN (Structure Reinforcing and Attribute Weakening Network) based API recommendation approach for Mashup creation. Specifically, we first design a feature extractor layer to capture structure relationship and attribute information from an API relation network graph by introducing a GAT2VEC framework, and obtain representation vectors corresponding to each API. Then, a matching evolving layer is proposed to capture the matching evolving process between APIs. At this layer, APIs are chosen incrementally to composite a Mashup, and the embedding vectors of the Mashup's existing composition features are updated adaptively based on diverse candidate APIs, by introducing a Deep Interest Network. Comprehensive experiments on a real-world dataset show that SRAWN outperforms the other state-of-the-art solutions.