Cost-aware routing dynamically dispatches user queries to models of varying capability to balance performance and inference cost.However, the routing strategy introduces a new security concern that adversaries may manipulate the router to consistently select expensive highcapability models.Existing routing attacks depend on either white-box access or heuristic prompts, rendering them ineffective in realworld black-box scenarios.In this work, we propose R 2 A, which aims to mislead black-box LLM routers to expensive models via adversarial suffix optimization.Specifically, R 2 A deploys a hybrid ensemble surrogate router to mimic the black-box router.A suffix optimization algorithm is further adapted for the ensemble-based surrogate.Extensive experiments on multiple open-source and commercial routing systems demonstrate that R 2 A significantly increases the routing rate to expensive models on queries of different distributions.Code and examples: https://github.com/ thcxiker/R2A-Attack.

User interactions with mobile applications (apps) are accompanied by continuous visual changes in the Graphical UI (GUI), guiding task completion and feedback. These changes help users complete intended tasks or assess the appropriateness of their actions, typically conveyed through visual cues such as appearance and color. While such visual changes are effective for sighted users, they are inaccessible to blind users, creating substantial barriers to GUI interaction. To address these challenges, we propose VisualDroid , a method based on a multi-modal large language model (LLM) for testing and classifying GUI visual changes using a tailored three-hop reasoning prompting framework. VisualDroid achieved an F1 score of 94.7% in 34 apps from 17 domains, surpassing all baseline methods. When evaluated on five open source apps from F-Droid, our method enabled developers to resolve three identified issues, with two still under review. In terms of efficiency and cost, our method indicates minimal resource consumption.

ABSTRACT Background In recent years, machine learning (ML) APIs have emerged as a valuable resource for addressing complex problems, such as image recognition. However, developers should use ML APIs carefully, as they have their own characteristics different from traditional ones: an ML API has its own training data set, a concrete target task, and its output is often the probability. As a result, developers may use an inappropriate API, and the program can still run without reporting errors, especially as there are many similar ML APIs provided by different platforms. Methods This paper proposes an approach called DAOR to help developers use ML APIs properly in their tasks. First, a comparative analysis of ML APIs is conducted, leveraging information from documentation and user reviews to identify comparable APIs. This involves extracting differences from the documentation, categorized into inputs, functions, and outputs, and summarizing key information from user reviews using GPT‐driven prompts. Finally, a visualization framework is designed to summarize and show the results. Evaluation and Results To evaluate the approach, a series of experiments is conducted based on the ML APIs from two famous platforms, Amazon Web Service AI and IBM Watson. The results show that useful information for distinguishing similar ML APIs can be gained, and it is helpful for developers to use the ML APIs correctly.

Logic synthesis is a crucial step in integrated circuit design, and area optimization is an indispensable part of this process. However, the area optimization problem for large-scale Fixed Polarity Reed-Muller (FPRM) circuits is an NP-hard problem. To address this problem, we divide Boolean circuits into small-scale circuits based on the idea of divide-and-conquer using the proposed grouping decomposition mechanism. Each small-scale Boolean circuit is transformed into an FPRM circuit by a polarity transformation algorithm. To ensure the circuit’s functionality remains unaffected, we integrate FPRM circuits into an FPRM and Boolean (RM-TB) dual logic circuit based on the proposed gate-level integration. However, the area optimization problem of RM-TB dual logic circuits is a multi-task, high-dimensional, and multi-extremal combinatorial optimization problem. Therefore, we propose a Multipopulation Multitasking Optimization Algorithm (MMuOA) that integrates self-evolution with a multitasking equilibrium optimizer and cross-task evolution through knowledge sharing and transfer. This forms a dynamic optimization framework for simultaneously searching for the optimal polarity corresponding to the minimal area of RM-TB dual logic circuits. Moreover, we propose an Area Optimization Approach (AOA) for an RM-TB dual logic circuit with the minimum area using the MMuOA. Experimental results based on the Microelectronics Center of North Carolina (MCNC) Benchmark test circuits demonstrate the effectiveness and superiority of the AOA compared to the state-of-the-art area optimization approach.

Graphical User Interfaces (GUIs) serve as the primary medium for user interaction with mobile applications (apps). Within these GUIs, editable text views, buttons, and other visual elements exhibit different states following user actions. However, developers often present these states only in various colors without providing textual hints for blind users. This results in significant difficulties for blind users to discern the transitions in component states, thereby hindering their ability to proceed with subsequent actions. Traditional rule-based methods and attribute settings often struggle to adapt to diverse component styles and fail to address the component state changes influenced by context. Recently, pre-trained Large Language Models (LLMs) have demonstrated their generalization ability to various downstream tasks. In this work, we leverage LLMs and propose a tool called C omponent st a te s distinguishing GPT (CasGPT) to automatically distinguish component states in GUIs and provide corresponding textual hints, thereby aiding blind users in app usage. Our experiments demonstrate that CasGPT is a lightweight approach capable of accurately distinguishing component states (accuracy = 86.5%). The usefulness of our method is validated through a user study, where participants expressed positive attitudes toward it. Also, we compare and find that our method outperforms other open source LLMs and different versions of GPT.

Malware detection is a critical issue in software engineering as it directly threatens user information security. Existing approaches often focus on individual modality (either source code or binary code) for the detection, but it ignores to effectively exploit the complementary information between them. This limits the detection performance, especially in complex and evasive malware scenarios. In this paper, we take Android applications written in Java as objects, and provide a novel fine-grained multimodal fusion method with large pre-trained models to combine the features from source and binary codes for the malware detection. For the source code modality, we employ the graphical user interface (GUI) as a framework to segment the source code into snippets, and use a pre-trained programming language model to extract feature representations. For the binary code modality, we convert binary code into grayscale images and fine-tune a pre-trained vision model to extract features indirectly. We then implement cross-modal attention and devise a contrastive loss to align features across modalities, supplementing this with supervised classification loss to refine the multimodal fusion process specifically for malware detection. Our experiments, conducted using the Data-MD and Data-MC benchmarks, demonstrate that our approach achieves a precision of 0.977 and a recall of 0.984 in detecting malware. This underscores the advantages of using large pre-trained models for feature representation and the fusion of information across different modalities for effective malware detection.

Program comprehension is a fundamental activity in the field of software engineering. However, efficiently and accurately understanding source code poses significant chal-lenges, as source code with similar semantics can differ in syntax. Recent state-of-the-art research has demonstrated that combining deep learning techniques with structural information from source code, specifically AST-based static graphs, can en-hance the extraction of essential features from source programs. Control flow and data flow information in source programs can express richer semantics while existing studies often overlook their heterogeneous integration when constructing program static graphs. This oversight results in the loss of information about the type of static graph edges, potentially impeding program comprehension. In this paper, We model the source program by using a heterogeneous static graph and then use Relational Graph Con-volutional Network (R-GCN) for feature extraction. Specifically, we present an innovative method for constructing a program static graph, termed the Heterogeneous Abstract Code Graph (HACG), and then we employ R-GCN to generate representations based on HACG for code classification and code clone detection. We evaluate our method using two extensive source code datasets: CodeNet, introduced by IBM, and BigCloneBench. The experimental results demonstrate the superiority of our approach over existing methods, achieving a code classification accuracy of 97.38 % and an average F1-score of 98.34 % in code clone detection.

Logic synthesis is a crucial step in integrated circuit design, and power optimization is an indispensable part of this process. However, power optimization for large-scale Mixed Polarity Reed-Muller (MPRM) logic circuits is an NP-hard problem. In this article, we divide Boolean circuits into small-scale circuits based on the idea of divide and conquer using the proposed Dynamic Adaptive Grouping Strategy (DAGS) and the proposed circuit decomposition model (CDM). Each small-scale Boolean circuit is transformed into an MPRM logic circuit by a polarity transformation algorithm. Based on the gate-level integration, we integrate small-scale circuits into an MPRM and Boolean Dual Logic (RBDL) circuit. Furthermore, the power optimization problem of RBDL circuits is a multi-task, multi-extremal, high-dimensional combinatorial optimization problem, for which we propose an Adaptive Multi-task Intelligent Algorithm (AMIA), which includes global task optimization, population reproduction, valuable knowledge transfer (VKT), and local exploration to search for the lowest power for RBDL circuits. Moreover, based on the proposed Fast Power Decomposition Algorithm (FPDA), we proposed a Power Optimization Approach (POA) for an RBDL circuit with the lowest power using the AMIA. Experimental results based on Microelectronics Center of North Carolina (MCNC) Benchmark test circuits demonstrate the effectiveness and superiority of the POA compared to state-of-the-art POAes.

With the increasing interconnection of smart devices, users often desire to adopt the same app on quite different devices for identical tasks, such as watching the same movies on both their smartphones and TVs. However, the significant differences in screen size, aspect ratio, and interaction styles make it challenging to adapt Graphical User Interfaces (GUIs) across these devices. Although there are millions of apps available on Google Play, only a few thousand are designed to support smart TV displays. Existing techniques to map a mobile app GUI to a TV either adopt a responsive design, which struggles to bridge the substantial gap between phone and TV, or use mirror apps for improved video display, which requires hardware support and extra engineering efforts. Instead of developing another app for supporting TVs, we propose a semi-automated approach to generate corresponding adaptive TV GUIs, given the phone GUIs as the input. Based on our empirical study of GUI pairs for TVs and phones in existing apps, we synthesize a list of rules for grouping and classifying phone GUIs, converting them to TV GUIs, and generating dynamic TV layouts and source code for the TV display. Our tool is not only beneficial to developers but also to GUI designers, who can further customize the generated GUIs for their TV app development. An evaluation and user study demonstrate the accuracy of our generated GUIs and the usefulness of our tool.

Links between issues and pull requests (PRs) assist GitHub developers in tackling technical challenges, gaining development inspiration, and improving repository maintenance. In realistic repositories, these links are still insufficiently established. Aiming at this situation, existing works focus on issues and PRs themselves and employ text similarity with additional information like issue size to predict issue-PR links, yet their effectiveness is unsatisfactory. The limitation is that issues and PRs are not isolated on GitHub. Rather, they are related to multiple GitHub sources, including repositories and submitters, which, through their diverse relationships, can supply potential and crucial knowledge about technical domains, developmental insights, and cross-repository technical details. To this end, we proposeAutoIPLinker (AIPL), which introduces the heterogeneous graph to model multiple GitHub sources with their relationships. Further, it leverages the metapath-based technique to reveal and incorporate the potential information for a more comprehensive understanding of issues and PRs. Firstly, we identify 4 types of GitHub sources related to issues and PRs (repositories, users, issues, PRs) as well as their relationships, and model them into task-specific heterogeneous graphs. Next, we analyze information transmitted among issues or PRs to reveal which knowledge is crucial for them. Based on our analysis, we formulate a series of metapaths and employ the metapath-based technique to incorporate various information for learning the knowledge-aware embedding of issues and PRs. Finally, we can infer whether an issue and a PR can be linked based on their embedding. We evaluate the performance of AIPL on real-world data sets collected from GitHub. The results show that, compared to the baselines, AIPL can achieve average improvements of 15.94&, 15.19&, 20.52&, and 18.50& in terms of Accuracy, Precision, Recall, and F1-score.

UI animation is a widely adopted design element in the UI of Android apps. There are many animation APIs available for a variety of purposes, and developers can utilize them to realize the UI animations to avoid reinventing the wheel and thus improve the development efficiency. However, the number of animation APIs is as high as thousands and it is non-trivial for developers to systematically master their use. Facing such a problem, we construct a multi-modal real-time animation API recommendation model called U-A2A in this paper, which can provide the available animation API for developers of Android apps in real-time throughout the animation realization according to the multi-modal information, that is, the information of UI animation task and the animation API context of current program (i.e., the animation API sequence that has been used). The reason for considering the animation API context is that realizing a UI animation requires the use of multiple animation APIs and relevant animation APIs roughly follow a sequence. U-A2A consists of two important parts: feature extractor and predictor. The feature extractor, which is constructed based on 3D CNN and GRU, can gain the combined feature of UI animation task as well as animation API context. The predictor consists of a fully connected layer as well as a softmax layer, and it can predict and recommend the next available animation API according to the result from feature extractor. Furthermore, we use the development experience about animation APIs of existing app products as the basis to adjust the parameters of U-A2A, thereby completing the training work of recommendation model. The experimental result shows that when 1, 3, 5, and 10 animation APIs are considered, U-A2A can achieve 45.13%, 65.72%, 72.97% and 81.85% accuracy respectively, which is much higher than the baseline LUPE.

The Graphical User Interface (GUI) plays a critical role in the interaction between users and mobile applications (apps), aiming at facilitating the operation process. However, due to the variety of functions and non-standardized design, GUIs might have many accessibility issues, like the size of components being too small or their intervals being narrow. These issues would hinder the operation of low vision users, preventing them from obtaining information accurately and conveniently. Although several technologies and methods have been proposed to address these issues, they are typically confined to issue identification, leaving the resolution in the hands of developers. Moreover, it can be challenging to ensure that the color, size, and interval of the fixed GUIs are appropriately compared to the original ones. In this work, we propose a novel approach named AccessFixer (Accessibility IssuesFixing Method), which utilizes the Relational-Graph Convolutional Neural Network (R-GCN) to simultaneously fix three kinds of accessibility issues, including small sizes, narrow intervals, and low color contrast in GUIs. With AccessFixer, the fixed GUIs would have a consistent color palette, uniform intervals, and adequate size changes achieved through coordinated adjustments to the attributes of related components. Our experiments demonstrate the effectiveness and usefulness of AccessFixer in fixing GUI accessibility issues. After fixing 30 real-world apps, our approach solves an average of 81.2% of their accessibility issues. Compared with the baseline tool that can only fix size-related issues, AccessFixer not only fixes both the interval and color contrast of components, but also ensures that no new issues arise in the fixed results. Also, we apply AccessFixer to 10 open-source apps by submitting the fixed results with pull requests (PRs) on GitHub. The results demonstrate that developers approve of our submitted fixed GUIs, with 8 PRs being merged or under fixing. A user study examines that low vision users host a positive attitude toward the GUIs fixed by our method.

Mobile applications (apps) are integral to our daily lives, offering diverse services and functionalities. They enable sighted users to access information coherently in an extremely convenient manner. However, it remains unclear if visually impaired users, who rely solely on the screen readers (e.g., Talkback) to navigate and access app information, can do so in the correct and reasonable order. This may result in significant information bias and operational errors. Furthermore, in our preliminary exploration, we explained and clarified that the navigation sequence-related issues encountered by visually impaired users could be categorized into two types: unintuitive navigation sequence and unapparent focus switching. Considering these issues, in this work, we proposed a method named RGNF (Re-draw GUI Navigation Flow). It aimed to enhance the understandability and coherence of accessing the content of each component within the Graphical User Interface (GUI), together with assisting developers in creating well-designed GUI navigation flow (GNF). This method was inspired by the characteristics identified in our preliminary study, where visually impaired users expected navigation to be associated with close position and similar shape of GUI components that were read consecutively. Thus, our method relied on the principles derived from the Gestalt psychological model, aiming to group GUI components into different regions according to the laws of proximity and similarity, thereby redrawing the GNFs. To evaluate the effectiveness of our method, we calculated sequence similarity values before and after redrawing the GNF, and further employed the tools proposed by Alotaibi et al. to measure the reachability of GUI components. Our results demonstrated a substantial improvement in similarity (0.921) compared to the baseline (0.624), together with the reachability (90.31%) compared to the baseline GNF (74.35%). Furthermore, a qualitative user study revealed that our method had a positive effect on providing visually impaired users with an improved user experience.

The graphical user interface (GUI) is crucial for users to interact with mobile devices. However, accessibility issues in the GUI, such as undersized text and redundant information, lead to understanding and operating obstacles for billions of low vision users in our society. To alleviate this situation, academia and industry have proposed various accessibility-related methods. Still, their over-dependence on specific detection rules and their inability to automatically repair the GUI source code limit them in helping developers resolve these issues. In this paper, we propose a novel method, named AGAA, for capturing and repairing undersized text and redundant information issues in the GUI. The evaluation on 12 real-world apps and the user study on 36 low vision users demonstrate that AGAA is effective in resolving these issues and is useful in improving the mobile device experience for low vision users, respectively.

UI animations, such as card movement and menu slide in/out, provide appealing user experience and enhance the usability of mobile applications. In the process of UI animation implementation, it is difficult for developers to identify suitable APIs for the animation to be implemented from a large number of APIs. Fortunately, the huge app market contains millions of apps, and they can provide valuable data resources for solving this problem. By summarizing the API usage for the same or similar animations in apps, reusable knowledge can be mined for the API recommendation. In this paper, we propose a novel method Animation2API, which mines the knowledge about APIs from existing apps and recommends APIs for UI animations. Different from existing text-based API recommendation approaches, Animation2API takes the UI animation in GIF/video format as query input. Firstly, we construct a database containing mappings between UI animations and APIs by analyzing a broad set of apps. Then, we build a UI animation feature extractor, which can be used to gain temporal-spatial feature vectors of UI animations. By comparing the temporal-spatial feature vectors between UI animations, we identify animations that are similar to the query animation from the database. Finally, we summarize the APIs used for implementing these animations and recommend a list of APIs for developers. The empirical evaluation results show that our method can achieve 82.66%Success rateand outperform the baseline Guru by 230.77% and 184.95% in terms ofPrecisionandRecallwhen considering twenty APIs. In the user study, we take the scenarios of using web search and ChatGPT to implement animations as baselines, and the results show that participants can complete animations faster (14.54%) after using Animation2API. Furthermore, participants’ positive feedbacks on the questionnaire indicate the usefulness of Animation2API.

Abstract The rapid development of Apps not only brings huge economic benefit but also causes increasingly fierce competition. In such a situation, developers are required to develop and update innovative functions to attract and retain users. Afterwards, analysing the functions of similar products can help developers formulate a well‐designed plan at the beginning of development as well as make updated strategies during the version update process. However, although there have been some methods that can be applied to extract the features from App descriptions to achieve this purpose to some extent, the features they obtained do not cover the details of App functions. Therefore, to conduct an in‐depth research on App functions, a novel method is proposed to extract App features with detailed information and an approach to integrate the gained results for further helping developers obtain the valuable knowledge better is provided. Subsequently, a series of experiments is carried out to evaluate our method. The results reveal that the proposed method can mine the features with detailed information from descriptions and integrate them effectively and also can assist developers to compare with other competitors and develop a better competitive analysis scheme.

Filter pruning, a representative model compression technique , has been widely used to compress and accelerate sophisticated deep neural networks on resource-constrained platforms. Nevertheless, most studies focus on reducing the cost of model inference, whereas the heavy burden of the pruning optimization process is neglected. In this paper, we propose MaskACC, a mask-aware convolutional computation method, which accelerates the prevailing mask-based filter pruning process on modern CPU platforms. MaskACC dynamically reorganizes the tensors used in convolutions with the mask information to avoid unnecessary computations, thereby improving the computational efficiency of the pruning process. Evaluation with state-of-the-art neural network models on CPU cloud platforms demonstrates the effectiveness of our method, which achieves up to 1.61 × speedup under commonly-used pruning rates, compared to conventional computations.

Abstract With the rapid increase in the number of Apps, the requirement of users has also become extremely complex. Developers have to continuously acquire innovative requirements that provide the guideline for developing more competitive products. However, traditional methods to acquire requirements are not suitable for the App development due to the disadvantage that it cannot interact with users directly. Besides, some methods that use text and data analysis to acquire requirements automatically are hard to expand innovative products because they are often confined to the specific App or the same domain. Therefore, to attract more new users, developers try to find new portable inspiration from other domains for enriching the functions of the App. In this article, we propose a feature extraction method from the descriptions of Apps and use similarity matching to acquire cross‐domain requirements. Our experiments have verified that the Precision, the Recall, and the F‐measure are all as high as 80% of our feature extraction method. Besides, the requirements list we recommend also makes a good performance in terms of reusability with the average Reuse Rank of 59.33% and average Adjusted Functional Points of 7.49, the adaptability gets an average score of 3.3, and the average score of operability is 3.

To prevent JavaScript developers from reinventing wheels, npm ecosystem provides numerous third-party libraries for developers to realize relevant functionalities. Npm displays the tags provided by the creators for these packages to help developers find suitable ones. However, not all creators have the habit of tagging their packages, and thus npm cannot provide tag information of a lot of packages for developers to help them understand the package functionalities effectively. Considering that many tags are unrelated to the functionality of packages, we propose a method to find out the tags that are important to distinguish the functionality categories of packages and assign them to untagged packages for assisting developers in the process of retrieving the packages. Firstly, we analyze the attribute of existing tags in npm to establish category tags (functionality categories). Then, we further mine the readme of tagged packages to generate keywords for each category tag. Finally, our method identifies category tags for untagged packages by measuring the similarity between their readme and the keywords of category tags. The evaluation demonstrates that our approach has a good performance in assigning category tags to untagged packages.

Abstract With the rapid prevalence of smartphones and the dramatic proliferation of mobile applications, people tend to do everything at their fingertips, including some sensitive activities, such as bank transfers. This makes security become one important factor when recommending apps to users. However, most existing methods recommend apps only on the basis of the apps' functionalities. Even when some methods take security into account, they usually roughly group apps with functionalities and identify the products using extra permissions as risky, but this ignores a common phenomenon that these permissions may be used only to achieve the corresponding functionalities. In this paper, we propose an app recommendation method considering both functionalities and security. For functionalities, we summarized them from app descriptions and further evaluated their completion quality in different products by analyzing their related reviews. For security, we cluster apps with similar functionalities and quality and analyze the permissions of apps in a more comparable range. In this way, our method recommends apps with higher completion quality of functionalities and security degree to users according to their demands. We conducted experiments on apps collected from six categories of Google Play, and the results show that our method has a good recommendation effect.

Abstract Application programming interfaces (APIs) play an important role in the increasingly competitive mobile application development industry, as they can greatly improve the efficiency of app development. However, finding proper APIs is often time‐consuming for the gap between the knowledge of APIs and app features. To solve this problem, we give an approach to summarize the wisdom of developers contained in the products in app stores and establish the system of API knowledge indexed by app features for the API recommendation. First, we extract features from the app descriptions and define the feature framework. Second, we parse the APK files of apps to gain the methods in code and APIs called by them and further introduce such API knowledge into the feature framework by utilizing method names as bridges. Finally, according to features in developers' queries, we locate corresponding feature nodes in the API knowledge system and recommend related API knowledge to developers. We conduct experiments based on 38,952 apps from five categories on Google Play, and the experimental results show that our approach has a good recommendation effect for the queries on app features.

To support the iterative development process of Apps, the goal model is not only established to describe the requirements at the early stage but also used for identifying the updating strategy in every iteration. In this process, reviews from users provide valuable information for developers to analyse the model with users sentiments. In this study, the authors combine the goal model with reviews for supporting the evolution of Apps. First, the authors introduce the reviews into the goal model as a new factor by comparing keywords. Second, the users sentiments in reviews are mined, and two kinds of information are gained by analysing the model to help developers make decisions on which goals to be improved in next version: one kind of information is about users sentiments on the goals to evaluate whether users like them; another kind is the impact of updating one goal to others. To validate the proposed approach, they conducted experiments and a survey based on the Apps in Google Play. The results show that the proposed approach can establish relationships between goals and reviews reasonably and further provide useful information for optimising the evolution strategy of the App.

Abstract Goal model is an important model in requirements engineering, and it can describe features and their relationships for supporting the development of apps. Since an app evolves continually, the goal model also needs to be updated with new requirements to guide the whole process. As the feedback of users, reviews provide an abundant resource of user requirements for updating the goal model. In this paper, we propose an approach to help developers (a) analyze reviews to gain the information of user requirements by training a classifier and defining keyword‐based linguistic rules as well as grammar‐based rules and (b) update the goal model with the extracted information, including improving existing goals and extending the model with new goals. In addition, we design a framework to represent results so that they can be understood by developers easily. According to our experiments based on the data in Google Play, the F‐measure of classifier on reviews can reach 75.76%, and the average precision for extracting requirements‐related information from reviews is 84.04%, then we can map the information to goals with the F‐measure of 70.21%. Furthermore, the survey on 22 developers shows that the information provided by us is useful for updating the goal model.

Summary Compared with traditional software, the domain analysis of apps is conducted not only in the early stage of software development to gain knowledge of a particular domain but also runs throughout each iteration of apps to help developers understand evolution trends of the domain for maintaining their competitiveness. In this paper, we propose an approach to analyze app descriptions combined with reviews in App stores automatically and construct a feature‐based domain state model (FDSM) in the form of state machine to support the domain analysis of apps. In FDSM, the domain knowledge up to a certain moment together is defined as a state. Initial state summarizes the high‐level knowledge by gaining topics of app descriptions, whereas each transition is generated based on the information gained within one period of time and describes the change from the current state to the next one. Furthermore, user opinions in reviews are introduced into the model to quantify the value of information for helping developers get key domain knowledge efficiently. To validate the proposed approach, we conducted a series of experiments based on Google Play. The results show that FDSM can provide valuable information for supporting domain analysis, especially in the evolution process of apps.

Abstract Reviews in App stores are a massive and fast‐growing data resource for developers to understand user experiences and their needs. Studies show that users often express their sentiments on App features in reviews, and this information is important for the development and evolution of Apps. To help developers gain such information efficiently, this paper proposes a method using App descriptions, another typical data in App stores, to guide the analysis of reviews. Firstly, we extract App features from descriptions, then summarize them to gain topics of App features as high‐level information; the results are formalized as a topic‐based domain model (TBDM). Secondly, we train classifiers of reviews based on the model to establish the relationships between user sentiments and App features. Finally, a quantified method is given to analyze the model based on developer preferences for recommending and summarizing reviews. To evaluate our approach, experiments were conducted using the App descriptions and reviews collected from Google Play. The results indicate that the approach can classify reviews to their related App features effectively (average F measure is 86.13%), and provides useful information for overall analyzing App features in a domain and identifying (dis)advantages of an App.