Supporting Software Developers with a Holistic Recommender System

“Code search engines and recommendation systems are available for retrieving API and code usage examples [30]–[33], code fragments implementing specific features [34]–[37] and crowd knowledge [38], [39].”

“Equally important are recommenders that support developers in finding APIs and code usage examples [21, 22, 31, 45, 64], code fragments implementing specific features [41, 51, 53, 66] or generally useful crowdsourced knowledge for a given implementation task [50].”

“, which systematically mines community based resources such as forums or Stack Overflow to propose related changes [46].”

“flow discussions to suggest context-specific code snippets and documentation [5], [25], [26], [27], [29], [31], [38], [41].”

“One advantage of recruiting locally is that researchers can be fairly certain that most of their sample actually has a computer science/programming background (e.g., [1], [2], [3], [4], [5], [6]).”

“For example, tools like Mica [64], Assieme [33], and Libra [56] improve existing general-purpose search engines by concentrating programming-specific information to help developers locate the most relevant API choices or other resources; BluePrint [15] and Seahawk [55] mined software repositories and online Q&A forums…”

“Among the various resources available on the Web, Q&A Websites — and in particular Stack Overflow— have been the basis of many recommender systems [8], [29], [30], [39].”

“Following the previous guidelines [70], [90] on interpreting the effect size (trivial: | d | ≤ 0 .”

“Other tools exploit StackOverflow discussions to suggest code snippets and documentation [9], [31], [32], [34], [36], [47], [50].”

“propose a holistic recommendation system Libra, which integrates the IDE and the web browser [47].”

“On the other hand, the main challenge of external [17].”

“Libra [32] augments this approach by including relevant terms from the user’s recent browsing history.”

“Recent work proposed approaches to help developers browse heterogeneous data during maintenance tasks [17], [20], [21].”

“Such approaches [61], [62], [63], [64], [65], [66], [67], [68], [69], [70], [71] mine software repositories to find representative API usage examples for assisting developers when trying to use a class or method, or when finding code examples showing how to implement a given task/feature.”

“, [21]-[23]), the automatic generation of code for different purposes (e.”

“2016), programming-related information (Ponzanelli et al. 2017), APIs (Yuan et al.”

“For example, recommendation of developments artifacts [42, 43, 44, 45], contextual information [46], libraries [47], meth-ods [48], commands [49] and even pull-requests commenter [50].”

“Indeed, it has been found that, on one hand, human software developers spent most of their time in activities such as information seeking (Ponzanelli et al., 2017) or interacting (Ciccozzi et al., 2017).”

“Libra [21] is a holistic recommender system which can support software developers by reorganizing browser search results.”

“A wide range of studies focus on the conventions of developers when writing Java programs and their potential applications.”

“The previous version of M AESTRO [17], [57] recommends relevant SO posts for manual fixing of REs, while Prompter [58], [59] and Li-bra [60] suggest posts to assist during implementation.”

“However, some summarization approaches leverage new techniques, for instance HoliRank [28], [29], an extension of PageRank [30] devised to analyze data in a holistic fashion.”

“Researchers might use eye-trackers not only for data collection, but also for the design of online approaches that leverage a developer’s previously viewed information to continuously assist in the identification of task-relevant text [30].”

“In the same domain, Libra [30] integrates recommendations but it also considers, in addition to context extracted from the integrated development environment, resources opened by the user such as URLs or documents to better understand his/her context.”

“This concept was later expanded in Libra [18], which provides comprehensive web resource recommendations directly within the IDE. CrossRec [16] recommends third-party libraries by analyzing attached code.”

“…Almost all API recommendation tasks rely on two sources of information, natural language queries that developers use to search [4], [5], [18], [19], [20], [21], [22], [23], [24], [25], and contextual information about existing code on the client side [26], [27], [28], [29], [30], [31], [32], [33].”

“[1], this information is often highly heterogeneous, making it difficult for developers to access and integrate all the information they need to complete their tasks effectively.”

“Tools like Libra [53] aim at refining the web search results complemented by the code context of the software developer in their IDEs.”

“1) Datasets: For this investigation, we chose artifact types that relate to program comprehension tasks [28], [29].”

“…developer, we will measure the completion percentage as the number of sub-tasks correctly completed divided by the total number of sub-tasks [26], [27], and completion time , measured in seconds passed from the beginning of the task to the moment in which the participants says they have completed…”

“Libra adds a plot to Google’s search result page visualizing results according to the number and diversity of overlapping code terms, helping guide developers toward pages with more details or alternative solutions [34].”

“In the early years of API recommendation, researchers approached the API recommendation problem using statistical approaches such as Indexing [23, 54, 55, 68, 69, 80, 89], Tf-idf [18, 42, 97, 98], bag-of-words (BOW) [78], PageRank [64] and Learn-to-rank [102].”

“The use of data mining algorithms for recommending source code changes has been steadily increasing in a variety of fields, including industrial, scientific, and commercial ones (Borg et al., 2017; Robillard et al., 2010; Ponzanelli et al., 2017).”

“Newcomers possess limited knowledge and experience [9, 10].”

“As a first step to help locate relevant information to a task, many systems assist in locating pertinent artifacts (e.g. [4, 18]).”

“Ponzanelli et al. presented a holistic recommender system called Libra to help analyze developers' semantic relationships by constructing a holistic meta-information model of developers [28].”