Leveraging crowd knowledge to curate documentation for agile software industry using deep learning and expert ranking
Agility gives the ability to be highly responsive to changes and make improvements in the way the agile project is documented. The just-in-time and just barely good enough documentation may miss out on important executable specifications. Interestingly, the frequently asked ‘how-to-do’ questions on...
Uloženo v:
| Vydáno v: | Multimedia systems Ročník 29; číslo 3; s. 1799 - 1813 |
|---|---|
| Hlavní autor: | |
| Médium: | Journal Article |
| Jazyk: | angličtina |
| Vydáno: |
Berlin/Heidelberg
Springer Berlin Heidelberg
01.06.2023
Springer Nature B.V |
| Témata: | |
| ISSN: | 0942-4962, 1432-1882 |
| On-line přístup: | Získat plný text |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
| Shrnutí: | Agility gives the ability to be highly responsive to changes and make improvements in the way the agile project is documented. The just-in-time and just barely good enough documentation may miss out on important executable specifications. Interestingly, the frequently asked ‘how-to-do’ questions on the popular question answering (Q&A) websites like stack overflow are strong indicators of gaps in documentation and respondent answers can complement conventional software documentation practices. Social interaction within these QA websites generates partially structured content commonly referred to as crowd knowledge that can offer a peer-reviewed re-documentation by integrating the answers of these ‘how-to-do’ concerns. But finding the best, value-added answer to the question which can contribute toward an enriched and curated documentation is computationally difficult. Moreover, query duplicates can cause seekers to spend more time finding these best answers. As a solution, the research proffers a novel question-answering crowd documentation model (QACDoc) which is based on a socially mediated documentation mechanics involving the dynamics of community-based web. Firstly, duplicate questions are detected using Siamese neural architecture where two identical hierarchical attention networks are used to generate vectors for similarity matching. Semantic matching is done using Manhattan distance function, and a multi-layer perceptron is trained to output the predictions. Next, all respondents of semantically matched questions are grouped to form an intent-based crowd and lastly, top
k
-experts are identified using representative social presence features for expert ranking. The crowd documentation is then filtered to only include answers of these identified experts. |
|---|---|
| Bibliografie: | ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 14 |
| ISSN: | 0942-4962 1432-1882 |
| DOI: | 10.1007/s00530-020-00741-x |