In the paper, a novel technique QUICKAR was proposed. It helps suggest semantically relevant queries for concept location by not only estimating word semantic similarity or relevance from the source code but also leveraging the crowdsourced knowledge from Stack Overflow. The paper describes in detail the query reformulation process of QUICKAR and the principle of building a word adjacency list database, and through the 510 query request experiment on two systems ecf, eclipse.pde.ui, it is found that compared with a baseline technology Rocchio's expansion, QUICKAR can improve 66% of the initial query quality on average while preserving the quality of 10%, which is highly promising.
Contributions of The Paper
Mining 500K questions from Stack Overflow using Stack Exchange Data Explorer and building a word adjacency list database for query reformulation by performing standard natural language preprocessing and capturing co-occurring words.
Introduce and evaluate a new technique QUICKAR, which takes the idea of applying contextual words from Stack Overflow questions and answers for identifying semantically similar software-specific words for the query reformulation in software maintenance tasks.
Comments
The paper clearly describes the algorithms and evaluation results in query reformulation using pseudo-code, workflow diagrams, and data tables. And it is a good analytical method to analyze the evaluation results by answering three presupposed questions.
In the algorithm, we use the cosine similarity measure for the contextual similarity between words for candidate terms from the project and check the co-occurrence frequency between words for candidate terms from SO to calculate the R score. The paper did not mention the specific reason why we use these two methods in this way, can we switch these two methods for these two candidate terms sets? (Maybe I miss reading this point in the paper)
Publisher
Shiwen(Lareina) Yang
Link to The Paper
https://web.cs.dal.ca/~masud/papers/masud-ASE2016b-pp.pdf
Name of The Authors
Mohammad Masudur Rahman, Chanchal K. Roy
Year of Publication
2016
Summary
In the paper, a novel technique QUICKAR was proposed. It helps suggest semantically relevant queries for concept location by not only estimating word semantic similarity or relevance from the source code but also leveraging the crowdsourced knowledge from Stack Overflow. The paper describes in detail the query reformulation process of QUICKAR and the principle of building a word adjacency list database, and through the 510 query request experiment on two systems ecf, eclipse.pde.ui, it is found that compared with a baseline technology Rocchio's expansion, QUICKAR can improve 66% of the initial query quality on average while preserving the quality of 10%, which is highly promising.
Contributions of The Paper
Comments
The paper clearly describes the algorithms and evaluation results in query reformulation using pseudo-code, workflow diagrams, and data tables. And it is a good analytical method to analyze the evaluation results by answering three presupposed questions.
In the algorithm, we use the cosine similarity measure for the contextual similarity between words for candidate terms from the project and check the co-occurrence frequency between words for candidate terms from SO to calculate the R score. The paper did not mention the specific reason why we use these two methods in this way, can we switch these two methods for these two candidate terms sets? (Maybe I miss reading this point in the paper)