TY - GEN
T1 - Automatic Query Quality Prediction Using Feed Forward Neural Network
AU - Swathi, B. P.
AU - Geetha, M.
AU - Suhas, M. V.
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - Text retrieval activity involves retrieving the source code which software developers conduct on a daily basis. Current methods for retrieving source code rely on regular expressions and assume that the developers searching the code base are well-versed in the source code. Software developers ought to possess freedom to formulate queries on the codebase using natural language sentences as opposed to a search being a keyword or pattern-based, which are difficult to remember. When a bug is raised, developers who are new to the source code base may issue an NLQ to the code base to obtain the code where they must provide a fix. Nevertheless, the performance of a text search is highly dependent on query quality, and it succeeds when the textual query is of good quality. Predicting query quality in advance on a source code retrieval system can alert developers when a query is unlikely to succeed, thereby saving them time and effort from going through a long list of results. In this paper, the quality of the query is predicted using a back-propagation approach to train a feed-forward neural network (multi-layer perceptron). The model is trained and evaluated on a dataset that is created from the source code documentation. Natural language processing pre-retrieval metrics are exploited to study the significance of training the model. The model proves to be a good predictor of query quality with 85.75% accuracy.
AB - Text retrieval activity involves retrieving the source code which software developers conduct on a daily basis. Current methods for retrieving source code rely on regular expressions and assume that the developers searching the code base are well-versed in the source code. Software developers ought to possess freedom to formulate queries on the codebase using natural language sentences as opposed to a search being a keyword or pattern-based, which are difficult to remember. When a bug is raised, developers who are new to the source code base may issue an NLQ to the code base to obtain the code where they must provide a fix. Nevertheless, the performance of a text search is highly dependent on query quality, and it succeeds when the textual query is of good quality. Predicting query quality in advance on a source code retrieval system can alert developers when a query is unlikely to succeed, thereby saving them time and effort from going through a long list of results. In this paper, the quality of the query is predicted using a back-propagation approach to train a feed-forward neural network (multi-layer perceptron). The model is trained and evaluated on a dataset that is created from the source code documentation. Natural language processing pre-retrieval metrics are exploited to study the significance of training the model. The model proves to be a good predictor of query quality with 85.75% accuracy.
UR - http://www.scopus.com/inward/record.url?scp=85174821729&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85174821729&partnerID=8YFLogxK
U2 - 10.1109/INDISCON58499.2023.10270300
DO - 10.1109/INDISCON58499.2023.10270300
M3 - Conference contribution
AN - SCOPUS:85174821729
T3 - 2023 IEEE 4th Annual Flagship India Council International Subsections Conference: Computational Intelligence and Learning Systems, INDISCON 2023
BT - 2023 IEEE 4th Annual Flagship India Council International Subsections Conference
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 4th IEEE Annual Flagship India Council International Subsections Conference, INDISCON 2023
Y2 - 5 August 2023 through 7 August 2023
ER -