TY - JOUR
T1 - Location-based deep factorization machine model for service recommendation
AU - Wang, Qingren
AU - Zhang, Min
AU - Zhang, Yiwen
AU - Zhong, Jinqin
AU - Sheng, Victor S.
N1 - Funding Information:
This work is partially supported by the National Natural Science Foundation of China [No.62006003, U1936220], the National Key R&D Program of China under Grant [No.2019YFB1704101], the Natural Science Foundation of Anhui Province of China [No.2008085QF307], and the Anhui Foundation for Science and Technology Major Project under Grant [No.18030901034]. Yiwen Zhang is the corresponding author of this paper.
Publisher Copyright:
© 2021, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.
PY - 2022
Y1 - 2022
N2 - The era of everythingasaservice led to an explosion of services with similar functionalities on the internet. Quickly obtaining a high-quality service has become a research focus in the field of service recommendation. Studies show that quality of service (QoS) predictions are an effective way to discover services with high quality. However, sparse data and performance fluctuation challenge the accuracy and robustness of QoS prediction. To solve these two challenges, this paper proposes a location-based deep factorization machine model, namely LDFM, by employing information entropy and location projection of users and services. Particularly, our LDFM can be decomposed into three phases: i) extending a raw QoS dataset without introducing additional information, where LDFM projects the existing users (services) in the direction of their position vectors to increase the number of users (services) as well as the number of records that users invoke services; ii) mining a sufficient number of potential features behind the behaviors of users who invoke services, where LDFM employs a factorization machine to extract potential features of breadth with low dimensions (i.e., one and two dimensions) and utilizes deep learning to seek potential depth features with high dimensions; and iii) weighting extracted features within various dimensions, where LDFM employs information entropy to strengthen the positive effects of valid features while reducing the negative impacts generated by biased features. Our experimental results (including t-test analyses) show that our proposed LDFM always performs well under different user-service matrix densities and performs better than existing start-of-the-art methods in terms of the accuracy and robustness of QoS predictions.
AB - The era of everythingasaservice led to an explosion of services with similar functionalities on the internet. Quickly obtaining a high-quality service has become a research focus in the field of service recommendation. Studies show that quality of service (QoS) predictions are an effective way to discover services with high quality. However, sparse data and performance fluctuation challenge the accuracy and robustness of QoS prediction. To solve these two challenges, this paper proposes a location-based deep factorization machine model, namely LDFM, by employing information entropy and location projection of users and services. Particularly, our LDFM can be decomposed into three phases: i) extending a raw QoS dataset without introducing additional information, where LDFM projects the existing users (services) in the direction of their position vectors to increase the number of users (services) as well as the number of records that users invoke services; ii) mining a sufficient number of potential features behind the behaviors of users who invoke services, where LDFM employs a factorization machine to extract potential features of breadth with low dimensions (i.e., one and two dimensions) and utilizes deep learning to seek potential depth features with high dimensions; and iii) weighting extracted features within various dimensions, where LDFM employs information entropy to strengthen the positive effects of valid features while reducing the negative impacts generated by biased features. Our experimental results (including t-test analyses) show that our proposed LDFM always performs well under different user-service matrix densities and performs better than existing start-of-the-art methods in terms of the accuracy and robustness of QoS predictions.
KW - Deep learning
KW - Factorization machine
KW - Information entropy
KW - Quality of service
KW - Service recommendation
UR - http://www.scopus.com/inward/record.url?scp=85122661800&partnerID=8YFLogxK
U2 - 10.1007/s10489-021-02998-9
DO - 10.1007/s10489-021-02998-9
M3 - Article
AN - SCOPUS:85122661800
JO - Applied Intelligence
JF - Applied Intelligence
SN - 0924-669X
ER -