A Service Recommendation Algorithm Based on Self-Attention Mechanism and DeepFM

A Service Recommendation Algorithm Based on Self-Attention Mechanism and DeepFM

Li Ping Deng, Bing Guo, Wen Zheng
Copyright: © 2023 |Pages: 18
DOI: 10.4018/IJWSR.331691
Article PDF Download
Open access articles are freely available for download

Abstract

This article proposes a recommendation model based on self-attention mechanism and DeepFM service, the model is SelfA-DeepFM. The method firstly constructs the service network with DTc-LDA model to mine the potential relationship between Mashup and API, which not only fully considers the text attributes but also combines the network structure information to effectively mitigate the sparsity of the service data. Secondly, service clustering to obtain numerical feature similarities. Finally, the self-attention mechanism is used to capture the different importance of feature interactions, and the DeepFM model is used to mine the complex interaction information between multidimensional features to predict and rank the quality score of API services to recommend suitable APIs. To verify the performance of the model, the authors use the real data crawled from the ProgrammableWeb platform to conduct multiple groups of experiments. The experimental results show that the model significantly improves the accuracy of service recommendation.
Article Preview
Top

Introduction

With the emergence of new software technologies such as cloud computing, mobile computing, and blockchain, APIs (Application Programming Interface) is playing an increasingly important role in software development (Liang et al.,2019). Driven by the API economy, many companies, such as Google, Amazon, and Microsoft, have released Web-based open APIs, often to allow third parties to access their critical resources in a programmable way, and as a result the number of open APIs on the Internet has grown rapidly (Lian et al.,2021). In fact, the number of API services is showing a trend of multiplication (Almarimi, N. et al.,2019), with as many as 20,373 API services in the Programmable-Web website as of November 2019 (Deng et al., 2021). In this context, it has become an important problem in the field of service computing to quickly and effectively find API services to meet the Mashup needs of developer users from such a large-scale collection of services, and it is attempted to be solved by service recommendation (Kang et al.,2021; Jiang et al.,2021).

In the field of service computing, there have been many studies on recommending suitable APIs for Mashup creation, where collaborative filtering (CF) algorithms have played an important role (Cao et al.,2020). However, existing CF-based recommendation methods suffer from data sparsity and cold start, which lead to poor recommendation results. To alleviate the data sparsity problem, traditional recommendation methods often convert user attributes and item attributes into a generic feature vector and use supervised learning to predict recommendation scores, such as FM (Rendle et al.,2011), Wide& Deep (Cheng et al.,2016), DeepFM (Guo et al.,2017), xDeepFM (Lian et al.,2018), etc. Despite the superiority of deep learning-based FM, these approaches ignore the relationship between Web service structural properties (e.g., Mashup-API, Mashup-Mashup, API-API), which can affect the accuracy of recommendations (Cao&Liu et al.,2019). Deep learning-based recommendation methods are introduced into Web networks to solve the problem of data sparsity, and many studies have shown that the introduction of knowledge graphs in Web service recommendation can fully exploit the internal information of knowledge graphs to improve the recommendation effect (Jiang et al.,2021; Wang et al.,2017). However, these methods focus more on mining the structural attributes of the knowledge graph and ignore extracting the auxiliary information(e.g., tags) of the nodes, which affects the accuracy of recommendations (Cao et al.,2023). In response to the shortcomings of existing recommendation methods, this paper proposes a service recommendation SelfA-DeepFM model .

The main contributions of this paper are summarized as follows:

  • 1.

    A very effective general framework for service recommendation is proposed to connect API and Mashup related information to improve the performance of service recommendation. Among others, constructing service networks and service clustering models can be done with the DTc-LDA model mentioned in this framework, or other models can be used.

  • 2.

    The SelfA-DeepFM model is proposed. Using the Self-Attention Mechanism to capture the different importance of feature interactions. The DeepFM model is used to mine the complex interaction information between multi-dimensional features to predict and rank the quality score of API services to recommend suitable APIs, thus improving the recommendation effect.

  • 3.

    In order to effectively alleviate the sparsity of services, the DTc-LDA model is first used to construct a service network, which provides a comprehensive portrayal of the relationships between Web services and explores the potential relationships between Mashup and APIs, which not only fully considers the text properties, but also combines the network structure information.

  • 4.

    A series of experiments were conducted on a real Web dataset, and the experimental results show that the SelfA-DeepFM model effectively improves the effect of service recommendation .

Complete Article List

Search this Journal:
Reset
Volume 21: 1 Issue (2024): Forthcoming, Available for Pre-Order
Volume 20: 1 Issue (2023)
Volume 19: 4 Issues (2022): 1 Released, 3 Forthcoming
Volume 18: 4 Issues (2021)
Volume 17: 4 Issues (2020)
Volume 16: 4 Issues (2019)
Volume 15: 4 Issues (2018)
Volume 14: 4 Issues (2017)
Volume 13: 4 Issues (2016)
Volume 12: 4 Issues (2015)
Volume 11: 4 Issues (2014)
Volume 10: 4 Issues (2013)
Volume 9: 4 Issues (2012)
Volume 8: 4 Issues (2011)
Volume 7: 4 Issues (2010)
Volume 6: 4 Issues (2009)
Volume 5: 4 Issues (2008)
Volume 4: 4 Issues (2007)
Volume 3: 4 Issues (2006)
Volume 2: 4 Issues (2005)
Volume 1: 4 Issues (2004)
View Complete Journal Contents Listing