Goal-Driven Context-Aware Next Service Recommendation for Mashup Composition Xihao Xie1 Jia Zhang1 Rahul Ramachandran2 Tsengdar J. Lee3 and

2025-05-06 0 0 762.33KB 16 页 10玖币

侵权投诉

Goal-Driven Context-Aware Next Service

Recommendation for Mashup Composition

Xihao Xie1, Jia Zhang1, Rahul Ramachandran2, Tsengdar J. Lee3, and

Seungwon Lee4

1Department of Computer Science, Southern Methodist University, USA

{xihaox, jiazhang}@smu.edu

2NASA/MSFC, USA

rahul.ramachandran@nasa.gov

3Science Mission Directorate, NASA Headquarters, USA

tsengdar.j.lee@nasa.gov

4NASA/JPL, USA

seungwon.lee@jpl.nasa.gov

Abstract. As service-oriented architecture becoming one of the most

prevalent techniques to rapidly deliver functionalities to customers, in-

creasingly more reusable software components have been published on-

line in forms of web services. To create a mashup, it gets not only time-

consuming but also error-prone for developers to ﬁnd suitable services

from such a sea of services. Service discovery and recommendation has

thus attracted signiﬁcant momentum in both academia and industry.

This paper proposes a novel incremental recommend-as-you-go approach

to recommending next potential service based on the context of a mashup

under construction, considering services that have been selected to the

current step as well as its mashup goal. The core technique is an al-

gorithm of learning the embedding of services, which learns their past

goal-driven context-aware decision making behaviors in addition to their

semantic descriptions and co-occurrence history. A goal exclusionary neg-

ative sampling mechanism tailored for mashup development is also de-

veloped to improve training performance. Extensive experiments on a

real-world dataset demonstrate the eﬀectiveness of our approach.

Keywords: Service recommendation ·Mashup creation ·Service em-

bedding

1 Introduction

In the last two decades, a huge number of software components have been pub-

lished onto the Internet in forms of software services (or so-called APIs or ser-

vices). Up to June 15th, 2022, ProgrammableWeb5, the largest online repository

of APIs, has accumulated 24,471 services. Such remotely accessible services en-

able software developers to compose existing services into mashups, easier and

faster than before without creating everything from scratch. However, such a

5https://www.programmableweb.com/

arXiv:2210.14127v1 [cs.SE] 25 Oct 2022

2 X. Xie et al.

sea of services makes it time-consuming and error-prone for mashup develop-

ers to select suitable service candidates. Thus, service recommendation-powered

mashup development has attracted signiﬁcant momentum in recent years.

Fig. 1 illustrates a possible development process for a mashup randomly

selected from ProgrammableWeb named Shared Count6. As shown in Fig. 1,

a mashup creation is an incremental process with multiple rounds of service

selection, e.g., Twitter is selected ﬁrst followed by Facebook. In each round, an

instant recommender suggests the “next” suitable services.

Fig. 1. Incremental process of mashup creation.

In our position, an eﬀective recommender shall guide a mashup development

process as a journey of exploration in a forest (i.e., an embedding space), led

by a light beam (i.e., the mashup goal), as illustrated in the right-hand side

dotted box of Fig. 1. At any point of decision making, e.g., the green dot, the

services that have already been selected (i.e., Twitter at step t1 and Facebook

at step t2) represent the position of the “traveler” at the time. The problem

thus becomes how to make the next step (i.e., select the next service) toward

the ultimate goal (i.e., the red star representing the embedding of the textual

mashup goal “shared count is a small utility that will fetch social media shares

for a url”). We coin a term context to represent a decision making point during a

mashup development process, which comprises selected services and the mashup

goal each being denoted as a contextual item throughout this paper.

To tackle this problem, this paper proposes a goal-driven, context-aware ma-

chine learning method capable of recommending “next” suitable services in each

step of the incremental process of mashup creation. In contrast to the state-

of-the-art NLP-based word embedding [3] and transaction-based shopping cart

embedding and recommendation [11], we favor both mashup context and goal

synergistically. Our approach mainly comprises two modules: an oﬄine module

of representation learning and an online module of next candidate service recom-

6https://www.programmableweb.com/mashup/shared-count

Goal-Driven Context-Aware Service Recommendation 3

mendation. In the oﬄine module, we learn embeddings of contextual items and a

global attention vector over a historical mashup repository. Note that each con-

textual service may make a diﬀerent contribution scale to recommending next

service, and we apply an attention mechanism to learn such scales. In the on-

line module, once the context of an ongoing mashup is embedded based on the

trained parameters, we conduct matrix calculation to rank the probabilities of

potential services in descending order. Top K services will be recommended for

the next round to help mashup developers speed up service selection.

To the best of our knowledge, we make the ﬁrst eﬀort to learning the em-

bedding of goal-driven mashup context. For each service, in addition to learning

its semantic descriptions and service co-occurrence, we learn its decision making

behaviors from past mashup development provenance. Furthermore, our pro-

posed learning framework is extensible for considering other dimensions of data

such as user proﬁle. The main contributions of our work are three-fold. First,

we propose a novel machine learning algorithm that is capable of learning goal-

driven context embedding during mashup development. Second, we propose a

goal-exclusionary negative sampling strategy tailored for a rapid training pro-

cess for mashup development. Third, we demonstrate the eﬀectiveness of our

approach based on extensive experiments on a real-world dataset.

The remainder of this paper is organized as follows. Section 2 summaries the

related work. Section 3 deﬁnes the problem and presents our approach in detail.

Section 4 explains how to train and learn relevant parameters. Section 5 presents

and analyzes experimental results. Finally, Section 6 closes with conclusions.

2 Related Work

In this section, we will discuss closely related work from two aspects: service

representation learning and next service recommendation.

2.1 Service Representation Learning

In recent years, researchers have explored many ways to represent services in a

vector space to support downstream service recommendation applications. Here

are some examples. Gu et al. [2] propose an approach to recommending a bundle

of services by leveraging latent Dirichlet allocation (LDA) to embed services and

mashups in a topic space. Wang et al. [4] propose Service2vec to learn service

representations based on a constructed service network. Zhang et al. [5] incorpo-

rate service users’ perceptions into service proﬁles to form more comprehensive

service representations. Menzi et al. [6] embed services into a low-dimensional

vector space based on a constructed context-aware service knowledge graph to

support service recommendation.

In contrast to their work, our approach learns service embeddings from past

service selection decision making behaviors, in addition to service and mashup

descriptions and correlations.

2.2 Next Service Recommendation

Next service recommendation aims to recommend candidate services for devel-

opers in each step of a multi-round process of mashup creation. Cao et al. [1]

4 X. Xie et al.

design a collaborative-ﬁltering (CF)-based algorithm based on a two-level topic

model to recommend services. Zhang et al. [7] propose an approach to extracting

relationships of people, services and workﬂows from historical usage data into

a social network to proactively recommend services in a workﬂow composition

process. Liu et al. [8] develop a generalized sequential pattern algorithm to mine

frequent composition patterns of mashups, and design an interactive recommen-

dation algorithm to assist mashup creation. By modeling the relations between

services and service-based systems (SBSs) into a heterogeneous information net-

work, Xie et al. [9] measure semantic similarities between SBSs and use content

ﬁltering technique to recommend next service. Kirubananthan et al. [16] propose

a method to support long-term service composition recommendation according

to user ratings.

Our work diﬀers from the aforementioned studies in two aspects. First, we

introduce a model to embed the context including not only existing services but

also mashup goals. In this way, our model can learn goal-driven next service

prediction behaviors from past experiences. Second, we model diﬀerent contri-

butions of diﬀerent contextual items with an attention mechanism.

3 Context-aware Next Service Recommendation

Fig. 2. Blueprint of proposed approach.

Fig. 2 depicts an overview of our goal-driven context-aware service recom-

mendation framework. Given a partial mashup under construction, all services

already selected and the mashup goal form the context of the recommendation

problem. As shown in Fig. 2, for each contextual service item, its semantic em-

bedding and auxiliary embedding (to be learned) are fused into an intermediate

vector. Based on the services selected, an attention function (to be learned) will

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

Goal-DrivenContext-AwareNextServiceRecommendationforMashupCompositionXihaoXie1,JiaZhang1,RahulRamachandran2,TsengdarJ.Lee3,andSeungwonLee41DepartmentofComputerScience,SouthernMethodistUniversity,USAfxihaox,jiazhangg@smu.edu2NASA/MSFC,USArahul.ramachandran@nasa.gov3ScienceMissionDirectorate,NASAHeadq...

展开>> 收起<<

Goal-Driven Context-Aware Next Service Recommendation for Mashup Composition Xihao Xie1 Jia Zhang1 Rahul Ramachandran2 Tsengdar J. Lee3 and.pdf

共16页,预览4页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Goal-Driven Context-Aware Next Service Recommendation for Mashup Composition Xihao Xie1 Jia Zhang1 Rahul Ramachandran2 Tsengdar J. Lee3 and

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: