Full length articleA personalized recommendation method under the cloud platform based on users’ long-term preferences and instant interests
Introduction
In the past decades, emerging technologies such as big data, cloud services, cloud manufacturing, cloud computing, and the Internet of Things have been rapidly developing and gradually integrating with the design services industry. The emerging technologies are fast becoming a key instrument in promoting more personalized design services [1]. Cloud services become a new service model. In the era of cloud computing and big data, cloud services have been developed [2]. Cloud services focus on user-centric, aiming at providing more personalized product design solutions for users [3]. A key aspect of the cloud platform that allows user-centric and high-quality services is proactively recommending personalized services to users based on an accurate acquisition of user preferences. Large amounts of online text data need to be mined from cloud platforms, which are updated in real-time [3]. By analyzing these data, user preferences for product attributes can be accurately obtained, thus facilitating the product development and service recommendations, which is of great importance in the industrial design field. Most of the existing studies on cloud platforms focus on task reorganization and resource matching [4], task preference and decision-making [5], architecture and management [6], transaction models and services [7]. In summary, there is a lack of front-end research to capture users’ interest preferences for product attributes, to enable personalized recommendations.
The cloud platform is able to deliver virtualized resources, services and manufacturing to users [8]. It aims at achieving a high level of sharing of resources in the cloud and providing users with quality and efficient services at all times throughout the design process. The recommendation system based on the cloud platform helps users find the best information for their needs in an overloaded search space, and it can predict whether a user will like a given item or not [9]. In order to achieve personalized cloud service recommendations, besides the accurate access to user preferences, high quality and efficient recommendation algorithms are required to enable proactive customization of services for users and prediction of services based on their preferences [10]. The traditional recommendation systems focus on analyzing static associations between users and products from their interactions with them, mainly represented by traditional collaborative filtering models [10]. In recent years, researchers started to focus on several sequence recommendation scenarios, while the early work is almost exclusively based on sequence pattern mining [11]. Deep learning for time series data has made huge breakthroughs in recent years, and is widely used in natural language processing (NLP), social media and recommendation systems [12]. The recommendation algorithm can enhance the user stickiness and improve service quality. Initially, different types of information service terminals analyze big data on user behavior using the algorithmic technology to profile users, and constantly recommend information services on the same topic, in different forms and categories to users according to their interest preferences. This has a guiding effect on the expansion of knowledge and upgrading of interests of users within a certain period of time, thus performing the accurate matching of information services and information users. However, the immature algorithm technology prevents users from developing new user preferences, and thus creates an “information cocoon” [13]. In information dissemination, audiences only pay attention to what they like and get caught in an “echo chamber” of similar information, which gradually turns into a cocoon-like silkworm referred to as the “information cocoon” [14]. Users’ interests change over time and are classified into long-term preferences and instant interests [15]. Long-term preferences [15], [16], [17] refer to users’ more stable preferences over a longer period of time, such as what they are accustomed to, what they are more familiar with, and what they are professionally skilled at. In contrast to long-term preferences, instant interests [15], [16], [17] refer to the content that users are interested in at the current moment, representing content that users are not currently specialized in or unfamiliar with but may develop in the long term in the future. While most of the recommendation algorithms singularly consider users’ long-term preferences, and fail to take into account the instant interest of users. Consequently, it is quite common for information to become “more personalized and narrowly focused”. There is increasing concern that some instant interest is negatively inhibiting the development of user interest. Therefore, it is necessary to recommend both long-term preferences and instant interest to the user, and it is a key question of how to make a reasonable distribution of long-term preferences and instant interest. This paper aims at providing a personalized recommendation method based on user preferences by analyzing the problems in traditional personalized recommendation methods, combined with the characteristics of the user data of the cloud platform.
This study addresses the following issues: (1) The insufficient mining of resources on the front-end of the cloud platform. (2) The relationship between long-term preferences and instant interests. (3) The insufficient personalization caused by the “information cocoon” of the recommendation algorithm. Therefore, a BiRDA (Residual bi-directional Recurrent Neural Network with Dual Attentive mechanism) Recommendation is developed. We consider the co-opetition relationship between long-term preferences and instant interests, model long-term preferences and instant interests separately by Residual bi-directional Recurrent Neural Network, and then, compound long-term preferences and instant interests by dual attention mechanism. The aim is to improve the resource utilization rate on the cloud platform and to accurately personalized recommendation services for users. The main contributions include:
- •
A BiRDA model is proposed. Residual connectivity is used in the bidirectional RNN encoder to improve the feature extraction rate and to accurately extract the long-term preferences and instant interests features of users. And through the dual attention mechanism, the text prediction is made more in line with the user's personalized needs;
- •
The proposed BiRDA model is designed and analyzed algorithmically;
- •
Detailed experiments were conducted to obtain user data from three cloud platforms to evaluate the proposed BiRDA. We compare the proposed BiRDA model with the five state-of-the-art models and six variants of BiRDA model for experiments. It is shown that the BiRDA model is effective in solving the personalized recommendation problem.
The remainder of this paper is organized as follows. Section 2 reviews the literature related to cloud service recommendation systems and the evolution of recommendation models. Section 3 introduces the concepts related to the co-opetition games, the Residual bi-directional Recurrent Neural Network (Res-biRNN)) and the Dual attention mechanisms. Section 4 proposes and discusses a BiRDA recommendation model based on Res-BiRNN and dual attention mechanism. Section 5 performs example validation and analysis to show the cases where the proposed approach outperforms other models. The paper is finally concluded and the implications and limitations of the improved approach are discussed in Section 6.
Section snippets
Cloud service recommendation system
The cloud design platform provides users with hardware and software resources related to collaborative product design, as well as human resources such as suitable designers and manufacturers [18]. The cloud design platform supports the complete life cycle of accelerated application development, deployment and delivery [19]. The collaborative product design process of the cloud platform includes acquiring user requirements, determining design goals, decomposing design tasks, matching designers,
Co-opetition game theory
Game theory is considered a classical mathematical model in the economic field. It generally refers to the process in which multiple rational players with the information they know, choose a strategy that maximizes their interests under the mutual constraints of each player. In recent years, it has been widely used in the computer and application field, while several research results have emerged [49].
The co-opetition theory was first proposed by Brandenburger in 1996 [49], and co-opetition
Modeling framework
In order to improve the quality of product design service recommendations for the dynamic needs of users under the cloud platform, a personalized recommendation method which aggregates long-term preferences and instant interests is proposed (BiRDA). The BiRDA model predicts what users may need next, based on their long-term preferences and instant interest behavior. The model framework is illustrated in Fig. 3. It consists of three main steps (i.e., extracting personalized user needs, modeling
Case study
Considering the highly condensed question title text and answer data, as well as the coverage and variability of the preference words, the user’s question and answer records are captured. Therefore, considering the UniorcngeCDS cloud platform (https://wenda.orangecds. com/), the XMAKE cloud platform (https://wenda.yungongchang.com/) and the Asksubarme cloud platform (https://ask.asketchup.com/) as examples, the answer records and question records of users are obtained. Stable long-term
Conclusion
With the rapid development of cloud platforms, a significant amount of online text data has been accumulated, providing opportunities to understand users’ behaviors and provide information about personalized service. To solve the problem of insufficient mining of front-end resources in the cloud platform and insufficient personalization caused by the “information cocoon” of the recommendation algorithm, we first analyze the relationship between long-term preferences and immediate interests, we
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work was supported by the Ministry of Education of Humanities and Social Science Project (Grant Number: 21YJCZH113) and the Natural Science Foundation of Hebei Province (Grant Number: G2021202008).
References (69)
- et al.
Customer requirement-driven design method and computer-aided design system for supporting service innovation conceptualization handling
Adv. Eng. Inf.
(2020) - et al.
Cloud-based design and manufacturing: a computer-aided design manufacturing and design innovation
Comput. Des.
(2015) - et al.
Extracting and summarizing affective features and responses from online product descriptions and reviews: a Kansei text mining approach
Eng. Appl. Artif. Intell.
(2018) - et al.
Neural attentive travel package recommendation via exploiting long-term and short-term behaviors
Knowledge-Based Syst.
(2021) - et al.
Top-N personalized recommendation with graph neural networks in MOOCs
Comput. Educ. Artif. Intell.
(2021) - et al.
Cloud services recommendation: reviewing the recent advances and suggesting the future research directions
J. Netw. Comput. Appl.
(2017) - et al.
Research on multi-objective decision-making under cloud platform based on quality function deployment and uncertain linguistic variables
Adv. Eng. Inf.
(2019) - et al.
An AI-based open recommender system for personalized labor market driven education
Adv. Eng. Inf.
(2022) - et al.
Probabilistic forecasting with temporal convolutional neural network
Neurocomputing
(2020) - et al.
DeepAR: probabilistic forecasting with autoregressive recurrent networks
Int. J. Forecast.
(2020)