東華大學圖書館 |

A Machine Learning Approach to Solve the Complete Cold Start Problem in Recommendation Systems: Building the P2P Energy Trading Recommendation System with Theory of Consumption Values.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	A Machine Learning Approach to Solve the Complete Cold Start Problem in Recommendation Systems: Building the P2P Energy Trading Recommendation System with Theory of Consumption Values./
作者:	Shan, Shan.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2022,
面頁冊數:	250 p.
附註:	Source: Dissertations Abstracts International, Volume: 84-05, Section: A.
Contained By:	Dissertations Abstracts International84-05A.
標題:	Recommender systems. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=29730484
ISBN:	9798352978078

A Machine Learning Approach to Solve the Complete Cold Start Problem in Recommendation Systems: Building the P2P Energy Trading Recommendation System with Theory of Consumption Values.
Shan, Shan.

A Machine Learning Approach to Solve the Complete Cold Start Problem in Recommendation Systems: Building the P2P Energy Trading Recommendation System with Theory of Consumption Values. - Ann Arbor : ProQuest Dissertations & Theses, 2022 - 250 p.

Source: Dissertations Abstracts International, Volume: 84-05, Section: A.

Thesis (Ph.D.)--University of Northumbria at Newcastle (United Kingdom), 2022.

This item must not be sold to any third party vendors.

Recommendation Systems (RSs) belong to the subclass of information filtering systems, which is to determine any given user's "rating" of particular items. However, Cold Start Recommendations (CSR) pose significant challenges due to insufficient historical ratings of information. CSR is divided into Incomplete Cold Start Recommendation (ICSR) and Complete Cold Start Recommendation (CCSR) in historical ratings or user information; the former formulates recommendations to new users and new items, while the latter contributes to new systems. Various approaches deal with cold start challenges, including collaborative filtering, content-based fliting, the hybrid approach, increasing the diversity of data sources, context-awareness, and user profile construction. These approaches are commonly applied in the ICSR, though they are not as evident in the CCSR. Few previous studies have focused on solutions to CCSR; solving CCSR is significantly important due to the popularity of the recommendation system, the increasing new system, the improved recommendation performance, and the capability of reducing user input. The Peer to Peer (P2P) energy trading recommendation system (ETRS) is an example of a new system facing the complete cold start challenge. This research compares the advantages and drawbacks of different strategies to solve CSR and selects user profile construction to solve CCSR in P2P ETRS. First, a mixed-method data collection process was used, including semi-structured expert interviews, user-generated content analysis, and a survey to generate insights contributing to an efficient P2P energy trading user profile dataset. Second, a P2P ETRS is proposed by comparing the Decision Tree classifier and Ensemble classifier under different users' cultural backgrounds. Different cultural backgrounds may have different viewpoints on P2P energy trading. Thus, the system is evaluated with a large body of real-world data acquired from two case-study regions, one is the western dataset, and one is the Chinese dataset. The proposed recommendation system achieves high performance by delivering personalised recommendations. In addition, previous research applied to constructing the user profile dataset, demographic information, and user personalities, neglecting the motivations behind the product selections under different cultural backgrounds. Thus, this research analysed the driving motivations of P2P energy trading with the theory of consumption values under different cultural backgrounds. This research has two significant contributions. First, this study identified the user profile construction strategy with a machine learning approach to solve the CCSR problem in the P2P ETRS. This strategy can effectively build the original user profile dataset, reduce the user input, and evaluate the recommended model. In addition, the research adopted ensemble learning algorithms to improve the performance of the recommended model. In terms of system performance, the accuracy of the Decision Tree (DT) model achieves an 86.1% for the Chinese dataset and a 68.5% for the western data set. In addition, ensemble learning algorithms can improve the classification of both datasets, as CatBoost performed best in the ensemble model, with an accuracy rate of 94.3% on the Chinese dataset and 79.8% on the western dataset. Second, this research created a bridge between machine learning and the traditional social science theory. A framework for a P2P energy trading recommendation system within the theory of consumption values was proposed to understand the collaborative consumption behaviour. Previous studies of consumption value theory had applied to collaborative consumption, including Airbnb and Uber. Unlike the previous research, P2P energy trading paid more attention to the feeling of uncertainty on the emotional value rather than enjoyment; That is to say, consumers believe the uncertainty of the supply of using renewable energy is more important than the consideration of enjoyment of the experience. Most notably, the essential driving values in the two datasets are different, although income can impact P2P product selection in both datasets. The feeling of uncertainty played the most crucial role in the Chinese dataset, while the western dataset focused on sustainability.

ISBN: 9798352978078Subjects--Topical Terms:

3562220
Recommender systems.

A Machine Learning Approach to Solve the Complete Cold Start Problem in Recommendation Systems: Building the P2P Energy Trading Recommendation System with Theory of Consumption Values.
LDR:05605nmm a2200361 4500 001 2394417
005 20240422070851.5
006 m o d
007 cr#unu||||||||
008 251215s2022 ||||||||||||||||| ||eng d
020 $a 9798352978078
035 $a (MiAaPQ)AAI29730484
035 $a (MiAaPQ)Northumbria_49920
035 $a AAI29730484
040 $a MiAaPQ $c MiAaPQ
100 1 $a Shan, Shan. $3 3763891
245 1 2 $a A Machine Learning Approach to Solve the Complete Cold Start Problem in Recommendation Systems: Building the P2P Energy Trading Recommendation System with Theory of Consumption Values.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2022
300 $a 250 p.
500 $a Source: Dissertations Abstracts International, Volume: 84-05, Section: A.
500 $a Advisor: Li, Honglei;Yang, Longzhi;Lin, Zhibin.
502 $a Thesis (Ph.D.)--University of Northumbria at Newcastle (United Kingdom), 2022.
506 $a This item must not be sold to any third party vendors.
520 $a Recommendation Systems (RSs) belong to the subclass of information filtering systems, which is to determine any given user's "rating" of particular items. However, Cold Start Recommendations (CSR) pose significant challenges due to insufficient historical ratings of information. CSR is divided into Incomplete Cold Start Recommendation (ICSR) and Complete Cold Start Recommendation (CCSR) in historical ratings or user information; the former formulates recommendations to new users and new items, while the latter contributes to new systems. Various approaches deal with cold start challenges, including collaborative filtering, content-based fliting, the hybrid approach, increasing the diversity of data sources, context-awareness, and user profile construction. These approaches are commonly applied in the ICSR, though they are not as evident in the CCSR. Few previous studies have focused on solutions to CCSR; solving CCSR is significantly important due to the popularity of the recommendation system, the increasing new system, the improved recommendation performance, and the capability of reducing user input. The Peer to Peer (P2P) energy trading recommendation system (ETRS) is an example of a new system facing the complete cold start challenge. This research compares the advantages and drawbacks of different strategies to solve CSR and selects user profile construction to solve CCSR in P2P ETRS. First, a mixed-method data collection process was used, including semi-structured expert interviews, user-generated content analysis, and a survey to generate insights contributing to an efficient P2P energy trading user profile dataset. Second, a P2P ETRS is proposed by comparing the Decision Tree classifier and Ensemble classifier under different users' cultural backgrounds. Different cultural backgrounds may have different viewpoints on P2P energy trading. Thus, the system is evaluated with a large body of real-world data acquired from two case-study regions, one is the western dataset, and one is the Chinese dataset. The proposed recommendation system achieves high performance by delivering personalised recommendations. In addition, previous research applied to constructing the user profile dataset, demographic information, and user personalities, neglecting the motivations behind the product selections under different cultural backgrounds. Thus, this research analysed the driving motivations of P2P energy trading with the theory of consumption values under different cultural backgrounds. This research has two significant contributions. First, this study identified the user profile construction strategy with a machine learning approach to solve the CCSR problem in the P2P ETRS. This strategy can effectively build the original user profile dataset, reduce the user input, and evaluate the recommended model. In addition, the research adopted ensemble learning algorithms to improve the performance of the recommended model. In terms of system performance, the accuracy of the Decision Tree (DT) model achieves an 86.1% for the Chinese dataset and a 68.5% for the western data set. In addition, ensemble learning algorithms can improve the classification of both datasets, as CatBoost performed best in the ensemble model, with an accuracy rate of 94.3% on the Chinese dataset and 79.8% on the western dataset. Second, this research created a bridge between machine learning and the traditional social science theory. A framework for a P2P energy trading recommendation system within the theory of consumption values was proposed to understand the collaborative consumption behaviour. Previous studies of consumption value theory had applied to collaborative consumption, including Airbnb and Uber. Unlike the previous research, P2P energy trading paid more attention to the feeling of uncertainty on the emotional value rather than enjoyment; That is to say, consumers believe the uncertainty of the supply of using renewable energy is more important than the consideration of enjoyment of the experience. Most notably, the essential driving values in the two datasets are different, although income can impact P2P product selection in both datasets. The feeling of uncertainty played the most crucial role in the Chinese dataset, while the western dataset focused on sustainability.
590 $a School code: 5060.
650 4 $a Recommender systems. $3 3562220
650 4 $a Filtering systems. $3 3763892
650 4 $a Decision making. $3 517204
650 4 $a Computer science. $3 523869
650 4 $a Information science. $3 554358
650 4 $a Information technology. $3 532993
650 4 $a Web studies. $3 2122754
690 $a 0984
690 $a 0723
690 $a 0489
690 $a 0646
710 2 $a University of Northumbria at Newcastle (United Kingdom). $3 3545556
773 0 $t Dissertations Abstracts International $g 84-05A.
790 $a 5060
791 $a Ph.D.
792 $a 2022
793 $a English
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=29730484