以Citespace分析跨域推薦領域之研究現狀與趨勢

Abstract

傳統個性化推薦存在數據稀疏性以及冷啟動問題，跨域推薦可以整合來自多個領域的數據，為解決這兩個問題提供新的思路與方法，從而提高推薦的準確性與多樣性。目前利用視覺化作為輔助技術的跨域推薦綜合研究甚少，因此本文以Citespace為視覺化輔助，以Web of Science資料庫2000~2020年的相關文獻作為研究對象，通過科學知識圖譜，從宏觀到微觀對文獻的代表國家、關鍵詞、共被引文獻等進行視覺化分析，討論跨域推薦領域常用技術的優劣勢與適用範圍，為當前面臨的挑戰與未來研究的方向提供建議： 1、跨域推薦自2001年誕生，2014年之前發展較為緩慢，因2014年遷移學習技術提出，迅速發展。因此，建議可以合理利用前人的研究結果。 2、成熟的合作網絡可以善用各機構優勢及專業，推進研究發展。因此建議不同機構可以以優勢主題形成合作網絡，例如北京科技大學、雪梨科技大學、浙江大學均擅長遷移學習技術，因此這些機構可以此作為主要研究方向，形成合作網絡。 3、對跨域推薦系統的評價與應用關鍵詞主要集中在冷啟動問題、個性化、社交媒體上。可以預見，個性化的用戶需求結合社交媒體資料的冷啟動問題解決，會在一定程度上推進跨域推薦系統在更多領域的應用與發展。 4、跨域推薦在提高推薦多樣性與系統穩健性的相關研究較少。此外，跨領域也給推薦也帶來了新的問題，例如不同域之間的資料異構性，以及在多個資料來源的情況下，如何平衡多數據源的比例，如何快速處理多個域轉移。 5、藉由傳感器可以將跨域推薦系統應用於其他更多生活場景，例如教育與健康醫療領域與食物、醫療、運動領域結合，提供客製化飲食搭配、疾病預處理、運動建議等與人們生活相關的服務。

Traditional personalized recommendation suffers from data sparsity and cold start problems. Cross-domain recommendation(CDR) can integrate data from multiple domains and provide new ideas and methods to solve these two problems, thus improving the accuracy and diversity of recommendations. Therefore, in this paper, we use Citespace as the visualization aid and take the relevant literature of the web of science database from 2000 to 2020 as the research object, and conduct the visualization analysis from macro to the micro of the representative countries, keywords, and co-cited literature through scientific knowledge mapping, and discuss the advantages and disadvantages of CDR. The strengths and weaknesses of common techniques in the field of recommendation and their applicability are discussed to provide suggestions for the current challenges and future research directions: 1. CDR has been born in 2001 and developed slowly before 2014, and developed rapidly due to the proposed migration learning techniques in 2014. Therefore, it is suggested that the results of previous studies can be used rationally. 2. A mature collaborative network can make good use of the strengths and specialties of each institution to advance research development. For example, the University of Science and Technology Beijing, University of Technology Sydney, and Zhejiang University all specialize in migratory learning technology, so these institutions can take this as the main research direction and form a cooperative network. 3. The key terms for the evaluation and application of CDR systems are mainly focused on cold start problems, personalization, and social media. It can be expected that personalized user needs to be combined with social media data cold start problem solving, which will to a certain extent promote the application and development of CDR systems in more areas. 4. CDR in improving the recommendation of diversity and system robustness of the relevant research is relatively small. In addition, cross-domain also brings new problems to the recommendation, such as data heterogeneity between different domains, how to balance the ratio of multiple data sources in the case of multiple data sources, and how to quickly handle multiple domain transfers. 5. The sensor can be applied to other CDR systems in more life scenarios, such as education and health care field and food, medical, and sports field combined to provide a customized diet The sensor can be used in other life scenarios, such as education and health care, food, medical care, and sports, to provide customized diet, disease pre-treatment, and sports advice.