Cross-Domain Collaborative Filtering Recommendation without Overlapping Users and Items
Issued Date
2025-01-01
Resource Type
Scopus ID
2-s2.0-105032394095
Journal Title
Jcsse 2025 22nd International Joint Conference on Computer Science and Software Engineering
Start Page
75
End Page
80
Rights Holder(s)
SCOPUS
Bibliographic Citation
Jcsse 2025 22nd International Joint Conference on Computer Science and Software Engineering (2025) , 75-80
Suggested Citation
Tupwong A., Marukatat R. Cross-Domain Collaborative Filtering Recommendation without Overlapping Users and Items. Jcsse 2025 22nd International Joint Conference on Computer Science and Software Engineering (2025) , 75-80. 80. doi:10.1109/JCSSE67377.2025.11297910 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/115772
Title
Cross-Domain Collaborative Filtering Recommendation without Overlapping Users and Items
Author(s)
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
In recent years, single-domain recommendation methods have seen rapid development and significant success. However, they encounter problems that hinder their performance, such as sparse ratings in user-item rating matrix and cold start for new users. Cross-domain recommendation alleviates these issues by leveraging rating information from a well-established source domain and conveying it to a target domain. Such techniques usually rely on the existence of shared users or items in the two domains. However, such overlap is seldom found in real-world scenarios, which complicates the application of these methods. Therefore, our research proposes a cross-domain collaborative filtering recommendation method that does not rely on overlapping users or items. Instead, it matches items across domains based solely on their English text descriptions, such as movie overviews from the Movie domain and book descriptions from the Book domain. We evaluated a few different combinations of word embeddings (Word2Vec, Doc2Vec, and GloVe) and semantic similarity methods (cosine similarity and soft cosine similarity) for cross-domain item matching. The results indicate that soft cosine similarity, enhanced by Word2Vec word embeddings, was the best combination to capture complex semantic relationships in textual data. Moreover, sparsity reduction was found to be critical for improving the accuracy of recommendations.