Visualising the Evolution of Cross-Domain Recommendation Systems

Sammanfattning

Introduction: The thesis performed a diachronic survey and semantic network analysis (SemNA) of a corpus of cross-domain recommendation (CDR) abstracts to visualise the evolution of the field. By comparing the semantic clusters in network graph representations of the corpus, the thesis could map the evolution of the CDR domain between 2005 and April 2025. Research Question: “How have the semantic clusters in the cross-domain recommendation literature changed between the periods 2005–2015, 2016–2020, and 2021–2025?” Method: A literature search was conducted in ACM, IEEE, Scopus and Web of Science, which yielded 1247 articles after deduplication and removal of unrelated records. The abstracts were divided into three time periods, yielding three corpora. SemNA was conducted in two steps: first, network graphs were generated with the VOSviewer software, then, the networks were structurally analysed and compared. Results: The results show how the semantic clusters changed from fundamental recommendation system concepts such as filtering techniques 2005–2015, to machine learning 2016–2020, towards artificial intelligence 2021–2025. Furthermore, the results indicate a shift in CDR and society from long-format media such as movies and books towards short-format media such as images and videos. Discussion: The findings demonstrate that SemNA can effectively reveal both technical evolution and societal shifts through academic literature analysis, representing the first diachronic visualisation of the CDR domain. SemNA enabled a mapping of the CDR landscape, but was limited in its potential to comprehensively analyse literature. Future research should combine SemNA with manual reading of the full texts.