EUVIMEDEUVIMED entdecken
Crossref

Improvement of a semantic movie retrieval method based on weighted aggregation of multi-source embeddings

А. І. Плюснін, Р. M. Малий

Scientific Bulletin of UNFU · 25.06.2026 · 36 (3)

DOI: 10.36930/40360304 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Scientific Bulletin of UNFU, online erschienen am 25.06.2026. Verfasst von А. І. Плюснін und 1 weiteren Autor:innen. Die Arbeit nennt 14 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The features of improving semantic movie retrieval were investigated by combining heterogeneous textual sources that describe different aspects of film content. A weighted aggregation approach was analyzed using dense vector representations generated from metadata, user reviews, and English subtitles on a dataset of 100 movies and 50 natural-language queries with predefined relevant movies. Cosine similarity, Recall@5, Hit Rate@5, Mean Reciprocal Rank, bootstrap confidence intervals, and the Wilcoxon signed-rank test were appliedto evaluate standalone and combined configurations. It was established that review-based embeddings achieved the highest standalone effectiveness in semantic movie retrieval, with Recall@5 = 0.63, Hit Rate@5 = 0.90, and MRR = 0.78, outperforming metadata and subtitle embeddings. It was found that reviews are better aligned with natural-language queries because they contain evaluative descriptions, thematic summaries, emotional impressions, and references to narrative elements. It was determined that subtitle embeddings, despite the largest text volume, did not provide the best ranking quality of relevant movies because dialogue contains filler phrases, context-dependent utterances, and semantically neutral fragments. It was confirmed that weighted aggregation of heterogeneous textual representations improved retrieval effectiveness of relevant movies compared with standalone representations: the balanced source-weight configuration configuration achieved the highest mean MRR = 0.89, whereas the reviews-heavy configuration achieved the highest mean Recall@5 = 0.73. Statistical testing of the experimental results was conducted and confirmed significant improvements of weighted configurations over the best standalone textual source for selected metrics; however, the differences between the balanced and reviews-heavy configurations were not statistically significant. It was therefore substantiated that, these configurations should be interpreted as comparable weighted strategies with different retrieval profiles. Query categories were analyzed, and it was established that descriptive queries tend to align with metadata, subjective queries with reviews, and contextual or character-based queries with several sources. The benefits of this study include determining the contribution of metadata, reviews, and subtitles to semantic movie retrieval and confirming the effectiveness of weighted source combination. Future research should extend the dataset, compare several embedding models, and develop adaptive query-dependent weighting strategies.

Zitieren

APA

Плюснін, А. І., & Малий, Р. M. (2026). Improvement of a semantic movie retrieval method based on weighted aggregation of multi-source embeddings. Scientific Bulletin of UNFU, 36(3). https://doi.org/10.36930/40360304

Vancouver

Плюснін АІ, Малий РM. Improvement of a semantic movie retrieval method based on weighted aggregation of multi-source embeddings. Scientific Bulletin of UNFU. 2026;36(3). doi:10.36930/40360304

BibTeX
@article{2026,
  author = {Плюснін, А. І. and Малий, Р. M.},
  title = {Improvement of a semantic movie retrieval method based on weighted aggregation of multi-source embeddings},
  journal = {Scientific Bulletin of UNFU},
  year = {2026},
  volume = {36},
  number = {3},
  doi = {10.36930/40360304}
}
Kostenlos starten →