TU Wien:Einführung in Information Retrieval VU (Hanbury)/Prüfung Music Retrieval 2025-01-21

1. Audio fingerprints should be designed to be sufficiently entropic to be independent of absolute positions in time.
2. Text retrieval methods have applications in music retrieval.
3. Item-based collaborative filtering scales better in real-world scenarios than user-based collaborative filtering when pre-calculating similarity lookup tables.
4. In contrast to text processing, in audio processing, determining semantic units and delimiters is trivial.
5. When using a recommender system, users need to actively express their information needs.
6. By picking the local maximum values in the spectrogram, fingerprint calculation becomes robust to equalization and absolute magnitude values.
7. One underlying assumption of collaborative filtering used for recommending items is that users who had similar taste in the past, will have similar taste in the future.
8. Calculating the adjusted cosine similarity on item profiles is equivalent to calculating the Pearson correlation on the transposed user-item rating matrix used for user-based collaborative filtering.
9. As with text, the query-by-example paradigm can be applied in the music domain for numerous tasks.
10. In item-based recommendation, the item similarity function must stem from rating data in the user-item matrix.
11. Matrix factorization approaches permit to represent users and items in a joint latent space.
12. The so-called gray sheep problem occurs for users who are the first to rate obscure items, not benefitting from better matches with other users.
13. Discrete Cosine Transform can decompose any periodic audio signal into sine waves of different frequencies and amplitudes.
14. Recommender systems are information filters.
15. Off-line evaluation of recommender systems might put too much emphasis on historic data and neglect current and future user needs.
16. MFCCs are well-suited features for melody-based retrieval.
17. In music recommendation, the "portfolio effect" might be a desired feature of the system.
18. With features extracted from audio, an inverted index as used in text retrieval can not be used.
19. For music retrieval models, the temporal order of extracted features is always a central requirement.
20. Audio fingerprinting has the goal to find other similar sounding music tracks to the query.