Music Similarity Vector Construction via Recursive Embedding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining music similarity are time-consuming, biased towards popular songs, unreliable due to metadata inaccuracies, and computationally expensive, especially when dealing with large datasets of music artists and songs.
Innovation Solution
The Music Mapper system constructs coordinate vectors for music items by recursively embedding them into a multi-dimensional space using techniques like Landmark MDS or Fast Sparse Embedding, allowing for efficient similarity inference and playlist generation without requiring extensive computational audio matching analyses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If collaborative filtering techniques are used to combine user preferences, then similarity lists can be generated, but less well known music is rarely identified as similar
Solution Approach 1:
The patent introduces an intermediary similarity graph structure that mediates between user preferences and music similarity. Instead of directly using user lists, the system builds a graph where music items are nodes and similarity relationships are edges, allowing less popular music to be connected through intermediate nodes to popular music, thus resolving the bias toward well-known songs.
Solution Approach 2:
The patent transforms the one-dimensional user preference lists into a multi-dimensional similarity graph structure. By representing music items as nodes in a graph with multiple dimensions of similarity relationships, the system can identify similar music beyond what traditional linear user lists capture, including less popular tracks.
2Productivity
If metadata comparison is used to determine similarity, then matching songs can be selected quickly, but inaccuracies in metadata lead to incorrect similarity results
Solution Approach 1:
The patent implements feedback mechanisms where the similarity graph is continuously updated based on actual listening patterns and user interactions. This feedback loop allows the system to correct metadata inaccuracies by adapting the similarity relationships based on real-world usage data, maintaining both speed and accuracy.
Solution Approach 2:
The similarity graph acts as an intermediary layer between metadata and actual music similarity. Instead of directly comparing metadata fields, the system uses the graph structure to mediate similarity determination, allowing it to overcome metadata errors by relying on established similarity relationships in the graph.
3Productivity
If beat structure analysis is used to identify similar music, then BPM matching can be performed, but the method performs poorly on non-beat oriented music
Solution Approach 1:
The patent creates a universal similarity graph framework that can handle multiple music genres and styles uniformly. The graph structure accommodates different types of similarity relationships (temporal, spectral, structural, stylistic) without requiring separate analysis methods for each genre, making the system versatile across all music types including non-beat oriented genres.
Solution Approach 2:
The system dynamically changes the parameters used for similarity measurement based on the music genre and characteristics. For beat-oriented music, BPM and rhythmic parameters are emphasized, while for non-beat oriented music, the system switches to spectral, harmonic, and stylistic parameters, allowing the same framework to adapt to different music types.
4Measurement precision
If conventional MDS algorithms are used to embed music similarity graphs, then similarity can be visualized, but computational overhead becomes prohibitively large for large datasets
Solution Approach 1:
The patent segments the large music similarity graph into manageable components and uses incremental embedding techniques. Instead of processing the entire graph at once with conventional MDS, the system builds the embedding incrementally by adding nodes and edges in smaller batches, significantly reducing computational overhead while maintaining embedding accuracy for large datasets.
Solution Approach 2:
The system performs preliminary actions by pre-computing and storing similarity relationships in the graph structure during data ingestion. This preliminary processing allows the actual MDS embedding to be performed more efficiently on a pre-organized data structure, reducing the computational burden during the embedding phase.
Data Source
AI summary
A “Music Mapper” automatically constructs a set coordinate vectors for use in inferring similarity between various pieces of music. In particular, given a music similarity graph expressed as links between various artists, albums, songs, etc., the Music Mapper applies a recursive embedding process to embed each of the graphs music entries into a multi-dimensional space. This recursive embedding process also embeds new music items added to the music similarity graph without reembedding existing entries so long a convergent embedding solution is achieved. Given this embedding, coordinate vectors are then computed for each of the embedded musical items. The similarity between any two musical items is then determined as either a function of the distance between the two corresponding vectors. In various embodiments, this similarity is then used in constructing music playlists given one or more random or user selected seed songs or in a statistical music clustering process.