Peer To Patent

Home > Archived Applications > 20110112994 > Prior Art > Detail

Pre-Grant Publication Number: 20110112994

Musical piece recommendation system, musical piece recommendation method, and musical piece recommendation computer program

Please help the USPTO examine the application by evaluating the relevance of the publicly submitted prior art to the patent application. Peer To Patent forwards the Top 10 most relevant prior art submissions and their annotations to the USPTO.

Review this prior art and click on the thumbs up (or down) to indicate whether this submission should be forwarded to the USPTO.
If you login then you can add an annotation by typing in the box at the bottom of the screen to comment on the relevance of the prior art to the claims of the patent application.

Prior Art Detail

Annotations(0)

#1297Hybrid Collaborative and Content-based Music Recommendation

Applies to Claims 1,6,7

Submitted by: Damion JosephsLast updated: over 1 year ago

0 thumbs up 0 thumbs down

Summary / Description

Summary / Description	"This paper presents a hybrid music recommendation method that solves problems of two prominent conventional methods: collaborative filtering and content-based recommendation." "The colloaborative methods recommend pieces by considering someone else's ratings of those pieces." This was a paper published by Masataka Goto (among others)

Basic Information

Type of Prior Art	Print Publication

Publication Title *	Hybrid Collaborative and Content-based Music Recommendation

Author	Kazuyoshi Yoshii, Masataka Goto, Kazunori Komatani, Tesuya Ogata, Hiroshi G. Okuno

ISBN

Page Range	1-6

Medium	Other printed publication

Publication Date *	October 10, 2006

URL

Notes / To Do

Notes	Perhaps, this paper may be used as a basis for a 102 rejection - the publication was presented at: ISMIR 2006, 7th International Conference on Music Information Retrieval, Victoria, Canada, 8-12 October 2006, Proceedings. 2006 (Oct. 10, 2006)

Excerpt

Excerpt

"Content-based methods try to rank musical pieces on the basis of music-content similarity by representing user preferences in the music-content space." "To solve this problem, we associate rating and content data with newly-introduced variables that represents user preferences. In this paper, we use a Bayesian network called a three-way aspect model proposed by Popescul et al. [11]. This model has a set of latent variables that directly describe substantial preferences that cannot be observed. Those preferences are statistically estimated with theoretical proof. This will contribute to reliable recommendation."

Relevance

Claims

A musical piece recommendation system comprising:a rating history storage section that stores a plurality of rating histories obtained as a result of rating of at least one of a plurality of musical pieces performed by each of a plurality of users;

an acoustic feature storage section that stores acoustic features extracted from audio signals of the plurality of musical pieces;

a musical piece recommendation section that includes a probabilistic model including a plurality of variables and a plurality of parameters and that outputs a preference ranking of the plurality of musical pieces in response to a musical piece recommendation request from one of the users; and

a basic training section that receives the plurality of rating histories and the plurality of acoustic features as inputs and that estimates the plurality of parameters in the probabilistic model to maximize a likelihood of the plurality of parameters,

wherein the probabilistic model included in the musical piece recommendation section includes a musical piece variable node that stores variables related to the plurality of musical pieces, a user variable node that stores variables related to the plurality of users, a feature variable node that stores variables related to the plurality of acoustic features, and a latent variable node that stores a plurality of conceptual topics as latent variables,

a plurality of topic selection probabilities are defined between the user variable node and the latent variable node, each of the topic selection probabilities corresponding to a probability with which when one of the users is selected, the user selects each of the plurality of topics,

a plurality of musical piece selection probabilities are defined between the latent variable node and the musical piece variable node, each of the musical piece selection probabilities corresponding to a probability with which when one of the topics is selected, each of the plurality of musical pieces is selected,

a plurality of feature occurrence probabilities are defined between the latent variable node and the feature variable node, each of the feature occurrence probabilities corresponding to a probability with which when one of the topics is selected, each of the variables related to the plurality of acoustic features occurs for the topic,

the probabilistic model is obtained by formulating a three-way aspect model in which the topic selection probability, the musical piece selection probability, and the feature occurrence probability are independent of each other,

the formulated three-way aspect model includes the topic selection probability, the musical piece selection probability, and the feature occurrence probability as the plurality of parameters in a formula for calculating a generation probability of an event in which a certain user listens to a certain acoustic feature in a certain musical piece,

the musical piece recommendation section is configured to output, in response to a musical piece recommendation request from one of the users, a preference ranking of the plurality of musical pieces for the user using the topic selection probability for the user, the musical piece selection probability, and the feature occurrence probability, and

the musical piece recommendation system further comprises:

a first incremental training section that monitors the rating history storage section and that each time a change is made to the rating history or a new user is added, performs updating of or addition of the topic selection probability for the user for which the change is made to the rating history or for the new user such that the likelihood determined by the basic training section is kept maximized; and

a second incremental training section that monitors the acoustic feature storage section and that each time a new musical piece is added to perform addition to the acoustic features, performs addition to the musical piece selection probability related to the added musical piece such that the likelihood determined by the basic training section is kept maximized.

Relevance

Paper discusses latent variable system and training matrix. "The content of each piece is represented as a single vector of several acoustic features extracted from the corresponding audio signal."

Claim Chart

All

A musical piece recommendation method for causing a computer to perform the steps of:storing in a rating history storage section a plurality of rating histories obtained as a result of rating of at least one of a plurality of musical pieces performed by each of a plurality of users;

extracting a plurality of acoustic features from audio signals of the plurality of musical pieces and storing in an acoustic feature storage section the plurality of acoustic features extracted from the plurality of musical pieces;

outputting, in response to a musical piece recommendation request from one of the users, a preference ranking of the plurality of musical pieces using a probabilistic model, as a musical piece recommendation step, wherein the probabilistic model includes a musical piece variable node that stores variables related to the plurality of musical pieces, a user variable node that stores variables related to the plurality of users, a feature variable node that stores variables related to the plurality of acoustic features, and a latent variable node that stores a plurality of conceptual topics as latent variables, a plurality of topic selection probabilities are defined between the user variable node and the latent variable node, each of the topic selection probabilities corresponding to a probability with which when one of the users is selected, the user selects each of the plurality of topics, a plurality of musical piece selection probabilities are defined between the latent variable node and the musical piece variable node, each of the musical piece selection probabilities corresponding to a probability with which when one of the topics is selected, each of the plurality of musical pieces is selected, a plurality of feature occurrence probabilities are defined between the latent variable node and the feature variable node, each of the feature occurrence probabilities corresponding to a probability with which when one of the topics is selected, each of the variables related to the plurality of acoustic features occurs for the topic, the probabilistic model is obtained by formulating a three-way aspect model in which the topic selection probability, the musical piece selection probability, and the feature occurrence probability are independent of each other, the formulated three-way aspect model includes the topic selection probability, the musical piece selection probability, and the feature occurrence probability as a plurality of parameters in a formula for calculating a generation probability of an event in which a certain user listens to a certain acoustic feature in a certain musical piece, and the probabilistic model is configured to output, in response to a musical piece recommendation request from one of the users, a preference ranking of the plurality of musical pieces for the user using the topic selection probability for the user, the musical piece selection probability, and the feature occurrence probability;

receiving the plurality of rating histories and the plurality of acoustic features as inputs and estimating the plurality of parameters in the probabilistic model to maximize a likelihood of the plurality of parameters, as a basic training step, the basic training step being performed before execution of the musical piece recommendation step;

monitoring the rating history storage section, and each time a change is made to the rating history or a new user is added, performing updating of or addition of the topic selection probability for the user for which the change is made to the rating history or for the new user such that the likelihood determined in the basic training step is kept maximized, as a first incremental training step, the first incremental training step being performed after execution of the basic training step; and

monitoring the acoustic feature storage section, and each time a new musical piece is added to perform addition to the acoustic features, performing addition to the musical piece selection probability related to the added musical piece such that the likelihood determined in the basic training step is kept maximized, as a second incremental training step, the second incremental training step being performed after execution of the basic training step.

Relevance

Uses a latent variable system with a Bayesian network

Claim Chart

Some

The musical piece recommendation method according to Claim 6,wherein the basic training step comprises the steps of:calculating a plurality of rating vectors from the plurality of rating histories;

calculating a plurality of content vectors from the plurality of acoustic features,

calculating the likelihood of the parameters for the plurality of rating vectors and the plurality of content vectors, as a likelihood calculation step;

updating the plurality of topic selection probabilities, the plurality of musical piece selection probabilities, and the plurality of feature occurrence probabilities to maximize the likelihood, as an updating step; and

terminating the likelihood calculation step and the updating step when it is determined that an increase in the likelihood has converged.

Relevance

There is use of what is called "preference vectors"

Claim Chart

All

0 days left