KR20140111152A - Music recommendation system and method thereof - Google Patents

Abstract

The present invention relates to a music recommendation system and a method thereof, which can recommend music reflecting musical preference of a user using AHP, a hierarchial decision making tool, and can recommend music with high preference by reflecting musical preference of a user continuously through feedback based on a Bayesian network. The music recommendation system according to the present invention includes: a user terminal which receives an AHP element comprising musical characteristics reflected with comparative advantage and then transmits the AHP element to the music recommendation server, and displays music recommended from the music recommending server; and the music recommendation server which calculates a recommendation value through calculation based on AHP on the basis of the AHP element by being connected to the user terminal in communication, and recommends the music on the basis of the calculated recommendation value.

Description

[0001] MUSIC RECOMMENDATION SYSTEM AND METHOD THEREOF [0002]

The present invention relates to a personalized adaptive music recommendation system and method that reflects user preferences.

The present invention relates to a music recommendation system and method, and more particularly, to a music recommendation system using a hierarchical decision-making tool, AHP, which recommends music reflecting user's musical preference, So that music with high preference can be recommended. The present invention relates to a music recommendation system and method based on a Bayesian network.

In recent years, due to the spread of smart devices and the development of communication networks, it has become possible to listen to music through online using smart devices, not limited to places.

According to this tendency, sites providing sound sources are developing various technologies for providing music according to the user's taste, but also providing music reflecting the preference.

The music recommendation system and method have been disclosed in Japanese Patent Application Laid-Open No. 10-2008-0002348 as one of the techniques of providing music reflecting the preferences.

The music recommendation system and method described above determine an area from a maximum concentrated node having the highest degree of waveform concentration detected from sound source data to a certain number of nodes as a representative melody, present a test music to a user, After the preferred representative melody is collected, the representative melody of the sound source data is compared with the representative melody of the user to recommend sound source data similar to the representative representative melody of the user.

Such a music recommendation system and method can recommend a song suitable for a user's taste based on a user's preference.

However, representative tones are composed of at least one of intro, verse, transitional bridge, primary bridge, chorus, hook, interlude, and out tro. The order may also be varied. That is, the representative melody is widely varied according to each music, and it is difficult to recommend the music to the user because the music is not diversified and limited to the genre of some music, and the process of collecting the representative melody of the user, It takes a considerable amount of time in the process of detecting the music, and it takes a long response time until the music is recommended.

Another technique of music recommendation is disclosed in Japanese Patent Laid-Open No. 10-2009-0000232, which describes a system and method for recommending context-aware case-based music.

The situation-aware case-based music recommendation system and method uses case-based reasoning technique to extract information about music that has been heard by at least one other user similar to the user in past situations similar to external situation information and information about other users And recommends music based on the collected information. At this time, the external situation information includes at least one of season, month, day, weather, minimum temperature, maximum temperature, and average temperature.

The context awareness case based music recommendation system and method have an advantage that the music can be recommended to the user based on the information that other users have heard in the environment most similar to the user environment.

However, when music is recommended to the user, the music is recommended based on the past situation of the user other than the past situation of the user, so that the music preferred by the user is recommended rather than the music preferred by the user, There is a problem that each music preference is not properly reflected.

SUMMARY OF THE INVENTION The present invention has been made in order to solve the above problems, and it is an object of the present invention to provide an apparatus and method for recommending music reflecting a musical preference of a user using a hierarchical decision-making tool, AHP, A music recommendation system based on a Bayesian network and a method for recommending music adapted to a user's preference by continuously reflecting a musical preference of a user through the music recommendation system.

In order to solve the above problems, a music recommendation system based on a Bayesian network according to the present invention receives an AHP element having music characteristics reflecting a comparative advantage and transmits the received AHP element to a music recommendation server, A user terminal; And a music recommendation server for calculating a recommendation value based on the AHP-based operation based on the received AHP element in communication with the user terminal, and recommending music based on the calculated recommendation value; And a control unit for controlling the control unit.

According to another aspect of the present invention, there is provided a music recommendation method based on a Bayesian network, comprising: a position detection step of detecting a position of a user from a GPS; An importance level detecting step of detecting importance of an AHP element made up of music characteristics through a pair comparison; A music recommendation step of recommending music through Bayesian network reasoning to the AHP element whose importance is selected based on the detected user's position; And a music display step of displaying the recommended music; And a control unit for controlling the control unit.

The present invention has the effect of improving the satisfaction with the recommended music by recommending the music reflecting the musical preference of the user using the hierarchical decision-making tool AHP.

The present invention has the effect of improving the satisfaction with the recommended music by recommending the music continuously reflecting the musical preference of the user through the feedback based on the Bayesian network.

1 is a configuration diagram of a music recommendation system according to an embodiment of the present invention.
2 is a configuration diagram of a user terminal and a music recommendation server in a music recommendation system according to an embodiment of the present invention.
FIG. 3 is a diagram illustrating a hierarchical structure according to an embodiment of the AHP element in the music recommendation system according to the present invention.
4A to 4D are diagrams illustrating an example of a progress process displayed on a user terminal in the music recommendation method according to the present invention.
5 is a flowchart illustrating a process of performing a music recommendation method according to an embodiment of the present invention.
FIG. 6 is a graph showing a music recommendation satisfaction using the music recommendation system according to the present invention.

The terms and words used in the present specification and claims should not be construed as limited to ordinary or dictionary meanings and the inventor can properly define the concept of the term to describe its invention in the best possible way And should be construed in accordance with the principles and meanings and concepts consistent with the technical idea of the present invention.

Therefore, the embodiments described in the present specification and the configurations shown in the drawings are merely the most preferred embodiments of the present invention and are not intended to represent all of the technical ideas of the present invention. Therefore, various equivalents It should be understood that water and variations may be present.

Before describing the present invention with reference to the accompanying drawings, it should be noted that the present invention is not described or specifically described with respect to a known configuration that can be easily added by a person skilled in the art, Let the sound be revealed.

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a music recommendation system, which recommends music reflecting musical preferences of a user using a hierarchical decision-making tool, continuously reflects musical preferences of a user through feedback based on a Bayesian network, And a music recommendation system based on a Bayesian network that can improve the satisfaction.

FIG. 1 is a configuration diagram of a music recommendation system according to an embodiment of the present invention, and FIG. 2 is a configuration diagram of a user terminal and a music recommendation server in a music recommendation system according to an embodiment of the present invention.

The user terminal 100 refers to a device that can be connected to the server 200 via a wired / wireless communication network. The user terminal 100 may be a personal computer, a notebook, a smart phone, a tablet, A module 120, a feedback module 130, and a display module 140.

The sensing module 110 senses the position of the user using the GPS. At this time, it is possible to include the user's personal information, and the user's personal information may include the user's name, gender, and age.

The input module 120 is a module that receives the importance of the music characteristics preferred by the user. In this case, the music characteristics may be classified into a release year, a genre, and a group, and the classified music characteristics may be further subdivided. Therefore, the user inputs a music characteristic that he or she desires to be recommended according to the surrounding situation and psychological factors (joy, sadness, depression, calmness, etc.). Here, the music characteristic is used as an AHP element.

The feedback module 130 is for increasing or decreasing the preference for the recommended music as a module for sending feedback on the recommended music. That is, the feedback module 130 is for continuously reflecting the musical preference of the user, and feeds back the non-preferred music including the reason for the recommended music.

The display module 140 provides a user-centered interface to display the input module 120, the feedback module 130, and the recommended music.

The sensing module 110, the input module 120, the feedback module 130 and the display module 140 constituting the user terminal may be provided to the user terminal 100 as one application, Music recommendation through communication with the music recommendation server 200 described later can be performed.

The music recommendation server 200 is connected to the user terminal 100 through a communication network and recommends music based on the AHP based on the music characteristics transmitted from the user terminal 100. The feedback data is transmitted through the Bayesian network inference to the user- A music feature storage module 230, a context storage module 240, and a sound source providing module 250. The AHP calculation module 210, the Bayesian network reasoning module 220, the music characteristics storage module 230, the context storage module 240,

The AHP operation module 210 is a module for recommending music based on music characteristics reflecting preferences for the music to be heard.

The AHP element can be composed of a reference element that classifies music characteristics into major categories and a detailed element that re-classifies reference elements. For example, the AHP element may be composed of a release year, a genre, and a group as a reference element, and each reference element may be represented by a detail element of a lower layer. FIG. 3 of the attached drawing shows an upper layer and a lower layer And the connection relationship of the elements of the hierarchy.

The AHP element can be entered differently depending on the user's situation and condition. In other words, even if a music recommendation is received from the same location, the recommended music can be changed by inputting the importance of the reference element and the detailed element differently according to the state in which the ballad music is desired to be heard or the state in which the user wishes to listen to the dance music according to the psychological state of the user do.

For music recommendation, when the reference element is divided into release year, genre, and group, the lower level of the release year may be composed of the year in which each music is released, or may be grouped into groups of 10 years.

In addition, the lower level of the genre can be composed of religious music, new age, pop and song, and religious music and New Age have subdivided into domestic and foreign elements, , Rock / dance, and ballads.

The group can be composed of solos and groups as a lower layer, and distinguishes whether the singer providing the sound source is a solo singer or a group singer.

The hierarchical structure of the AHP elements shown in FIG. 3 is according to one embodiment, and may be configured to be changeable according to the tendency and intention of the user.

When the AHP elements are separated into layers and the comparative advantage of their causal relationships is determined, a recommendation value for music is calculated. Music recommendations are made based on this recommendation value. Music recommendation through the AHP calculation process consists of pair comparison, importance calculation and recommendation value calculation process.

(1) Pair comparison

The pair comparison compares pairs of elements in the immediate lower layer that contribute to achieving the goals of the elements in the upper layer so as to create a matrix by comparing the reference elements in the immediate lower layer with each other, The degree of contribution to the reference element of the upper layer is given. At this time, relative importance can be given on the scale of 9 points.

At this time, if the reference element of the layer to be compared is composed of n reference elements, the number of pairs of comparisons is compared with n (n-1) / 2 times.

When separating the reference element among the AHP elements into genre, release year, and group, each user assigns relative importance to each element.

For example, a genre can be selected to have a significance of five times and one times greater than a year of release and a group, respectively. In other words, the genre has a relative importance of 5 compared to the release year, and has a relative importance of 2 compared to the group. On the contrary, the release year has a relative importance of 1/5 of the genre, .

In addition, detailed elements can also be given relative importance. For example, if the relative importance of the genre is chosen to be higher than the importance of the release year and whether it is a group, a detailed comparison of the genre's details is made and a detailed comparison of the genre is made. When the detailed elements of the genre are composed of religious music, new age, pop and songs as shown in the figure, a pair comparison for each detail element is displayed, and the user compares the genre with a response to the genre.

Likewise, if the importance of a song among the detailed elements of a genre is selected to be higher than the other sub-elements, it is natural that a pair comparison of the availability is made.

This paired comparison is to select a comparative advantage for which of the two selected elements will be the more important factor, a pair comparison for the reference elements and a pair comparison for the detail elements. For example, if the reference element is divided into genre, release year, and group, the pair comparison is configured to select a user comparison advantage by setting the genre, release year, genre and group, release year and group.

FIG. 4A shows a pairwise comparison of the sub-elements with respect to whether or not the reference elements are grouped. Solo and group comparisons are made by dividing the details of the group by solo and group, and by choosing the solo and group importance.

(2) Calculation of Importance

The importance calculation is the process of calculating the importance of the relationship between the user's preference and other factors. If the relative importance of comparison of n elements to be subjected in one layer as w _i, a _ij of the pair-wise comparison matrix can be estimated by w _i / w _j. Therefore, the relation between a _ij and w _i can be expressed as a _ij = w _i / w _j (i, j = 1, 2, 3, ... n).

Therefore, the matrix A consisting of a _ij

로 나타낼 수 있다.

.

The importance calculation is obtained by dividing the horizontal axis sum of matrix A by the number of reference elements. The value obtained by adding the vertical axis of significance is 1.

At this time, the process that can be done in conjunction with the importance calculation is consistency verification. The coherence test verifies the consistency of the calculated significance. The coherence rate is a measure of the coherence index (CI) and consistency ratio (CR) to verify that they are consistent. If the coherence ratio (CR) is less than 0.1, it is determined to be consistent.

The coherence index (CI) is calculated by the following equation (1).

Where λ _max is the maximum eigenvalue of the pair of comparison matrices divided by the matrix product / significance sum divided by n, n is the number of reference elements, and CI is the consistency index.

The consistency ratio (CR, Consistency Ratio) as a value of the coherence index is calculated as shown in Equation (2).

Where CI is the consistency index, RI is the random number, and CR is the coherence rate.

In particular, the RI value is a value that can be varied depending on the number of elements. That is, a value of 1 to 9 is arbitrarily set to create a reciprocal matrix and the average coherence index of the matrix is calculated.

(3) Calculating the recommended value

The recommendation value calculation is a process of calculating the music to be recommended by combining the user's preferred factors.

By calculating the overall importance of the user's preferred basic elements, the user is identified with the preferred AHP element and the music is recommended through the identified element.

Therefore, if the recommendation value is calculated, music having an upper recommended value is selected from among the calculated recommendation values, and music is recommended. At this time, the recommended music may be a plurality of music having the recommended value of the upper level.

Based on the AHP - based music characteristics, the recommended value is calculated as a result, and music is recommended based on the recommended value. The recommended music is displayed on the user terminal 100. FIG. 4B shows a recommended music list, in which a music recommendation position, an AHP element, a result of AHP, and a satisfaction evaluation item are displayed. The recommended music list may not be a music that positively reflects a user's musical preference.

Accordingly, the user feeds back some unsatisfactory music among the recommended music to the music recommendation server 200 together with the reasons for dissatisfaction.

The feedback on reasons for dissatisfaction with recommended music is reflected in the recommendation of music.

At this time, the feedback is configured to be reflected when the music is recommended through reasoning of the Bayesian network.

The Bayesian network reasoning module 220 is a module for reflecting the contents of the feedback during music recommendation through Bayesian network reasoning about feedback.

The main characteristic of Bayesian networks is consistency and completeness in that there is only one probability distribution satisfying the constraints imposed on the Bayesian network. In other words, the probability can be calculated for all events that can be represented by a network variable only by probability distribution. Another feature of the Bayesian network is that there is an efficient algorithm that can calculate various probabilities without explicitly calculating the overall probability distribution. In particular, this algorithm can be applied regardless of the type of network. However, the efficiency and accuracy of the algorithm are affected by the network type and specific queries.

In the Bayesian network, the element represents the actual environment variable, and the arc represents the dependency between the variables. The network probability distribution can be calculated through the given data. After learning the network, if the variable for the situation is set, the conditional probability table and the independent condition are used based on the variable and the Bayesian inference algorithm Probability is calculated.

The music for the calculated probability is reflected in the recommendation value.

In the present invention, the probability is updated through the user's preferred feedback. That is, the probability that the user's favorite music is recommended is changed through the feedback of the user.

The feedback in the present invention can be configured to be input by being divided into implicit feedback and explicit feedback.

Implicit feedback can consist of general dissatisfaction reasons among recommended music dissatisfaction reasons. Common reasons for dissatisfaction can be determined through surveys. The reasons for dissatisfaction with the music were found to be non - favored singer, non - favored music, and inappropriate place. In addition, the percentage of non - preferred singers was the highest. Thus, the implicit feedback in the present invention can be designated as a non-preferred mantissa, and the interface of one embodiment to implicit feedback is shown in FIG. 4B.

Explicit feedback can consist of other reasons for dissatisfaction, except for reasons of dissatisfaction with explicit feedback. For example, if the reasons for dissatisfaction are non-favored singer, non-favored music, inappropriate place, etc., and the non-favored singer is the cause of dissatisfaction with implicit feedback, explicit feedback may consist of non-favorable music and inappropriate place, And the like. Thus, the explicit feedback corresponds to another dissatisfaction cause of the implicit feedback, which is indicated when "no" is selected for the implicit feedback shown in FIG. 4c, and is shown in FIG. 4d as one embodiment of the explicit feedback.

Therefore, since the Bayesian probability is varied through feedback, the music preferred by the user is recommended, so that it is possible to continuously reflect the preference and thus, an optimal music recommendation can be performed.

The music property storage module 230 stores and manages reference elements and detail elements corresponding to the AHP elements. As described above, the reference element can be composed of a release year, a genre, and a group, and the detailed elements according to the release year are composed of the year in which the music is announced. The detailed elements of the genre include religious music, And songs, and the detailed elements of grouping can be divided into solo and group. It is a matter of course that the sub-elements can be further divided into sub-sub-elements.

In addition, the music characteristics are stored and managed in addition to the AHP elements as well as artist (artist name), title (song name), file path (location where the sound source is stored) and mood (mood), and stored as an XML (extensible markup language) .

The XML file has the advantage that the user can edit the structured database and easily extract and use the music characteristics (element) through the XML parser.

The context storing module 240 stores and manages the position where the user has been recommended the music, the AHP element, the AHP result, and the feedback. The AHP result stores the artist name and song name, and feedback is stored for satisfaction and reasons for dissatisfaction.

That is, the context storage module 240 stores feedback based on the AHP element and reflecting the user's preference. At this time, the position of the user according to the feedback of the user is also stored.

When a signal requesting music recommendation is input, music recommendation is performed based on user feedback data stored in the context storage module 230 and user's location data. For example, if the recommended music based on the same AHP element at the same place or nearby is fed back to non-favored music and stored, the next music recommendation will be referred to as non-preferred music based on the same AHP element in the same place The probability is set low. Therefore, music recommendation in the same place can be performed based on the location of the user.

On the other hand, since the recommended music is reflected by applying the AHP element which is a music characteristic that the user desires to listen according to the mood of the user, even the music fed back to the non-preferred music can be re-recommended according to the AHP element inputted by the user.

In addition, the context storage module 240 stores the location of the user, that is, the location of the detected user terminal 100. If the cause of the feedback is unsatisfactory, the corresponding music based on the same location is not recommended again Lt; / RTI >

The sound source storage module 250 is a module for storing a sound source for the recommended music, and is a sound source in the file path of the music property storage module 230.

Accordingly, in order to listen to the recommended music, the music selected by the user's operation may be provided by streaming the sound source stored in the sound source storage module 250 to the user terminal 100 and outputting the stream. Here, it is a matter of course that the tone generator storage module 250 may be configured as a separate server.

5 is a flowchart illustrating a process of performing a music recommendation method based on a Bayesian network according to an embodiment of the present invention.

The music recommendation method based on the Bayesian network according to the present invention includes a position detection step S10, an importance degree detection step S20, a music recommendation step S30, a music display step S40, a feedback reception step S50, Step S60.

(1) Position detection step (S10)

The position detection step S10 is a step of detecting the position of the user based on the user position detected by the GPS using the GPS of the user terminal 100. [

(2) Importance level detection step (S20)

The importance level detection step S20 is a step of detecting the importance of the AHP element made up of the music characteristics through a pair comparison.

Music characteristics are inherent characteristics of music. The characteristics of music can include release year, genre, group, religious music, new age, pop, song, hip-hop, rock, dance, ballad, In this case, the musical characteristics are divided into the reference element and the detailed element, and the upper and lower layers are formed. The pair comparison is performed through the reference element, and the pair comparison can be performed through the detailed element.

(3) Music recommendation step (S30)

The music recommendation step (S30) is recommended by calculating a recommendation value for the comparative advantage of the selected AHP element based on the AHP and selecting a recommendation value calculated as an upper value as the recommendation music.

(4) Music display step (S40)

The music display step S40 is a step of transmitting and displaying the recommended music to the user terminal 100 displaying the music. At this time, the recommended music to be displayed may be composed of a plurality of.

(5) Feedback receiving step (S50)

The feedback receiving step S50 is a step of receiving user feedback on the indicated recommended music.

The recommended music may be the user's preferred music, but it may be the non-preferred music. Feedback is a reason for dissatisfaction with the recommended music, and can be configured to be received in terms of implicit feedback and explicit feedback.

Explicit feedback can consist of other reasons for dissatisfaction, except for reasons of dissatisfaction with explicit feedback. Implicit feedback consists of the most common causes of dissatisfaction and can be displayed before explicit feedback. Accordingly, the most common cause of dissatisfaction can be displayed first so that the user can receive feedback on the cause of dissatisfaction as soon as possible. This feedback is configured to be reflected in music recommendations through inference of the Bayesian network.

(6) Music recommendation step (S60)

The music re-recommendation step (S60) calculates a recommendation value reflecting the feedback based on the AHP element and the Bayesian network whose importance is selected based on the detected user's position, and re-recommending music based on the calculated recommendation value . Therefore, when the non-preferred music is recommended, it is configured to receive the feedback on the recommended music and to recommend the music by reflecting the recommendation.

Next, a music recommendation system was implemented and user satisfaction was evaluated in order to evaluate the user satisfaction with the music recommendation system and method according to the present invention.

The implementation environment of the system according to the present invention is configured as shown in [Table 1].

Item Specifications  Mobile Device  LG Vega X +  RAM  512MB  OS  Froyo  Language  Java for Android  Programming Tool  Eclipse

At this time, the music data stored for the 200 songs are stored in the memory.

As shown in FIG. 4A, the importance is input through a pair comparison of AHP elements, and music recommended through Bayesian network reasoning is output as shown in FIG. 4B.

The feedback of the user's dissatisfaction with the recommended music was input according to one embodiment of FIGS. 4c and 4d, and the user's preference through feedback was measured.

The preference was measured by dividing into 3 groups into 4 groups.

FIG. 6 is a graph showing satisfaction of music recommendation according to the present invention, and satisfaction of each group was low at the beginning of experiment.

The reason is that the prepared sound source data is not sufficient to satisfy a large number of users, and even if the standard suitable for the user's preference is recommended, unfamiliar music is generated. However, as the number of music recommendation increases, music satisfying the user preference is recommended, and the average satisfaction level of the users of each group is improved to 90% as the number of music recommendation increases.

The music recommendation system and method according to the present invention can improve the satisfaction with the recommended music by recommending the music reflecting the musical preference of the user using the hierarchical decision-making tool AHP, and the feedback based on the Bayesian network It is possible to improve the satisfaction with the recommended music by recommending the music that continuously reflects the user's musical preference.

1 to 6 have described only the main points of the present invention. As far as various designs can be made within the technical scope thereof, the present invention is not limited to Figs. 1 to 6 It is self-evident.

100: User terminal
110: sensing module 120: input module
130: feedback module 140: display module
200: Music recommendation server
210: AHP operation module 220: Bayesian network reasoning module
230: Music feature storage module 240: Context storage module
250: sound source providing module

Claims (6)

A music recommendation system comprising:
A user terminal 100 that receives an AHP element having music characteristics reflecting a comparative advantage and transmits the received AHP element to a music recommendation server 200 and displays recommended music provided from the music recommendation server 200; And
The recommendation value is calculated through AHP calculation based on the AHP element received in communication with the user terminal 100, and the music recommendation server 200 recommends music based on the calculated recommendation value; The music recommendation system comprising:
The method according to claim 1,
The music recommendation server 200
And a context storage module for storing and managing feedbacks including reasons for dissatisfaction with the music recommended by the user terminal 100, positions where the music is recommended, AHP elements, and AHP computation results,
Calculates a recommendation value using the feedback, and recommends music based on the calculated recommendation value.
The method of claim 2,
Wherein the calculated recommendation value is calculated by Bayesian network reasoning based on the feedback.
The method of claim 2,
The feedback
Implicit feedback designated as non-preferred
Non-favorable music, unfavorable place, and other explicit feedback designated as a music recommendation system based on Bayesian network.
In a music recommendation method,
A position detecting step (S10) of detecting the position of the user from the GPS;
An importance level detection step (S20) of detecting importance of AHP elements made up of music characteristics through a pair comparison;
A recommendation step (S30) of calculating a recommendation value based on the AHP of the AHP element whose importance is selected based on the detected user's position, and recommending music based on the calculated recommendation value; And
A music display step (S40) of displaying recommended music; Wherein the music recommendation method comprises:
The method of claim 5,
After the music display step S40,
A feedback receiving step (S50) of receiving user feedback on the indicated recommended music; And
A music re-recommending step (S60) for re-recommending music based on the received AHP element and the received feedback based on the Bayesian network based on the detected location of the user; The music recommendation method further comprising:

Images