Intelligent music playing method based on music integration platform

Abstract

The invention discloses an intelligent music playing method based on a music integration platform, and belongs to the technical field of music recommendation, and the method specifically comprises the steps: setting a monitoring period, and obtaining the playing data of a target object in the monitoring period; target music is determined according to the playing data, and the interestingness of any tag is calculated in sequence; obtaining audio information of music and performing feature extraction to obtain spectrum features of the music; calculating a spectrum feature similarity distance value between the to-be-recommended music and the target music; the method comprises the steps of obtaining a related label corresponding to any piece of to-be-recommended music, determining the interestingness of the recommended music according to the related label, correcting a similar distance value of the to-be-recommended music according to the interestingness, and performing music recommendation on a target object according to a correction result. According to the method, the robustness of similarity calculation of music recommendation and the accuracy and real-time performance of a recommendation algorithm are improved, so that the user satisfaction and the use experience are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of music recommendation, and in particular to a music intelligent playing method based on a music integration platform. Background Art

[0002] With the continuous improvement of productivity in modern society, people's entertainment and leisure activities have become diversified, and there are more choices for lifestyles. With the development and popularization of mobile Internet and mobile terminals, various mobile applications have gradually penetrated into people's work and life scenes. According to statistics, the usage rate of music applications remains at the top of the usage rate ranking of various entertainment applications. As one of the most important entertainment and spiritual pastimes for humans, music has shown new colors and power under the trend of mobility and socialization in the mobile Internet era, and plays an increasingly important role in people's social activities.

[0003] The current mainstream music recommendation system mainly uses collaborative filtering and LDA (Latent Dirichlet Allocation, topic model) technology. Collaborative filtering generates recommendations by analyzing users' listening behavior, but this method is easily affected by popular trends, resulting in the recommendation of popular songs. Although the traditional LDA model can reveal users' interest topics and corresponding song lists, it is very sensitive to training data and may cause errors in the training process due to corpus bias, thus affecting the accuracy of recommendations. Summary of the invention

[0004] The purpose of the present invention is to provide a music intelligent playing method based on a music integration platform to solve the following technical problems: The current mainstream music recommendation system mainly uses collaborative filtering and LDA (Latent Dirichlet Allocation, topic model) technology. Collaborative filtering generates recommendations by analyzing users' listening behavior, but this method is easily affected by popular trends, resulting in the recommendation of popular songs. Although the traditional LDA model can reveal users' interest topics and corresponding song lists, it is very sensitive to training data and may cause errors in the training process due to corpus bias, thus affecting the accuracy of recommendations.

[0005] The purpose of the present invention can be achieved through the following technical solutions: The music intelligent playing method based on the music integration platform is characterized by comprising the following steps: S1, setting a monitoring period, obtaining the playback data of the target object within the monitoring period, wherein the playback data includes the played music, the number of plays and the interactive behavior; the interactive behavior includes like, share and download; S2, taking the music played with interactive behavior as pending music, selecting any pending music and obtaining the number of times the pending music is played during the monitoring period, calculating the recommendation score of the pending music according to the number of times it is played, and if the recommendation score of the pending music is greater than or equal to a preset threshold, marking the pending music as target music; S3, obtain the recommendation scores of all played music and sum them up to obtain the total recommendation score W', take the played music without interactive behavior as the observed music, calculate the ratio of the recommendation score of any observed music to the total recommendation score, and mark the ratio as the interest X of the observed music, obtain the corresponding label of the target music and mark it as y1, y2, ..., yn, calculate the interest of the corresponding label of the target music in turn according to the interest of the observed music and mark it as P1, P2, ..., Pn; Obtain music that the user has not played and mark it as music to be recommended, obtain the frequency spectrum features of the target music and all the music to be recommended, calculate the similarity distance value between any music to be recommended and the target music according to the frequency spectrum features and mark it as C1, C2, ..., Cm; S4, obtain the label corresponding to any music to be recommended, compare the label corresponding to the music to be recommended with the label corresponding to the target music, obtain all overlapping labels, obtain the interest levels corresponding to all overlapping labels and sum them up, use the sum value as the interest level of the music to be recommended, correct the similarity distance value of the music to be recommended according to the interest level, sort all the music to be recommended in an ascending order according to the corrected similarity distance value, and recommend music to the target object according to the sorting result.

[0006] As a further solution of the present invention: In S2, the specific calculation process of the recommendation score is: ; Among them, ki is the number of times the i-th pending music is played, K is the number of times all music is played during the monitoring period, L and S are preset coefficients, and L is greater than S, and Ai is the total number of interactions of the i-th pending music.

[0007] As a further solution of the present invention: in S3, the target music corresponding labels include beat, emotion, singer name, genre and language.

[0008] As a further solution of the present invention: in S3, the calculation process of the interest degree is: Select any observed music, mark the interest of any corresponding label of the observed music as the interest of the observed music, obtain the corresponding label of the target music and calculate the interest of the corresponding label of the target music. The specific calculation formula is: ; Pm represents the interest of the target music corresponding to label m, i represents the played music corresponding to label m, Q represents the total amount of observed music corresponding to label m, Xi represents the interest of observed music i, m∈n, and n is the total number of corresponding labels.

[0009] As a further solution of the present invention: in S3, the specific process of spectrum feature extraction is: Get any music, divide it into frames according to the preset frame length, and get the audio frame sequence r1, r2, ..., rt, t is a positive integer, use Mel frequency cepstrum transform technology to generate the spectrum corresponding to each audio frame, select the frequency corresponding to the maximum peak in each spectrum, and get the frequency sequence p1, p2, p3, ..., pt, and calculate the variance of the frequency sequence through the variance formula; select all the amplitude values ​​of the spectrum corresponding to any audio frame, and calculate the power sum = (amplitude value) through the formula 2 , calculate the power sum corresponding to any audio frame, obtain the power sum sequence w1, w2, w3, ..., wt, calculate the variance logarithm of the power sum sequence; obtain the Mel spectrum of the sub-singing segment and input it into the preset neural network to obtain the eigenvalue of the target music timbre vector.

[0010] As a further solution of the present invention: the specific calculation process of the similarity distance value is: ; Among them, X is the variance of the target music frequency sequence, Y is the logarithm of the variance of the power and sequence of the target music, Z is the eigenvalue of the target music timbre vector, X' is the variance of the frequency sequence of the music to be recommended, Y' is the logarithm of the variance of the power and sequence of the music to be recommended, and Z' is the eigenvalue of the timbre vector of the music to be recommended.

[0011] As a further solution of the present invention: in S4, the specific process of correction is: Get the labels corresponding to the music to be recommended and the target music, and then get all the overlapping labels and mark them as y'1, y'2, ..., y'N, get the interest corresponding to all the overlapping labels and mark them as P1, P2, ..., PN, according to the calculation formula The corrected similarity distance value C' is calculated, where α is a preset correction coefficient and pi is the interest level of any overlapping label.

[0012] As a further solution of the present invention: if there are several target music, then obtaining any music to be recommended and calculating the similarity distance with the target music, sorting all the music to be recommended in ascending order according to the total similarity distance, and recommending music to the target object according to the sorting result; The calculation process of the total similarity distance is: ; Among them, C' i is the modified similarity distance value between the music to be recommended and any target music, Wi is the recommendation score of the ith target music, and h is the number of target music.

[0013] Beneficial effects of the present invention: In the present invention, the music played by the user during the monitoring period is first obtained, and the target music is determined according to whether there are interactive behaviors and the frequency of playing; in the music recommendation system, the user's interactive behaviors (such as likes, sharing and downloading) are important indicators for evaluating the user's preference for specific music, and these behaviors can directly reflect the user's preference for playing music. However, in actual application scenarios, a situation may occur: although a piece of music has specific preference parameters, its number of plays is relatively low. This may mean that the user has a high interest in the music when he heard it once, but did not choose to play it again in the subsequent time. Therefore, judging the user's liking for music based solely on the number of plays may result in insufficient accuracy. It is understandable that the higher the frequency of music playing, the higher the user's liking for the music. Frequent playback usually means that the user has a sustained interest and good impression of the music. Therefore, the target music is screened out through interactive behavior and playback frequency. Even when faced with music with a low number of playbacks but high preference parameters, the user's preference can be judged more accurately, thereby improving the relevance and satisfaction of the recommendation results, calculating the interest of each music tag, identifying the user's preference for different music features, and using the spectral features of the audio information to perform similarity calculation. It can be understood that the energy features of the audio can characteristically show the volume and intensity of the target music, and the frequency features of the audio can show the tone of the target music. At the same time, the Mel spectrum is input into a preset neural network, and the eigenvalues ​​of the timbre vector are extracted. The similarity distance is calculated according to the timbre, tone and pitch, and the similarity of different music can be distinguished. By correcting the similarity distance value according to the interest, the user's preference for different music tags can be balanced in the recommendation, and the personalization and relevance of the recommendation can be improved. By sorting and recommending the corrected similarity distance values, the present invention improves the robustness of the similarity calculation of music recommendation and the accuracy and real-time performance of the recommendation algorithm, allowing users to obtain richer and more personalized music recommendations, thereby improving user satisfaction and usage experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below in conjunction with the accompanying drawings.

[0015] Figure 1 It is a flowchart of the music intelligent playing method based on the music integration platform of the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] See also Figure 1 As shown, the present invention is a music intelligent playing method based on a music integration platform, comprising the following steps: S1, setting a monitoring period, obtaining the playback data of the target object within the monitoring period, wherein the playback data includes the played music, the number of plays and the interactive behavior; the interactive behavior includes like, share and download; S2, taking the music played with interactive behavior as pending music, selecting any pending music and obtaining the number of times the pending music is played during the monitoring period, calculating the recommendation score of the pending music according to the number of times it is played, and if the recommendation score of the pending music is greater than or equal to a preset threshold, marking the pending music as target music; S3, obtain the recommendation scores of all played music and sum them up to obtain the total recommendation score W', take the played music without interactive behavior as the observed music, calculate the ratio of the recommendation score of any observed music to the total recommendation score, and mark the ratio as the interest X of the observed music, obtain the corresponding label of the target music and mark it as y1, y2, ..., yn, calculate the interest of the corresponding label of the target music in turn according to the interest of the observed music and mark it as P1, P2, ..., Pn; Obtain music that the user has not played and mark it as music to be recommended, obtain the frequency spectrum features of the target music and all the music to be recommended, calculate the similarity distance value between any music to be recommended and the target music according to the frequency spectrum features and mark it as C1, C2, ..., Cm; S4, obtain the label corresponding to any music to be recommended, compare the label corresponding to the music to be recommended with the label corresponding to the target music, obtain all overlapping labels, obtain the interest levels corresponding to all overlapping labels and sum them up, use the sum value as the interest level of the music to be recommended, correct the similarity distance value of the music to be recommended according to the interest level, sort all the music to be recommended in an ascending order according to the corrected similarity distance value, and recommend music to the target object according to the sorting result.

[0018] In the present invention, the music played by the user during the monitoring period is first obtained, and the target music is determined according to whether there are interactive behaviors and the frequency of playing; in the music recommendation system, the user's interactive behaviors (such as likes, sharing and downloading) are important indicators for evaluating the user's preference for specific music, and these behaviors can directly reflect the user's preference for playing music. However, in actual application scenarios, a situation may occur: although a piece of music has specific preference parameters, its number of plays is relatively low. This may mean that the user has a high interest in the music when he heard it once, but did not choose to play it again in the subsequent time. Therefore, judging the user's liking for music based solely on the number of plays may result in insufficient accuracy. It is understandable that the higher the frequency of music playing, the higher the user's liking for the music. Frequent playback usually means that the user has a sustained interest and good impression of the music. Therefore, the target music is screened out through interactive behavior and playback frequency. Even when faced with music with a low number of playbacks but high preference parameters, the user's preference can be judged more accurately, thereby improving the relevance and satisfaction of the recommendation results, calculating the interest of each music tag, identifying the user's preference for different music features, and using the spectral features of the audio information to perform similarity calculations. It can be understood that the energy features of the audio can characteristically express the volume and intensity of the target music, and the frequency features of the audio can express the tone of the target music. At the same time, the Mel spectrum is input into the preset neural network, the eigenvalues ​​of the timbre vector are extracted, and the similarity distance is calculated based on the timbre, tone and pitch. The similarity of different music can be distinguished. By correcting the similarity distance value according to the interest, the user's preference for different music tags can be balanced in the recommendation, and the personalization and relevance of the recommendation can be improved. By sorting and recommending the corrected similarity distance values, users can obtain richer and more personalized music recommendations, thereby improving user satisfaction and usage experience.

[0019] In a preferred implementation of the present invention, in S2, the specific calculation process of the recommendation score is: ; Among them, ki is the number of times the i-th pending music is played, K is the number of times all music is played during the monitoring period, L and S are preset coefficients, and L is greater than S, and Ai is the total number of interactions of the i-th pending music.

[0020] In another preferred implementation of the present invention, in S3, the target music corresponding tags include beat, emotion, singer name, genre and language.

[0021] In another preferred implementation of the present invention, in S3, the calculation process of the interest degree is: Select any observed music, mark the interest of any corresponding label of the observed music as the interest of the observed music, obtain the corresponding label of the target music and calculate the interest of the corresponding label of the target music. The specific calculation formula is: ; Pm represents the interest of the target music corresponding to label m, i represents the played music corresponding to label m, Q represents the total amount of observed music corresponding to label m, Xi represents the interest of observed music i, m∈n, and n is the total number of corresponding labels.

[0022] In another preferred implementation of the present invention, in S3, the specific process of extracting spectrum features is: Get any music, divide it into frames according to the preset frame length, and get the audio frame sequence r1, r2, ..., rt, t is a positive integer, use Mel frequency cepstrum transform technology to generate the spectrum corresponding to each audio frame, select the frequency corresponding to the maximum peak in each spectrum, and get the frequency sequence p1, p2, p3, ..., pt, and calculate the variance of the frequency sequence through the variance formula; select all the amplitude values ​​of the spectrum corresponding to any audio frame, and calculate the power sum = (amplitude value) through the formula 2 , calculate the power sum corresponding to any audio frame, obtain the power sum sequence w1, w2, w3, ..., wt, calculate the variance logarithm of the power sum sequence; obtain the Mel spectrum of the sub-singing segment and input it into the preset neural network to obtain the eigenvalue of the target music timbre vector.

[0023] In another preferred implementation of the present invention, the specific calculation process of the similarity distance value is: ; Among them, X is the variance of the target music frequency sequence, Y is the logarithm of the variance of the power and sequence of the target music, Z is the eigenvalue of the target music timbre vector, X' is the variance of the frequency sequence of the music to be recommended, Y' is the logarithm of the variance of the power and sequence of the music to be recommended, and Z' is the eigenvalue of the timbre vector of the music to be recommended.

[0024] In another preferred implementation of the present invention, in S4, the specific process of correction is: Get the labels corresponding to the music to be recommended and the target music, and then get all the overlapping labels and mark them as y'1, y'2, ..., y'N, get the interest corresponding to all the overlapping labels and mark them as P1, P2, ..., PN, according to the calculation formula The corrected similarity distance value C' is calculated, where α is a preset correction coefficient and pi is the interest level of any overlapping label.

[0025] In another preferred implementation of the present invention, if there are several target music, obtaining any music to be recommended and calculating the similarity distance with the target music, sorting all the music to be recommended in ascending order according to the total similarity distance, and recommending music to the target object according to the sorting result; The calculation process of the total similarity distance is: ; Among them, C' i is the modified similarity distance value between the music to be recommended and any target music, Wi is the recommendation score of the ith target music, and h is the number of target music.

[0026] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A music intelligent playing method based on a music integration platform, characterized in that: The following steps are involved: S1, setting a monitoring period, obtaining the playback data of the target object within the monitoring period, wherein the playback data includes the played music, the number of plays and the interactive behavior; the interactive behavior includes like, share and download; S2, taking the music played with interactive behavior as pending music, selecting any pending music and obtaining the number of times the pending music is played during the monitoring period, calculating the recommendation score of the pending music according to the number of times it is played, and if the recommendation score of the pending music is greater than or equal to a preset threshold, marking the pending music as target music; S3, obtain the recommendation scores of all played music and sum them up to obtain the total recommendation score W', take the played music without interactive behavior as the observed music, calculate the ratio of the recommendation score of any observed music to the total recommendation score, and mark the ratio as the interest X of the observed music, obtain the corresponding label of the target music and mark it as y1, y2, ..., yn, calculate the interest of the corresponding label of the target music in turn according to the interest of the observed music and mark it as P1, P2, ..., Pn; Obtain music that the user has not played and mark it as music to be recommended, obtain the frequency spectrum features of the target music and all the music to be recommended, calculate the similarity distance value between any music to be recommended and the target music according to the frequency spectrum features and mark it as C1, C2, ..., Cm; S4, obtain the label corresponding to any music to be recommended, compare the label corresponding to the music to be recommended with the label corresponding to the target music, obtain all overlapping labels, obtain the interest levels corresponding to all overlapping labels and sum them up, use the sum value as the interest level of the music to be recommended, correct the similarity distance value of the music to be recommended according to the interest level, sort all the music to be recommended in an ascending order according to the corrected similarity distance value, and recommend music to the target object according to the sorting result.

2. The music intelligent playing method based on the music integrated platform according to claim 1 is characterized in that: In S2, the specific calculation process of the recommendation score is: ； Among them, ki is the number of times the i-th pending music is played, K is the number of times all music is played during the monitoring period, L and S are preset coefficients, and L is greater than S, and Ai is the total number of interactions of the i-th pending music.

3. The music intelligent playing method based on the music integrated platform according to claim 1 is characterized in that: In S3, the target music corresponding labels include beat, emotion, singer name, genre and language.

4. The music intelligent playing method based on the music integrated platform according to claim 1 is characterized in that: In S3, the calculation process of the interest degree is: Select any observed music, mark the interest of any corresponding label of the observed music as the interest of the observed music, obtain the corresponding label of the target music and calculate the interest of the corresponding label of the target music. The specific calculation formula is: ； Pm represents the interest of the target music corresponding to label m, i represents the observed music corresponding to label m, Q represents the total amount of observed music corresponding to label m, Xi represents the interest of observed music i, m∈n, and n is the total number of corresponding labels.

5. The music intelligent playing method based on the music integrated platform according to claim 1 is characterized in that: In S3, the specific process of spectrum feature extraction is: Get any music, divide it into frames according to the preset frame length, and get the audio frame sequence r1, r2, ..., rt, t is a positive integer, use Mel frequency cepstrum transform technology to generate the spectrum corresponding to each audio frame, select the frequency corresponding to the maximum peak in each spectrum, and get the frequency sequence p1, p2, p3, ..., pt, and calculate the variance of the frequency sequence through the variance formula; select all the amplitude values ​​of the spectrum corresponding to any audio frame, and calculate the power sum = (amplitude value) through the formula 2 , calculate the power sum corresponding to any audio frame, obtain the power sum sequence w1,w2,w3,...,wt, calculate the variance logarithm of the power sum sequence; obtain the Mel spectrum of the sub-singing segment and input it into the preset neural network to obtain the eigenvalue of the target music timbre vector.

6. The music intelligent playing method based on the music integrated platform according to claim 1 is characterized in that: The specific calculation process of the similarity distance value is: ； Among them, X is the variance of the target music frequency sequence, Y is the logarithm of the variance of the power and sequence of the target music, Z is the eigenvalue of the target music timbre vector, X' is the variance of the frequency sequence of the music to be recommended, Y' is the logarithm of the variance of the power and sequence of the music to be recommended, and Z' is the eigenvalue of the timbre vector of the music to be recommended.

7. The music intelligent playing method based on the music integrated platform according to claim 1 is characterized in that: In S4, the specific process of correction is: Get the labels corresponding to the music to be recommended and the target music, and then get all the overlapping labels and mark them as y'1, y'2, ..., y'N, get the interest corresponding to all the overlapping labels and mark them as P1, P2, ..., PN, according to the calculation formula The corrected similarity distance value C' is calculated, where α is a preset correction coefficient and pi is the interest level of any overlapping label.

8. The music intelligent playing method based on the music integrated platform according to claim 1 is characterized in that: It also includes if there are several target music, obtaining any music to be recommended and calculating the similarity distance with the target music, sorting all the music to be recommended in ascending order according to the total similarity distance, and recommending music to the target object according to the sorting result; The calculation process of the total similarity distance is: ； Among them, C' i is the modified similarity distance value between the music to be recommended and any target music, Wi is the recommendation score of the ith target music, and h is the number of target music.

Images