Audio search sorting adjustment method and device and medium

Abstract

The invention discloses an audio search sorting adjustment method and device and a medium, and relates to the technical field of audio search, and the method comprises the steps: obtaining at least one piece of to-be-adjusted audio data under a first search word, and extracting the audio features of the to-be-adjusted audio data; the audio features comprise Mel spectrum features and general audio features; inputting the audio features of the to-be-adjusted audio data into a prediction model so as to obtain prediction behavior data of the to-be-adjusted audio data through the prediction model; the predicted behavior data comprises a predicted playing completion rate, a predicted collection rate or a weighted average result of the predicted playing completion rate and the predicted collection rate; and determining a current sequence of at least one piece of to-be-adjusted audio data under the first search word, and adjusting the current sequence based on the predicted behavior data of the to-be-adjusted audio data. According to the method and the device, confusion caused by the same-name songs is effectively avoided, the quality of the songs and the preference of the user are comprehensively considered, and more accurate and personalized search is realized.

Description

Technical Field

[0001] The present invention relates to the field of audio search technology, and in particular to an audio search sort adjustment method, device and medium. Background Art

[0002] Existing audio search systems often focus on recommending popular music, ignoring less popular, high-quality, and unpopular songs or newly released high-quality music. This biased search result not only limits the user's choice range, but also reduces the diversity of music exploration. In addition, existing audio search systems cannot effectively distinguish between new songs, cover songs, or unpopular songs with the same name, and may ignore the overall quality of the song when searching.

[0003] Therefore, the above technical problems need to be solved by those skilled in the art urgently. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide an audio search ranking adjustment method, device and medium, which effectively avoids the confusion caused by songs with the same name, comprehensively considers the quality of the songs and the preferences of the users, and realizes a more accurate and personalized search. The specific scheme is as follows:

[0005] In a first aspect, the present application discloses a method for adjusting audio search ranking, comprising:

[0006] Acquire at least one audio data to be adjusted under the first search term, and extract audio features of the audio data to be adjusted; the audio features include Mel spectrum features and general audio representations;

[0007] Inputting the audio features of the audio data to be adjusted into a prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model; the predicted behavior data includes a predicted completion rate, a predicted collection rate, or a weighted average result of the predicted completion rate and the predicted collection rate;

[0008] Determine a current ranking of at least one of the audio data to be adjusted under the first search term, and adjust the current ranking based on the predicted behavior data of the audio data to be adjusted.

[0009] Optionally, before inputting the audio features of the audio data to be adjusted into the prediction model, the method further includes:

[0010] Acquire at least one original sample audio data under a second search term;

[0011] The original sample audio data whose click frequency is not greater than a preset click frequency threshold or whose play times are not greater than a preset play times threshold are eliminated to obtain standard sample audio data;

[0012] Acquire actual behavior data of the standard sample audio data; the actual behavior data includes a historical completion rate, a historical collection rate, and a weighted average result of the historical completion rate and the historical collection rate;

[0013] The prediction model is obtained by training actual behavior data based on standard sample audio data.

[0014] Optionally, the prediction model is obtained by training the actual behavior data based on the standard sample audio data, including:

[0015] Sorting the actual behavior data of the standard sample audio data in descending order to obtain a first sorting result;

[0016] Determine the first N standard sample audio data in the first sorting result as positive samples, and determine the last N standard sample audio data as negative samples; wherein N is a positive integer;

[0017] A positive sample pair is constructed according to any two of the positive samples, a negative sample pair is constructed according to the positive sample and the negative sample, the audio features of the positive sample pair or the negative sample pair are input into a first model to be trained, so as to obtain a multidimensional vector through the first model to be trained, and a first feature distance between the positive sample and a reference sample or a second feature distance between the negative sample and the reference sample is calculated according to the multidimensional vector; wherein the first model to be trained is a dual-tower network structure;

[0018] A first loss is calculated according to the first feature distance or the second feature distance and a first loss function until a first prediction model that meets the loss condition is obtained.

[0019] Optionally, inputting the audio features of the audio data to be adjusted into a prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model includes:

[0020] Inputting the audio features of the audio data to be adjusted into the first prediction model to obtain the multidimensional vector through the first prediction model, and calculating the corresponding feature distance between the audio data to be adjusted and the reference sample according to the multidimensional vector;

[0021] Accordingly, adjusting the current sorting based on the predicted behavior data of the audio data to be adjusted includes:

[0022] The current sorting is adjusted based on corresponding feature distances between the audio data to be adjusted and the reference samples.

[0023] Optionally, the prediction model is obtained by training the actual behavior data based on the standard sample audio data, including:

[0024] Sorting the actual behavior data of the standard sample audio data in descending order to obtain a second sorting result;

[0025] Inputting the audio features of the standard sample audio data into a second model to be trained, so as to obtain a first prediction ranking result of the standard sample audio data through the second model to be trained;

[0026] The second loss is calculated according to the second sorting result, the first predicted sorting result and a second loss function until a second prediction model that meets the loss condition is obtained.

[0027] Optionally, inputting the audio features of the audio data to be adjusted into a prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model includes:

[0028] Inputting the audio features of the audio data to be adjusted into the second prediction model to obtain a second prediction ranking result of the audio data to be adjusted through the second prediction model;

[0029] Accordingly, adjusting the current sorting based on the predicted behavior data of the audio data to be adjusted includes:

[0030] Determine a first position of the audio data to be adjusted in the current sorting, and determine a second position of the audio data to be adjusted in the second predicted sorting result;

[0031] The current sorting is adjusted according to the first position and the second position.

[0032] Optionally, the prediction model is obtained by training the actual behavior data based on the standard sample audio data, including:

[0033] Inputting the audio features of the standard sample audio data into a third model to be trained, so as to obtain the predicted behavior data of the standard sample audio data through the third model to be trained;

[0034] The third loss is calculated based on the predicted behavior data of the standard sample audio data, the actual behavior data of the standard sample audio data and the third loss function until a third prediction model that meets the loss condition is obtained.

[0035] Optionally, inputting the audio features of the audio data to be adjusted into a prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model includes:

[0036] Inputting the audio features of the audio data to be adjusted into the third prediction model to obtain the predicted behavior data of the audio data to be adjusted through the third prediction model;

[0037] Accordingly, adjusting the current sorting based on the predicted behavior data of the audio data to be adjusted includes:

[0038] Determine target behavior data threshold;

[0039] The current ranking is adjusted according to the magnitude relationship between the predicted behavior data of the audio data to be adjusted and the target behavior data threshold.

[0040] In a second aspect, the present application discloses an audio search ranking adjustment device, comprising:

[0041] An audio feature extraction module, used to obtain at least one audio data to be adjusted under the first search term, and extract audio features of the audio data to be adjusted; the audio features include Mel spectrum features and general audio representations;

[0042] A predicted behavior data acquisition module, used for inputting the audio features of the audio data to be adjusted into a prediction model, so as to obtain predicted behavior data of the audio data to be adjusted through the prediction model; the predicted behavior data includes a predicted completion rate, a predicted collection rate, or a weighted average result of the predicted completion rate and the predicted collection rate;

[0043] The audio search ranking adjustment module is used to determine the current ranking of at least one of the audio data to be adjusted under the first search term, and adjust the current ranking based on the predicted behavior data of the audio data to be adjusted.

[0044] In a third aspect, the present application discloses an electronic device, comprising:

[0045] Memory, used to store computer programs;

[0046] A processor is used to execute the computer program to implement the audio search ranking adjustment method disclosed above.

[0047] In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the aforementioned audio search ranking adjustment method disclosed above is implemented.

[0048] In a fifth aspect, the present application discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the audio search ranking adjustment method described in any one of claims 1 to 8.

[0049] It can be seen that the present application proposes an audio search ranking adjustment method, including: obtaining at least one audio data to be adjusted under a first search term, and extracting audio features of the audio data to be adjusted; the audio features include Mel spectrum features and general audio representations; inputting the audio features of the audio data to be adjusted into a prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model; the predicted behavior data includes a predicted completion rate, a predicted collection rate, or a weighted average of the predicted completion rate and the predicted collection rate; determining the current ranking of at least one of the audio data to be adjusted under the first search term, and adjusting the current ranking based on the predicted behavior data of the audio data to be adjusted. In summary, the present application obtains the audio features of the audio data to be adjusted, and then obtains the predicted behavior data of the audio data to be adjusted according to the prediction model, and adjusts the current ranking of at least one audio data to be adjusted under the first search term according to the predicted behavior data. Since the predicted behavior data includes the predicted completion rate, the predicted collection rate, or the weighted average of the predicted completion rate and the predicted collection rate, the present application comprehensively considers the quality of the song and the user's preferences to achieve a more accurate and personalized search. For less popular music or high-quality new songs that lack behavioral data, this search strategy can more fairly promote diverse music content, which not only increases the user's chances of discovering new music, but also brings wider user satisfaction and engagement to the music platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0051] Figure 1 A flow chart of an audio search ranking adjustment method disclosed in this application;

[0052] Figure 2 A schematic diagram of audio search sorting disclosed in this application;

[0053] Figure 3 A flowchart of a specific audio search ranking adjustment method disclosed in this application;

[0054] Figure 4 A flowchart of a specific audio search ranking adjustment method disclosed in this application;

[0055] Figure 5 A flowchart of a specific audio search ranking adjustment method disclosed in this application;

[0056] Figure 6 This is a schematic diagram of the structure of an audio search ranking adjustment device disclosed in this application;

[0057] Figure 7 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0058] 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.

[0059] Existing audio search systems often focus on recommending popular music, ignoring less popular, high-quality, and unpopular songs or newly released high-quality music. This biased search result not only limits the user's choice range, but also reduces the diversity of music exploration. In addition, existing audio search systems cannot effectively distinguish between new songs, cover songs, or unpopular songs with the same name, and may ignore the overall quality of the song when searching.

[0060] To this end, an embodiment of the present application proposes an audio search ranking adjustment scheme to avoid confusion caused by songs with the same name, taking into account the quality of the songs and the user's preferences, and achieving a more accurate and personalized search.

[0061] The present application embodiment discloses a method for adjusting the audio search ranking, see Figure 1 As shown, including:

[0062] Step S11: obtaining at least one audio data to be adjusted under the first search term, and extracting audio features of the audio data to be adjusted; the audio features include Mel-spectrogram features and general audio representations.

[0063] In this embodiment, the audio search sorting adjustment method can be implemented by a terminal device, such as a computer and a mobile device. Specifically, in the implementation process based on a computer, the MP3Helper software is used to sort the MP3 / wma files by clicking the relevant button. In the implementation process based on a terminal device, the music player application on the mobile phone or tablet performs the corresponding sorting. Furthermore, the audio search sorting adjustment method can also be implemented by a server, such as a music platform server, a search engine server, etc.

[0064] In this embodiment, the first search term may be a song title. After entering the song title in the input box, identification information of at least one audio will be obtained. The identification information represents the difference between audios with the same name, such as the singing version information and singer information of the audios with the same name. The singing version information includes recorded version, live version, chorus version, English version, etc. For example, see Figure 2 As shown, after entering the title of the target song in the input box, you will get search results for multiple different versions sung by different singers.

[0065] In this embodiment, after the audio data to be adjusted is obtained, the audio features of the audio data to be adjusted are extracted, wherein the audio features include Mel spectrum features and general audio representations. The Mel spectrum features are a time-frequency domain representation of the audio, covering information such as pitch and rhythm in music, and the general audio representation includes multi-dimensional information such as the audio melody, genre, sound quality, and labels.

[0066] Step S12: Input the audio features of the audio data to be adjusted into the prediction model to obtain the predicted behavior data of the audio data to be adjusted through the prediction model; the predicted behavior data includes the predicted completion rate, the predicted collection rate or the weighted average result of the predicted completion rate and the predicted collection rate.

[0067] At least one original sample audio data under the second search term is obtained, and the original sample audio data whose click frequency is not greater than the preset click frequency threshold or the number of plays is not greater than the preset number of plays threshold is eliminated to obtain standard sample audio data. This embodiment retains the actual behavior data whose click frequency is greater than the preset click frequency threshold, which can reduce the statistical deviation of the behavior data. This embodiment retains the actual behavior data whose number of plays is greater than the preset number of plays threshold, which helps to filter out noise in the search results.

[0068] Furthermore, actual behavior data of the standard sample audio data is obtained, and a prediction model is trained based on the actual behavior data of the standard sample audio data. The actual behavior data includes historical completion rate, historical collection rate, and weighted average of historical completion rate and historical collection rate. The completion rate is the ratio of users who click to play a song and listen to the end after seeing the search results. The collection rate is the ratio of users who collect the song after seeing the search results. The actual behavior data represents the user's actual satisfaction with the audio, and the predicted behavior data represents the user's predicted satisfaction with the audio.

[0069] After the prediction model is obtained, the audio features of the audio data to be adjusted are input into the prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model.

[0070] Step S13: Determine a current ranking of at least one of the audio data to be adjusted under the first search term, and adjust the current ranking based on the predicted behavior data of the audio data to be adjusted.

[0071] In this embodiment, after the predicted behavior data is obtained, the current ranking of at least one audio data to be adjusted under the first search term is determined, and the current ranking is adjusted based on the predicted behavior data of the audio data to be adjusted. Specifically, depending on the search scenario, this embodiment can directly adjust the current ranking of the audio data to be adjusted based on the predicted behavior data of the audio data to be adjusted, or can adjust the current ranking of the audio data to be adjusted based on the relationship between the predicted behavior data of the audio data to be adjusted and the actual behavior data. In subsequent embodiments, it will be explained how to design different prediction models for search scenarios, how to use corresponding prediction models to obtain predicted behavior data, and how to use the predicted behavior data to adjust the current ranking of the audio data to be adjusted.

[0072] It can be seen that the present application proposes an audio search ranking adjustment method, including: obtaining at least one audio data to be adjusted under a first search term, and extracting audio features of the audio data to be adjusted; the audio features include Mel spectrum features and general audio representations; inputting the audio features of the audio data to be adjusted into a prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model; the predicted behavior data includes a predicted completion rate, a predicted collection rate, or a weighted average of the predicted completion rate and the predicted collection rate; determining the current ranking of at least one of the audio data to be adjusted under the first search term, and adjusting the current ranking based on the predicted behavior data of the audio data to be adjusted. In summary, the present application obtains the audio features of the audio data to be adjusted, and then obtains the predicted behavior data of the audio data to be adjusted according to the prediction model, and adjusts the current ranking of at least one audio data to be adjusted under the first search term according to the predicted behavior data. Since the predicted behavior data includes the predicted completion rate, the predicted collection rate, or the weighted average of the predicted completion rate and the predicted collection rate, the present application comprehensively considers the quality of the song and the user's preferences to achieve a more accurate and personalized search. For less popular music or high-quality new songs that lack behavioral data, this search strategy can more fairly promote diverse music content, which not only increases the user's chances of discovering new music, but also brings wider user satisfaction and engagement to the music platform.

[0073] The present application embodiment discloses a specific audio search ranking adjustment method. Compared with the previous embodiment, this embodiment further explains and optimizes the technical solution. Figure 3 As shown, including:

[0074] Step S21: sorting the actual behavior data of the standard sample audio data in descending order to obtain a first sorting result.

[0075] Step S22: determine the first N standard sample audio data in the first sorting result as positive samples, and determine the last N standard sample audio data as negative samples; wherein N is a positive integer.

[0076] Step S23: construct a positive sample pair based on any two of the positive samples, construct a negative sample pair based on the positive sample and the negative sample, input the audio features of the positive sample pair or the negative sample pair into the first model to be trained to obtain a multidimensional vector through the first model to be trained, and calculate the first feature distance between the positive sample and the reference sample or the second feature distance between the negative sample and the reference sample based on the multidimensional vector; wherein, the first model to be trained is a dual-tower network structure.

[0077] Step S24: Calculate a first loss according to the first feature distance or the second feature distance and a first loss function until a first prediction model that meets the loss condition is obtained.

[0078] This embodiment trains a first prediction model, and the first prediction model predicts which audio is more likely to be preferred by the user based on the feature distance between the two audios. The training process of the first prediction model is introduced below: the actual behavior data of the standard sample audio data is sorted in descending order to obtain a first sorting result. Taking the actual behavior data as the completion rate as an example, the first N standard sample audio data in the first sorting result are samples with high completion rates, also called positive samples, and the last N standard sample audio data in the first sorting result are samples with low completion rates, also called negative samples. In this embodiment, any two positive samples are selected to form a positive sample pair, and any two negative samples are selected to form a negative sample pair, that is, under the same search term, <songs with high completion rates and songs with high completion rates> are positive sample pairs, and <songs with high completion rates and songs with low completion rates> are negative sample pairs. Further, the audio features of the positive sample pair or the negative sample pair are input into the first model to be trained to obtain a multidimensional vector through the first model to be trained, and the first feature distance between the positive sample and the reference sample, or the second feature distance between the negative sample and the reference sample is calculated based on the multidimensional vector, wherein the first model to be trained is a dual-tower network structure based on a convolutional network or a residual network, each tower processes an audio feature respectively, and shares weights to extract high-level representations, generates a multidimensional vector of fixed dimension through a fully connected layer, and calculates the first feature distance between the positive sample and the reference sample, or the second feature distance between the negative sample and the reference sample based on the multidimensional vector, and the feature distance includes Euclidean distance or cosine similarity. In order to optimize the model, this embodiment uses a first loss function to calculate the first loss until a first prediction model in which the first loss satisfies a preset loss condition is obtained, and the first loss function is: ; d(x,y) represents the characteristic distance between the x sample and the y sample, 𝑎 represents the reference sample, 𝑝 represents the positive sample, 𝑛 represents the negative sample, and margin represents the hyperparameter that controls the distance difference between positive and negative samples.

[0079] After obtaining the first prediction model, this embodiment inputs the audio features of the audio data to be adjusted into the first prediction model to obtain a multidimensional vector through the first prediction model, and calculates the corresponding feature distance between the audio data to be adjusted and the reference sample according to the multidimensional vector, and then adjusts the current sorting based on the corresponding feature distance between the audio data to be adjusted and the reference sample. Exemplarily, the reference sample is a sample with relatively accurate sorting, and the feature distances between multiple audio data to be adjusted and the reference sample are compared, and the audio data to be adjusted with a smaller distance is pulled up in the sorting position, and the audio data to be adjusted with a larger distance is lowered in the sorting position.

[0080] The embodiment of the present application discloses a specific method for adjusting the audio search ranking. Compared with the previous embodiment, this embodiment further explains and optimizes the technical solution. Figure 4 As shown, including:

[0081] Step S31: sorting the actual behavior data of the standard sample audio data in descending order to obtain a second sorting result.

[0082] Step S32: input the audio features of the standard sample audio data into the second model to be trained, so as to obtain a first predicted sorting result of the standard sample audio data through the second model to be trained.

[0083] Step S33: Calculate the second loss according to the second sorting result, the first prediction sorting result and the second loss function until a second prediction model that meets the loss condition is obtained.

[0084] This embodiment trains a second prediction model, which predicts the arrangement order of a group of audios. The training process of the second prediction model is described below: the actual behavior data of the standard sample audio data is sorted in descending order to obtain a second sorting result. Taking the actual behavior data as the completion rate as an example, the second sorting result can be<id1,id2,id3…idN> . Furthermore, the audio feature pairs of the standard sample audio data are input into the second model to be trained to obtain the first predicted sorting result of the standard sample audio data through the second model to be trained. The second model to be trained is a network structure based on a convolutional network or a residual network, which is used to extract local and deep features of the audio. Subsequently, these features generate feature vectors of fixed dimensions through several fully connected layers, and finally output the sorting result of each audio to indicate its relative position in the sorting. In order to optimize the model, a second loss function is used to calculate the second loss until a second prediction model is obtained in which the second loss satisfies the preset loss condition, wherein the second loss function includes: L=∑ i<j max(0,1-(s yi -s yj ));in, and Respectively represent the second to-be-trained model for the standard sample audio data y i and standard sample audio data y j From the prediction sorting results, we can see that the second loss function ensures that the audio with a high completion rate has a higher score than the audio with a low completion rate by minimizing the difference between the predicted order and the actual order, thereby learning a more reasonable order of the audio completion rate.

[0085] After obtaining the second prediction model, this embodiment inputs the audio features of the audio data to be adjusted into the second prediction model to obtain the second prediction ranking result of the audio data to be adjusted through the second prediction model, further determines the first position of the audio data to be adjusted in the current ranking, and determines the second position of the audio data to be adjusted in the second prediction ranking result, and adjusts the current ranking according to the first position and the second position. Exemplarily, if the first position of a new song (new song or cold song) in the current ranking is the last 20%, and the second position in the second prediction ranking result is the first 20%, the song is pulled up several positions.

[0086] The embodiment of the present application discloses a specific method for adjusting the audio search ranking. Compared with the previous embodiment, this embodiment further explains and optimizes the technical solution. Figure 5 As shown, including:

[0087] Step S41: input the audio features of the standard sample audio data into the third model to be trained, so as to obtain the predicted behavior data of the standard sample audio data through the third model to be trained.

[0088] Step S42: Calculate the third loss based on the predicted behavior data of the standard sample audio data, the actual behavior data of the standard sample audio data and the third loss function until a third prediction model that meets the loss condition is obtained.

[0089] This embodiment trains a third prediction model, and the third prediction model predicts the absolute arrangement order of a single audio. The training process of the third prediction model is introduced below: the audio features of the standard sample audio data are input into the third model to be trained, so as to obtain the predicted behavior data of the standard sample audio data through the third model to be trained, and the third loss is calculated according to the predicted behavior data of the standard sample audio data, the actual behavior data of the standard sample audio data and the third loss function, until the third prediction model that meets the loss condition is obtained. The third model to be trained is a network structure based on a convolutional network or a residual network, which is used to extract local and deep features of the audio. Subsequently, these features generate feature vectors of fixed dimensions through several fully connected layers, and finally output the predicted behavior data of each audio. In order to optimize the model, the third loss is calculated using the third loss function until a third prediction model whose third loss meets the preset loss condition is obtained, wherein the third loss function includes: ,in, Represents actual behavior data, represents the predicted behavior data of the standard sample audio data, and N represents the number of the standard sample audio data. It can be seen that this embodiment trains the model by minimizing the difference between the predicted value and the actual completion rate.

[0090] After obtaining the third prediction model, the audio features of the audio data to be adjusted are input into the third prediction model to obtain the predicted behavior data of the audio data to be adjusted through the third prediction model. Further, the target behavior data threshold is determined, and the current order of the audio data to be adjusted is adjusted according to the relationship between the predicted behavior data of the audio data to be adjusted and the target behavior data threshold. Exemplarily, if the predicted completion rate is > 0.4, then one position is pulled up for every 0.05, and if the completion rate is < 0.2, then one position is pulled down for every 0.05. It can be understood that the above values ​​are empirical values ​​and can be adjusted.

[0091] This application proposes a new search strategy that uses the predicted behavior data obtained from audio features to guide the search algorithm, optimize the sorting mechanism, avoid confusion caused by songs with the same name, and comprehensively consider the quality of the songs and user preferences. For less popular music or high-quality new songs that lack behavior data, this search strategy can more fairly promote diverse music content, which not only increases users' chances of discovering new music, but also brings wider user satisfaction and engagement to the music platform.

[0092] Correspondingly, the embodiment of the present application also discloses an audio search ranking adjustment device, see Figure 6 As shown, the device comprises:

[0093] The audio feature extraction module 11 is used to obtain at least one audio data to be adjusted under the first search term, and extract the audio features of the audio data to be adjusted; the audio features include Mel spectrum features and general audio representations;

[0094] The predicted behavior data acquisition module 12 is used to input the audio features of the audio data to be adjusted into the prediction model to obtain the predicted behavior data of the audio data to be adjusted through the prediction model; the predicted behavior data includes the predicted completion rate, the predicted collection rate or the weighted average result of the predicted completion rate and the predicted collection rate;

[0095] The audio search ranking adjustment module 13 is used to determine the current ranking of at least one of the audio data to be adjusted under the first search term, and adjust the current ranking based on the predicted behavior data of the audio data to be adjusted.

[0096] Among them, for more specific working processes of the above-mentioned modules, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.

[0097] It can be seen that the present application proposes an audio search ranking adjustment method, including: obtaining at least one audio data to be adjusted under a first search term, and extracting audio features of the audio data to be adjusted; the audio features include Mel spectrum features and general audio representations; inputting the audio features of the audio data to be adjusted into a prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model; the predicted behavior data includes a predicted completion rate, a predicted collection rate, or a weighted average of the predicted completion rate and the predicted collection rate; determining the current ranking of at least one of the audio data to be adjusted under the first search term, and adjusting the current ranking based on the predicted behavior data of the audio data to be adjusted. In summary, the present application obtains the audio features of the audio data to be adjusted, and then obtains the predicted behavior data of the audio data to be adjusted according to the prediction model, and adjusts the current ranking of at least one audio data to be adjusted under the first search term according to the predicted behavior data. Since the predicted behavior data includes the predicted completion rate, the predicted collection rate, or the weighted average of the predicted completion rate and the predicted collection rate, the present application comprehensively considers the quality of the song and the user's preferences to achieve a more accurate and personalized search. For less popular music or high-quality new songs that lack behavioral data, this search strategy can more fairly promote diverse music content, which not only increases the user's chances of discovering new music, but also brings wider user satisfaction and engagement to the music platform.

[0098] Furthermore, an embodiment of the present application also provides an electronic device. Figure 7 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram cannot be regarded as any limitation on the scope of use of the present application.

[0099] Figure 7 The present invention provides a schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a display screen 23, an input / output interface 24, a communication interface 25, a power supply 26, and a communication bus 27. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the audio search ranking adjustment method disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0100] In this embodiment, the power supply 26 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 25 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 24 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0101] In addition, the memory 22, as a carrier for storing resources, may be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon may include a computer program 221, and the storage method may be temporary storage or permanent storage. In addition to including a computer program that can be used to complete the audio search ranking adjustment method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 221 may further include a computer program that can be used to complete other specific tasks.

[0102] Furthermore, an embodiment of the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the aforementioned audio search ranking adjustment method disclosed above is implemented.

[0103] For the specific steps of the method, reference may be made to the corresponding contents disclosed in the aforementioned embodiments, which will not be described in detail here.

[0104] The various embodiments in this application are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0105] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0106] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0107] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0108] The above is a detailed introduction to an audio search ranking adjustment method, device, and storage medium provided by the present application. This article uses specific examples to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, based on the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for adjusting audio search ranking, characterized in that: include: Acquire at least one audio data to be adjusted under the first search term, and extract audio features of the audio data to be adjusted; the audio features include Mel spectrum features and general audio representations; Inputting the audio features of the audio data to be adjusted into a prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model; the predicted behavior data includes a predicted completion rate, a predicted collection rate, or a weighted average result of the predicted completion rate and the predicted collection rate; Determine a current ranking of at least one of the audio data to be adjusted under the first search term, and adjust the current ranking based on the predicted behavior data of the audio data to be adjusted.

2. The audio search ranking adjustment method according to claim 1, characterized in that: Before inputting the audio features of the audio data to be adjusted into the prediction model, the method further includes: Acquire at least one original sample audio data under a second search term; The original sample audio data whose click frequency is not greater than a preset click frequency threshold or whose play times are not greater than a preset play times threshold are eliminated to obtain standard sample audio data; Acquire actual behavior data of the standard sample audio data; the actual behavior data includes a historical completion rate, a historical collection rate, and a weighted average result of the historical completion rate and the historical collection rate; The prediction model is obtained by training actual behavior data based on standard sample audio data.

3. The audio search ranking adjustment method according to claim 2, characterized in that: The prediction model is obtained by training the actual behavior data based on the standard sample audio data, including: Sorting the actual behavior data of the standard sample audio data in descending order to obtain a first sorting result; Determine the first N standard sample audio data in the first sorting result as positive samples, and determine the last N standard sample audio data as negative samples; wherein N is a positive integer; A positive sample pair is constructed according to any two of the positive samples, a negative sample pair is constructed according to the positive sample and the negative sample, the audio features of the positive sample pair or the negative sample pair are input into a first model to be trained, so as to obtain a multidimensional vector through the first model to be trained, and a first feature distance between the positive sample and a reference sample or a second feature distance between the negative sample and the reference sample is calculated according to the multidimensional vector; wherein the first model to be trained is a dual-tower network structure; A first loss is calculated according to the first feature distance or the second feature distance and a first loss function until a first prediction model that meets the loss condition is obtained.

4. The audio search ranking adjustment method according to claim 3, characterized in that: The step of inputting the audio features of the audio data to be adjusted into a prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model includes: Inputting the audio features of the audio data to be adjusted into the first prediction model to obtain the multidimensional vector through the first prediction model, and calculating the corresponding feature distance between the audio data to be adjusted and the reference sample according to the multidimensional vector; Accordingly, adjusting the current sorting based on the predicted behavior data of the audio data to be adjusted includes: The current sorting is adjusted based on corresponding feature distances between the audio data to be adjusted and the reference samples.

5. The audio search ranking adjustment method according to claim 2, characterized in that: The prediction model is obtained by training the actual behavior data based on the standard sample audio data, including: Sorting the actual behavior data of the standard sample audio data in descending order to obtain a second sorting result; Inputting the audio features of the standard sample audio data into a second model to be trained, so as to obtain a first prediction ranking result of the standard sample audio data through the second model to be trained; The second loss is calculated according to the second sorting result, the first predicted sorting result and a second loss function until a second prediction model that meets the loss condition is obtained.

6. The audio search ranking adjustment method according to claim 5, characterized in that: The step of inputting the audio features of the audio data to be adjusted into a prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model includes: Inputting the audio features of the audio data to be adjusted into the second prediction model to obtain a second prediction ranking result of the audio data to be adjusted through the second prediction model; Accordingly, adjusting the current sorting based on the predicted behavior data of the audio data to be adjusted includes: Determine a first position of the audio data to be adjusted in the current sorting, and determine a second position of the audio data to be adjusted in the second predicted sorting result; The current sorting is adjusted according to the first position and the second position.

7. The audio search ranking adjustment method according to claim 2, characterized in that: The prediction model is obtained by training the actual behavior data based on the standard sample audio data, including: Inputting the audio features of the standard sample audio data into a third model to be trained, so as to obtain the predicted behavior data of the standard sample audio data through the third model to be trained; The third loss is calculated based on the predicted behavior data of the standard sample audio data, the actual behavior data of the standard sample audio data and the third loss function until a third prediction model that meets the loss condition is obtained.

8. The audio search ranking adjustment method according to claim 7, characterized in that: The step of inputting the audio features of the audio data to be adjusted into a prediction model to obtain predicted behavior data of the audio data to be adjusted through the prediction model includes: Inputting the audio features of the audio data to be adjusted into the third prediction model to obtain the predicted behavior data of the audio data to be adjusted through the third prediction model; Accordingly, adjusting the current sorting based on the predicted behavior data of the audio data to be adjusted includes: Determine target behavior data threshold; The current ranking is adjusted according to the magnitude relationship between the predicted behavior data of the audio data to be adjusted and the target behavior data threshold.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the audio search ranking adjustment method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: Used to store a computer program; wherein, when the computer program is executed by a processor, the audio search ranking adjustment method as described in any one of claims 1 to 8 is implemented.

11. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the audio search ranking adjustment method described in any one of claims 1 to 8 is implemented.

Images