Song requesting background recommendation method, device and equipment based on tablet terminal

Abstract

The invention provides a song requesting background recommendation method, device and equipment based on a tablet terminal, and relates to the technical field of data analysis. In the application, firstly, target user preference information is acquired, and historical user preference information and historical song requesting information are acquired; secondly, respectively carrying out semantic mining on the target user preference information and the historical user preference information to form target preference semantic features and historical preference semantic features; secondly, semantic mining is carried out on the historical song requesting information, and historical song requesting semantic features are formed; further, on the basis of preference fusion semantic features of the target preference semantic features and the historical preference semantic features, semantic adjustment is performed on the historical song-requesting semantic features to form adjusted song-requesting semantic features; and finally, performing semantic restoration based on the adjusted song requesting semantic features to form target recommended song information. On the basis of the content, the problem that in the prior art, the reliability of song requesting recommendation is relatively low can be solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, in particular to a song ordering background recommendation method, device and equipment based on a tablet terminal. BACKGROUND

[0002] With the rapid development of mobile Internet and smart devices, song ordering systems are increasingly popular in various occasions (such as KTV, music application software, etc.). The existing song ordering recommendation system usually recommends songs based on the basic information or simple history song ordering records of users, which leads to the problem of relatively low reliability of song ordering recommendation. SUMMARY

[0003] Therefore, the present application aims to provide a song ordering background recommendation method, device and equipment based on a tablet terminal to improve the problem of relatively low reliability of song ordering recommendation in the prior art.

[0004] To achieve the above-mentioned purpose, the present application adopts the following technical solutions: A song ordering background recommendation method based on a tablet terminal, comprising: obtaining target user preference information of a current user corresponding to a target tablet terminal, and obtaining historical user preference information and historical song ordering information, wherein the historical user preference information and the historical song ordering information correspond to a historical user corresponding to the target tablet terminal and / or a historical user corresponding to an associated tablet terminal of the target tablet terminal; respectively performing semantic mining on the target user preference information and the historical user preference information to form target preference semantic features and historical preference semantic features; performing semantic mining on the historical song ordering information to form historical song ordering semantic features; performing semantic adjustment on the historical song ordering semantic features based on preference fusion semantic features of the target preference semantic features and the historical preference semantic features to form adjusted song ordering semantic features; performing semantic restoration based on the adjusted song ordering semantic features to form target recommended song information, wherein the target recommended song information is used to display to the current user through the target tablet terminal to complete song ordering recommendation.

[0005] In the preferred selection of the present application, in the above-mentioned song ordering background recommendation method based on a tablet terminal, the step of performing semantic adjustment on the historical song ordering semantic features based on the preference fusion semantic features of the target preference semantic features and the historical preference semantic features to form adjusted song ordering semantic features comprises: performing semantic mapping on the historical preference semantic features to form historical preference mapping parameters, wherein the historical preference mapping parameters are used to represent the importance of each position; weighting mapping the historical preference semantic features based on the historical preference mapping parameters to realize semantic fusion of the target preference semantic features and the historical preference semantic features, and form preference fusion semantic features; performing semantic adjustment on the historical ordering semantic features based on the preference fusion semantic features to form adjusted ordering semantic features.

[0006] In a preferred selection of the present application, in the above-mentioned ordering background recommendation method based on a tablet terminal, the step of performing semantic adjustment on the historical ordering semantic features based on the preference fusion semantic features to form adjusted ordering semantic features comprises: In the first time step, perform semantic adjustment on the historical ordering semantic features based on the preference fusion semantic features to form intermediate adjusted semantic features in the first time step; In each of the second and subsequent time steps, perform semantic adjustment on the intermediate adjusted semantic features in the previous time step based on the preference fusion semantic features to form intermediate adjusted semantic features in the current time step; Determine the adjusted ordering semantic features based on the intermediate adjusted semantic features in the last time step.

[0007] In a preferred selection of the present application, in the above-mentioned ordering background recommendation method based on a tablet terminal, the step of performing semantic adjustment on the historical ordering semantic features based on the preference fusion semantic features to form adjusted ordering semantic features in each of the second and subsequent time steps comprises: In each of the second and subsequent time steps, after linear mapping the preference fusion semantic features and the intermediate adjusted semantic features in the previous time step, multiply the two linear mapping features obtained to form a first association parameter distribution; Obtain an association weight distribution corresponding to the current time step, and weighting the first association parameter distribution based on the association weight distribution to form a second association parameter distribution, wherein the association weight distribution is obtained by multiplying an initial weight distribution and a gating adjustment parameter determined based on the intermediate adjusted semantic features in the previous time step, and the initial weight distribution is obtained by learning sample data and corresponding labels; Perform normalization mapping on the second association parameter distribution to form a third association parameter distribution, and perform association adjustment on the linear mapping features of the intermediate adjusted semantic features in the previous time step based on the third association parameter distribution to form the intermediate adjusted semantic features in the current time step.

[0008] In the preferred selection of the present application, in the above-mentioned song ordering background recommendation method based on a tablet terminal, the step of respectively performing semantic mining on the target user preference information and the historical user preference information to form target preference semantic features and historical preference semantic features comprises: respectively performing semantic mining on user input preference information and environment information included in the target user preference information to form first preference semantic features and target environment semantic features; respectively performing semantic mining on user input preference information and environment information included in the historical user preference information to form second preference semantic features and historical environment semantic features; performing semantic fusion on the first preference semantic features and the target environment semantic features to form target preference semantic features; performing semantic fusion on the second preference semantic features and the historical environment semantic features to form historical preference semantic features.

[0009] In the preferred selection of the present application, in the above-mentioned song ordering background recommendation method based on a tablet terminal, the step of performing semantic fusion on the first preference semantic features and the target environment semantic features to form target preference semantic features comprises: in each forward diffusion stage, gradually applying a noise semantic feature to the target environment semantic feature to form a noise environment semantic feature corresponding to each forward diffusion stage; in each backward diffusion stage, based on the first preference semantic features, performing noise suppression on the noise environment semantic feature corresponding to the corresponding forward diffusion stage to form a noise suppression semantic feature corresponding to each backward diffusion stage; based on the noise suppression semantic feature corresponding to the last backward diffusion stage, determining the target preference semantic features.

[0010] In the preferred selection of the present application, in the above-mentioned song ordering background recommendation method based on a tablet terminal, the step of, in each backward diffusion stage, based on the first preference semantic features, performing noise suppression on the noise environment semantic feature corresponding to the corresponding forward diffusion stage to form a noise suppression semantic feature corresponding to each backward diffusion stage comprises: in the first backward diffusion stage, based on the first preference semantic features, performing cross-attention mining on the noise environment semantic feature corresponding to the last forward diffusion stage to achieve noise suppression, thereby forming a noise suppression semantic feature corresponding to the first backward diffusion stage; In each of the second and subsequent backward diffusion stages, based on the noise suppression semantic features corresponding to the previous backward diffusion stage, cross-attention mining is performed on the noise environment semantic features corresponding to the corresponding forward diffusion stage to achieve noise suppression, thereby forming the noise suppression semantic features corresponding to the current backward diffusion stage.

[0011] In a preferred selection of the present application, in the above-mentioned song ordering background recommendation method based on a tablet terminal, the step of performing semantic mining on the historical song ordering information to form historical song ordering semantic features comprises: For each historical song in the historical song ordering information, a plurality of convolutional processing is performed on the song spectrum graph of the historical song based on a local mask mechanism, thereby forming a plurality of song convolutional semantic features, wherein the local frequency regions masked by the local mask mechanism are at least partially different for any two convolutional processing. The plurality of song convolutional semantic features are added together, and the result of the addition operation is normalized and mapped to form the historical song ordering semantic features.

[0012] The present application also provides a song ordering background recommendation device based on a tablet terminal, comprising: A song ordering preference acquisition module is configured to acquire target user preference information of a current user corresponding to a target tablet terminal, and to acquire historical user preference information and historical song ordering information, wherein the historical user preference information and the historical song ordering information correspond to historical users of the target tablet terminal and / or historical users of associated tablet terminals of the target tablet terminal. A preference semantic mining module is configured to perform semantic mining on the target user preference information and the historical user preference information respectively, thereby forming target preference semantic features and historical preference semantic features. A song ordering semantic mining module is configured to perform semantic mining on the historical song ordering information, thereby forming historical song ordering semantic features. A semantic fusion adjustment module is configured to perform semantic adjustment on the historical song ordering semantic features based on a preference fusion semantic feature formed by the target preference semantic features and the historical preference semantic features, thereby forming adjusted song ordering semantic features. A song ordering semantic restoration module is configured to perform semantic restoration based on the adjusted song ordering semantic features, thereby forming target recommended song information, wherein the target recommended song information is used to display to the current user through the target tablet terminal to complete song ordering recommendation.

[0013] Based on the above, the present application also provides an electronic device, comprising: A memory is configured to store a computer program. A processor connected with the memory is used to execute the computer program stored in the memory to realize the above-mentioned song ordering background recommendation method based on a tablet terminal.

[0014] The song ordering background recommendation method based on a tablet terminal, the device and the equipment provided by the application firstly acquire target user preference information, and acquire historical user preference information and historical song ordering information; secondly, semantic mining is respectively performed on the target user preference information and the historical user preference information to form target preference semantic features and historical preference semantic features; then, semantic mining is performed on the historical song ordering information to form historical song ordering semantic features; further, based on preference fusion semantic features of the target preference semantic features and the historical preference semantic features, semantic adjustment is performed on the historical song ordering semantic features to form adjusted song ordering semantic features; finally, semantic restoration is performed based on the adjusted song ordering semantic features to form target recommended song information. Based on the above-mentioned content, since in the process of semantic mining and fusion, the target preference semantic features and the historical preference semantic features are firstly fused to form corresponding preference fusion semantic features, and then the historical song ordering semantic features are adjusted based on the preference fusion semantic features, the target preference semantic features and the historical preference semantic features are semantic features of the same dimension (i.e. both are representations of preferences), therefore, the accuracy of fusion is relatively higher, and the historical preference semantic features and the historical song ordering semantic features are also historical, and have relevance, therefore, the preference fusion semantic features formed by fusion are closer to the historical song ordering semantic features in the feature space, so that the reliability of semantic adjustment based thereon can be higher, therefore, the reliability of semantic restoration based on the obtained adjusted song ordering semantic features can be higher, so that reliable target recommended song information is obtained, and the problem of relatively low reliability of song ordering recommendation in the prior art is improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to make the above-mentioned purpose, features and advantages of the application more obvious and easy to understand, the following preferred embodiments are specifically described in detail below, and the accompanying drawings are referred to.

[0016] Figure 1 The structural block diagram of the electronic device provided by the embodiments of the application is shown.

[0017] Figure 2 The flowchart of the song ordering background recommendation method based on a tablet terminal provided by the embodiments of the application is shown.

[0018] Figure 3 The schematic diagram of noise suppression provided by the embodiments of the application is shown.

[0019] Figure 4 The schematic diagram of semantic mining provided by the embodiments of the application is shown.

[0020] Figure 5A block diagram of a tablet-based song ordering background recommendation device according to an embodiment of the present application is provided. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application but not all embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present application.

[0023] As Figure 1 shown, an electronic device is provided according to an embodiment of the present application. The electronic device can include a memory, a processor and a tablet-based song ordering background recommendation device.

[0024] In detail, the memory and the processor are directly or indirectly electrically connected to realize data transmission or interaction. For example, the memory and the processor can be electrically connected through one or more communication buses or signal lines. The tablet-based song ordering background recommendation device includes at least one software functional module stored in the memory in the form of software or firmware. The processor is configured to execute the executable computer programs stored in the memory, for example, the software functional modules and computer programs included in the tablet-based song ordering background recommendation device, to realize the tablet-based song ordering background recommendation method provided by the embodiments of the present application.

[0025] Optionally, the memory can be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read only memory (PROM), an erasable programmable read only memory (EPROM), an electrically erasable programmable read only memory (EEPROM) and the like.

[0026] The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc. The processor can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0027] It can be understood that Figure 1 The structure shown is only schematic, and the electronic device can further include more or fewer components than those shown in the figures, or have a different configuration from that shown in the figures, for example, it can further include a communication unit for information interaction with other devices (such as a tablet terminal, etc.). Figure 1 The structure shown is only schematic, and the electronic device can further include more or fewer components than those shown in the figures, or have a different configuration from that shown in the figures, for example, it can further include a communication unit for information interaction with other devices (such as a tablet terminal, etc.). Figure 1 The structure shown is only schematic, and the electronic device can further include more or fewer components than those shown in the figures, or have a different configuration from that shown in the figures, for example, it can further include a communication unit for information interaction with other devices (such as a tablet terminal, etc.).

[0028] In combination with Figure 2 The embodiments of the present application also provide a tablet terminal-based song ordering background recommendation method applicable to the above electronic device. The method steps defined by the flow of the tablet terminal-based song ordering background recommendation method can be implemented by the electronic device. The specific flow shown in Figure 2 will be described in detail below.

[0029] In step S110, target user preference information of a current user corresponding to a target tablet terminal is obtained, and historical user preference information and historical song ordering information are obtained.

[0030] In the embodiments of the present application, the electronic device can obtain target user preference information of a current user corresponding to the target tablet terminal, and obtain historical user preference information and historical song ordering information. The historical user preference information and the historical song ordering information correspond to a historical user corresponding to the target tablet terminal and / or a historical user corresponding to an associated tablet terminal of the target tablet terminal. That is, the historical user preference information and the historical song ordering information of the historical user corresponding to the target tablet terminal can be obtained. The historical user preference information and the historical song ordering information of the historical user corresponding to the associated tablet terminal of the target tablet terminal can also be obtained, for example, when the historical user of the target tablet terminal is less or none, the data corresponding to the associated tablet terminal can be obtained. The historical user preference information and the historical song ordering information of the historical user corresponding to the target tablet terminal and the historical user preference information and the historical song ordering information of the historical user corresponding to the associated tablet terminal of the target tablet terminal can also be obtained. It should be noted that the associated tablet terminal and the target tablet terminal can have a high song coincidence degree between the historical song ordering information, or the associated tablet terminal and the target tablet terminal can be close in the distance between the places of delivery, or the associated tablet terminal can be determined based on one or more other manners.

[0031] In step S120, the target user preference information and the historical user preference information are respectively subjected to semantic mining to form target preference semantic features and historical preference semantic features.

[0032] In the embodiments of the present application, after obtaining the target user preference information and the historical user preference information, the electronic device can respectively subject the target user preference information and the historical user preference information to semantic mining to form target preference semantic features and historical preference semantic features. That is, the target user preference information can be subjected to semantic mining to form target preference semantic features, and the historical user preference information can be subjected to semantic mining to form historical preference semantic features. It should be noted that semantic mining refers to mining potential semantic features in the corresponding information, and the potential semantic features can be represented in the form of vectors or matrices.

[0033] In step S130, the historical song ordering information is subjected to semantic mining to form historical song ordering semantic features.

[0034] In the embodiments of the present application, after obtaining the historical song ordering information, the electronic device can subject the historical song ordering information to semantic mining to form historical song ordering semantic features. That is, potential semantic features in the historical song ordering information can be mined and represented in the form of vectors or matrices.

[0035] In step S140, the historical point singing semantic feature is semantically adjusted based on the preference fusion semantic feature of the target preference semantic feature and the historical preference semantic feature, to form an adjusted point singing semantic feature.

[0036] In the embodiments of the present application, after the target preference semantic feature, the historical preference semantic feature and the historical point singing semantic feature are formed, the electronic device can perform semantic adjustment on the historical point singing semantic feature based on a preference fusion semantic feature of the target preference semantic feature and the historical preference semantic feature, to form an adjusted point singing semantic feature. That is, the target preference semantic feature and the historical preference semantic feature of the same dimension can be fused first, and the adjustment of the target semantic feature based on the historical semantic feature is realized, so as to ensure that the adaptation degree between the semantic features is higher when the historical point singing semantic feature which also belongs to the historical semantic feature is semantically adjusted, and thus the reliability of semantic adjustment is ensured, and the problem of easy semantic distortion caused by semantic adjustment based on a semantic feature with low adaptation degree is avoided, so that the semantic representation accuracy of the adjusted point singing semantic feature is higher, and thus the reliability of subsequent semantic restoration is ensured.

[0037] In step S150, the adjusted point singing semantic feature is used for semantic restoration, to form target recommended song information.

[0038] In the embodiments of the present application, after the adjusted point singing semantic feature is formed, the electronic device can perform semantic restoration based on the adjusted point singing semantic feature, to form target recommended song information. The target recommended song information is used for displaying to the current user through the target tablet terminal (for example, the electronic device can send the target recommended song information formed to the target tablet terminal for display, or the electronic device can also select part of the song names from the target recommended song information and send them to the target tablet terminal), to complete the point singing recommendation.

[0039] Based on the above, since in the process of semantic mining and fusion, the target preference semantic feature and the historical preference semantic feature are fused first to form the corresponding preference fusion semantic feature, and then the historical song semantic feature is adjusted based on the preference fusion semantic feature, the target preference semantic feature and the historical preference semantic feature are semantic features of the same dimension (i.e. both are representations of preferences), therefore, the accuracy of fusion is relatively higher, and the historical preference semantic feature and the historical song semantic feature are also at the historical level and have relevance, therefore, the preference fusion semantic feature and the historical song semantic feature formed by fusion are closer in the feature space, so that the reliability of the semantic adjustment based thereon can also be higher, therefore, the reliability of the semantic restoration based on the obtained adjusted song semantic feature can also be higher, so that reliable target recommended song information is obtained, and the problem of relatively low reliability of song recommendation in the prior art is improved.

[0040] In the first aspect, it needs to be explained that the specific content of the obtained target user preference information and historical user preference information is not limited, and can be selected according to actual needs.

[0041] For example, in an alternative embodiment, the target user preference information can be preference information input by the current user in the process of using the target tablet terminal to order songs, such as song-related preferences (such as song style, etc.) or other information that can affect song selection (such as age, occupation, etc.). In addition, it needs to be explained that different singing or listening environments will also affect the song selection of the current user, therefore, the target user preference information can also include environmental information, such as light, sound, etc.

[0042] In the second aspect, it needs to be explained that the specific way of respectively performing semantic mining on the target user preference information and the historical user preference information is not limited, and can be selected according to actual needs.

[0043] For example, in an alternative embodiment, in order to ensure that the target preference semantic feature and the historical preference semantic feature formed have high semantic representation ability, such as richer semantic information, the above step S120 can further include step S121, step S122, step S123 and step S124, and the specific content of each step is described as follows.

[0044] Step S121: respectively performing semantic mining on the user input preference information included in the target user preference information and the environmental information, to form a first preference semantic feature and a target environmental semantic feature.

[0045] In the embodiments of the present application, the user input preference information and the environment information included in the target user preference information can be respectively subjected to semantic mining to form first preference semantic features and target environment semantic features. Exemplarily, the user input preference information can be text information, based on which embedding processing can be performed on the user input preference information (for example, a word embedding model can be used), to obtain corresponding first preference semantic features. In addition, when the environment information is sound data, the time-domain sound data can be converted into a frequency spectrum diagram, and thus the frequency spectrum diagram can be subjected to convolution processing through a convolution network layer to obtain target environment semantic features. Alternatively, when the environment information is image data, the environment information can be subjected to convolution processing through a convolution network layer to obtain target environment semantic features.

[0046] In step S122, the user input preference information and the environment information included in the historical user preference information are respectively subjected to semantic mining to form second preference semantic features and historical environment semantic features.

[0047] In the embodiments of the present application, the user input preference information and the environment information included in the historical user preference information can be respectively subjected to semantic mining to form second preference semantic features and historical environment semantic features. The specific process of semantic mining can refer to the explanation and description of step S121. In addition, it should be noted that when there are multiple historical user preference information, the second preference semantic features and the historical environment semantic features corresponding to each historical user preference information can be obtained respectively, and then the second preference semantic features can be fused by averaging or the like to form final second preference semantic features, and the historical environment semantic features can be fused by averaging or the like to form final historical environment semantic features.

[0048] In step S123, the first preference semantic features and the target environment semantic features are subjected to semantic fusion to form target preference semantic features.

[0049] In the embodiments of the present application, after the first preference semantic features and the target environment semantic features are obtained, the first preference semantic features and the target environment semantic features can be subjected to semantic fusion to form target preference semantic features, that is, the preference semantic information of two dimensions is fused to obtain target preference semantic features with higher semantic representation richness.

[0050] In step S124, the second preference semantic features and the historical environment semantic features are subjected to semantic fusion to form historical preference semantic features.

[0051] In the embodiment of the present application, after the second preference semantic feature and the historical environment semantic feature are obtained, the second preference semantic feature and the historical environment semantic feature can be semantically fused to form a historical preference semantic feature, that is, the preference semantic information of the two dimensions is fused to obtain a historical preference semantic feature with higher semantic representation richness.

[0052] It can be understood that the specific manner of semantically fusing the first preference semantic feature and the target environment semantic feature in the above step S123 is not limited, for example, in an alternative embodiment, in order to ensure that the target preference semantic feature formed has higher semantic representation accuracy, the above step S123 can further include steps S123a, S123b and S123c, and the specific contents of each step are as follows.

[0053] Step S123a, in each forward diffusion stage, gradually apply a noise semantic feature to the target environment semantic feature to form a noise environment semantic feature corresponding to each forward diffusion stage.

[0054] In the embodiment of the present application, in each forward diffusion stage, a noise semantic feature can be gradually applied to the target environment semantic feature to form a noise environment semantic feature corresponding to each forward diffusion stage. It should be noted that since the environment information is susceptible to various disturbances in the process of collection, distortion problems exist, and therefore the distortion can be simulated by applying noise. For example, for the first forward diffusion stage, a random noise vector can be generated, and then the random noise vector can be added to the target environment semantic feature to form a noise environment semantic feature corresponding to the first forward diffusion stage. For the second forward diffusion node, the noise environment semantic feature corresponding to the first forward diffusion stage and the random noise vector can be added to form a noise environment semantic feature corresponding to the second forward diffusion stage. Based on this, by continuously increasing the forward diffusion stage, the amplitude of the applied noise can be gradually increased.

[0055] Step S123b, in each backward diffusion stage, based on the first preference semantic feature, noise suppression is performed on the noise environment semantic feature corresponding to the corresponding forward diffusion stage to form a noise suppression semantic feature corresponding to each backward diffusion stage.

[0056] In this embodiment of the invention, after forming the noise environment semantic features corresponding to each forward diffusion stage, noise suppression can be performed on the noise environment semantic features corresponding to the forward diffusion stage at each backward diffusion stage based on the first preferred semantic features, thus forming noise-suppressed semantic features corresponding to each backward diffusion stage. It should be noted that while the application of noise can restore distorted semantic information to a certain extent, it may also introduce real noise, affecting the expression of correct semantics. Therefore, noise can be suppressed using the first preferred semantic features with a correlation, thereby improving the accuracy of the obtained semantic features.

[0057] Step S123c: Based on the noise suppression semantic features corresponding to the last backward diffusion stage, determine the target preference semantic features.

[0058] In this embodiment of the application, after obtaining the noise suppression semantic features corresponding to the last backward diffusion stage, the target preference semantic features can be determined based on the noise suppression semantic features corresponding to the last backward diffusion stage. For example, the noise suppression semantic features corresponding to the last backward diffusion stage can be determined as the target preference semantic features.

[0059] It is understood that in step S123b above, the specific method of noise suppression for the semantic features of the noise environment corresponding to the forward diffusion stage is not limited. For example, in an alternative implementation, in order to ensure the reliability of noise suppression, that is, to balance noise suppression and actual semantic mining, step S123b above may include the following (in combination with...). Figure 3 (as shown) First, in the first backward diffusion stage, based on the first preference semantic features, cross-attention mining is performed on the noise environment semantic features corresponding to the last forward diffusion stage to achieve noise suppression, forming the noise suppression semantic features corresponding to the first backward diffusion stage. In other words, the cross-attention mechanism can be used to mine associated semantic information, thereby suppressing unrelated noise information. Secondly, in each subsequent backward diffusion stage, based on the noise suppression semantic features corresponding to the previous backward diffusion stage, cross-attention mining is performed on the noise environment semantic features corresponding to the corresponding forward diffusion stage to achieve noise suppression and form the noise suppression semantic features corresponding to the current backward diffusion stage. For example, based on the noise suppression semantic features corresponding to the first backward diffusion stage, cross-attention mining can be performed on the noise environment semantic features corresponding to the penultimate forward diffusion stage to form the noise suppression semantic features corresponding to the second backward diffusion stage.

[0060] For step S130, it is to be explained that the specific way of performing semantic mining on the historical karaoke information is not limited, and can be selected according to actual needs.

[0061] For example, in an alternative embodiment, in order to ensure that the formed historical karaoke semantic features have high semantic representation ability, the above step S130 can further include the following contents (in combination with Figure 4 Firstly, for each historical song in the historical karaoke information, a plurality of convolutional processing can be performed on the song spectrum of the historical song based on a local mask mechanism, to form a plurality of song convolutional semantic features, wherein for any two convolutional processing, the local frequency area masked by the local mask mechanism in the song spectrum is at least partially different; that is, due to different frequency components, the semantic information such as emotion represented is different, therefore, in order to enable the mined historical karaoke semantic features to fully represent the emotional information carried, thereby facilitating subsequent recommendation, different frequency areas can be masked and mined respectively, specifically, the increase of low frequency can convey emotions of melancholy or security; the strength and clarity of medium frequency can represent the urgency or authenticity of emotion; the enhancement of high frequency component can convey happiness, excitement or excitement; therefore, low frequency, medium frequency and high frequency can be respectively subjected to convolutional processing, so as to respectively mine semantic information in low frequency, semantic information in medium frequency and semantic information in high frequency; in addition, it is to be explained that the specific division of the three frequency bands is not limited, for example, in an example, low frequency can be 20Hz-250Hz, medium frequency can be 250Hz-2kHz, and high frequency can be 2kHz-20kHz; it is to be explained that for the local mask mechanism, before convolutional processing of low frequency, the local spectrum of medium frequency and high frequency can be masked, for example, the song spectrum is multiplied by a mask matrix, wherein the parameters of the area corresponding to low frequency in the mask matrix are equal to 1, and the parameters of the areas corresponding to medium frequency and high frequency are equal to 0; then, the result of multiplication is subjected to convolutional processing through a convolutional network layer; Secondly, the plurality of song convolutional semantic features are subjected to addition operation, and the result of the addition operation is subjected to normalization mapping to form a historical karaoke semantic feature; based on this, a historical karaoke semantic feature corresponding to a historical song can be formed, it is to be explained that when there are a plurality of historical songs, a plurality of corresponding historical karaoke semantic features can be fused by mean value or the like to form a final historical karaoke semantic feature.

[0062] For step S140, it is to be explained that the specific way of performing semantic adjustment on the historical karaoke semantic feature is not limited, and can be selected according to actual needs.

[0063] ​For example, in an alternative implementation, in order to guarantee the reliability of the formed adjustment point song semantic features, the step S140 can further include steps S141, S142 and S143, and the specific contents of each step are as follows.

[0064] In step S141, the historical preference semantic features are semantically mapped to form historical preference mapping parameters.

[0065] In the embodiments of the present application, the historical preference semantic features can be semantically mapped to form historical preference mapping parameters. For example, the historical preference semantic features can be linearly mapped (such as through a fully connected network layer, which can not change the feature size or dimension of the semantic features). Then, the result of linear mapping can be nonlinearly activated (such as through a sigmiod function), thereby obtaining the historical preference mapping parameters. The historical preference mapping parameters are used to represent the importance of each position (therefore, the value of each position parameter belongs to 0-1, wherein 0 represents that the importance of the corresponding position is very low, and the corresponding semantic information can be discarded, and 1 represents that the importance of the corresponding position is very high, and the corresponding semantic information can be completely retained).

[0066] In step S142, the historical preference semantic features are weighted mapped based on the historical preference mapping parameters to realize semantic fusion of the target preference semantic features and the historical preference semantic features, thereby forming preference fusion semantic features.

[0067] In the embodiments of the present application, after the historical preference mapping parameters are formed, the historical preference semantic features can be weighted mapped (such as bit-by-bit multiplication operation) based on the historical preference mapping parameters to realize semantic fusion of the target preference semantic features and the historical preference semantic features, thereby forming the preference fusion semantic features.

[0068] In step S143, the historical point song semantic features are semantically adjusted based on the preference fusion semantic features to form adjustment point song semantic features.

[0069] In the embodiments of the present application, after the preference fusion semantic features are formed, the historical point song semantic features can be semantically adjusted based on the preference fusion semantic features to form adjustment point song semantic features, that is, the preference fusion semantic features can be fused into the historical point song semantic features to realize semantic influence on the historical point song semantic features.

[0070] It can be understood that the specific manner of adjusting the historical karaoke semantic features in step S143 is not limited, and in order to guarantee the reliability of the semantic adjustment, a sufficient fusion operation can be performed to achieve this, based on which, step S143 can further include steps S143a, S143b and S143c, and the specific content of each step is as follows.

[0071] In step S143a, the historical karaoke semantic features are adjusted based on the preference fusion semantic features to form the intermediate adjustment semantic features of the first time step.

[0072] In the embodiment of the present application, in the first time step, the historical karaoke semantic features can be adjusted based on the preference fusion semantic features to form the intermediate adjustment semantic features of the first time step, that is, the preference fusion semantic features are fused into the historical karaoke semantic features.

[0073] In step S143b, in the second and each subsequent time step, the intermediate adjustment semantic features of the previous time step are adjusted based on the preference fusion semantic features to form the intermediate adjustment semantic features of the current time step.

[0074] In the embodiment of the present application, after the intermediate adjustment semantic features of the first time step are formed, in the second and each subsequent time step, the intermediate adjustment semantic features of the previous time step can be adjusted based on the preference fusion semantic features to form the intermediate adjustment semantic features of the current time step. That is, by fusing the preference fusion semantic features in each of the multiple time steps, the fusion is more sufficient.

[0075] In step S143c, the adjusted karaoke semantic features are determined based on the intermediate adjustment semantic features of the last time step.

[0076] In the embodiment of the present application, after the intermediate adjustment semantic features of the last time step are formed, the adjusted karaoke semantic features can be determined based on the intermediate adjustment semantic features of the last time step, for example, the intermediate adjustment semantic features of the last time step can be determined as the adjusted karaoke semantic features. Or, in other embodiments, the intermediate adjustment semantic features of the last time step can be further mined, such as self-attention processing, to obtain the adjusted karaoke semantic features.

[0077] It can be understood that the manner of semantic adjustment in steps S143a and S143b can be the same, and based on this, step S143b can further include steps b1, b2 and b3, and the specific content of each step is as follows.

[0078] In the second and each subsequent time step, after linear mapping the preference fusion semantic feature and the intermediate adjustment semantic feature of the previous time step, the two linear mapping features obtained are multiplied to form a first association parameter distribution.

[0079] In the second and each subsequent time step, after linear mapping the preference fusion semantic feature and the intermediate adjustment semantic feature of the previous time step, the two linear mapping features obtained are multiplied to form a first association parameter distribution. Illustratively, the preference fusion semantic feature can be multiplied by a first weight matrix to obtain a first linear mapping feature, and the intermediate adjustment semantic feature of the previous time step can be multiplied by a second weight matrix to obtain a second linear mapping feature, and then the first linear mapping feature and the second linear mapping feature can be dot producted to form the first association parameter distribution.

[0080] Step b2, obtaining an association weight distribution corresponding to the current time step, and weighting the first association parameter distribution based on the association weight distribution to form a second association parameter distribution.

[0081] In the second and each subsequent time step, after linear mapping the preference fusion semantic feature and the intermediate adjustment semantic feature of the previous time step, the two linear mapping features obtained are multiplied to form a first association parameter distribution. Illustratively, the preference fusion semantic feature can be multiplied by a first weight matrix to obtain a first linear mapping feature, and the intermediate adjustment semantic feature of the previous time step can be multiplied by a second weight matrix to obtain a second linear mapping feature, and then the first linear mapping feature and the second linear mapping feature can be dot producted to form the first association parameter distribution.

[0082] Step b3, performing a normalized mapping on the second correlation parameter distribution to form a third correlation parameter distribution, and performing a correlation adjustment on the linear mapping feature of the intermediate adjusted semantic feature of the previous time step based on the third correlation parameter distribution to form the intermediate adjusted semantic feature of the current time step.

[0083] In the embodiments of the present application, after the second correlation parameter distribution is formed, the second correlation parameter distribution can be normalized mapped to form a third correlation parameter distribution, and the linear mapping feature of the intermediate adjusted semantic feature of the previous time step can be adjusted based on the third correlation parameter distribution to form the intermediate adjusted semantic feature of the current time step. For example, the intermediate adjusted semantic feature of the previous time step can be multiplied by a third weight matrix, and then the result of the multiplication can be weighted and summed based on the third correlation parameter distribution to obtain the intermediate adjusted semantic feature of the current time step.

[0084] The fifth aspect needs to be explained for step S150, and the specific way of performing semantic restoration based on the adjusted song semantic feature is not limited, and can be selected according to actual needs.

[0085] For example, in an alternative implementation, the adjusted song semantic feature can be processed by a fully connected network layer to obtain a fully connected feature, and the size of the fully connected feature can be equal to the number of songs in the song library. For example, if the number of songs is n, the size of the fully connected feature is 1*n (i.e. one parameter in the fully connected feature corresponds to one song), and then a classification function (such as a softmax function) can be used to map the fully connected feature to form a probability distribution with a size of 1*n. Based on this, the maximum probability in the probability distribution can be determined, and then the song corresponding to the probability can be determined as the song in the target recommended song information. Alternatively, after obtaining the fully connected feature, an identity mapping can be performed on the fully connected feature, so that the song corresponding to the maximum parameter or parameters in the identity mapping can be determined as the song in the target recommended song information. Based on this, semantic restoration can be achieved to obtain the target song information.

[0086] In combination Figure 5 The embodiments of the present application also provide a tablet terminal-based song ordering background recommendation device applicable to the above-mentioned electronic device. The tablet terminal-based song ordering background recommendation device can include a song ordering preference acquisition module, a preference semantic mining module, a song semantic mining module, a semantic fusion adjustment module, and a song semantic restoration module.

[0087] The song ordering preference acquisition module is configured to acquire target user preference information of a current user corresponding to a target tablet terminal, and acquire historical user preference information and historical song ordering information, wherein the historical user preference information and the historical song ordering information correspond to a historical user corresponding to the target tablet terminal and / or a historical user corresponding to an associated tablet terminal of the target tablet terminal. Figure 2 The step S110 is shown, and the related content of the song ordering preference acquisition module can be referred to the description of the step S110.

[0088] The preference semantic mining module is configured to perform semantic mining on the target user preference information and the historical user preference information respectively, to form target preference semantic features and historical preference semantic features. Figure 2 The step S120 is shown, and the related content of the preference semantic mining module can be referred to the description of the step S120.

[0089] The song ordering semantic mining module is configured to perform semantic mining on the historical song ordering information, to form historical song ordering semantic features. Figure 2 The step S130 is shown, and the related content of the song ordering semantic mining module can be referred to the description of the step S130.

[0090] The semantic fusion adjustment module is configured to perform semantic adjustment on the historical song ordering semantic features based on a preference fusion semantic feature of the target preference semantic features and the historical preference semantic features, to form adjusted song ordering semantic features. Figure 2 The step S140 is shown, and the related content of the semantic fusion adjustment module can be referred to the description of the step S140.

[0091] The song ordering semantic restoration module is configured to perform semantic restoration based on the adjusted song ordering semantic features, to form target recommended song information, wherein the target recommended song information is used to display to the current user through the target tablet terminal, to complete song ordering recommendation. Figure 2 The step S150 is shown, and the related content of the song ordering semantic restoration module can be referred to the description of the step S150.

[0092] To sum up, the method, device and equipment for recommending songs in the background based on a tablet terminal are provided in the application. First, the target user preference information is obtained, and the historical user preference information and historical song ordering information are obtained. Second, the target user preference information and the historical user preference information are subjected to semantic mining respectively to form target preference semantic features and historical preference semantic features. Third, the historical song ordering information is subjected to semantic mining to form historical song ordering semantic features. Fourth, the historical song ordering semantic features are subjected to semantic adjustment based on the preference fusion semantic features of the target preference semantic features and the historical preference semantic features to form adjusted song ordering semantic features. Finally, the target recommended song information is formed based on the semantic restoration of the adjusted song ordering semantic features. Based on the above, the target preference semantic features and the historical preference semantic features are fused first to form corresponding preference fusion semantic features, and then the historical song ordering semantic features are subjected to semantic adjustment based on the preference fusion semantic features. The target preference semantic features and the historical preference semantic features are semantic features of the same dimension (i.e., both are representations of preferences), so the accuracy of the fusion is relatively higher. Moreover, the historical preference semantic features and the historical song ordering semantic features are both of the historical level and have relevance, so the preference fusion semantic features formed by the fusion are closer to the historical song ordering semantic features in the feature space, so the reliability of the semantic adjustment based thereon can be higher. Therefore, the reliability of the semantic restoration based on the obtained adjusted song ordering semantic features can be higher, so reliable target recommended song information is obtained, and the problem of relatively low reliability of song ordering recommendation in the prior art is improved.

[0093] In several embodiments provided by the embodiments of the present application, it should be understood that the disclosed apparatus and method can also be implemented by other manners. The apparatus and method embodiments described above are only illustrative. For example, the flowchart and block diagram in the drawings show the possible implementation architecture, function and operation of the apparatus, method and computer program product according to the embodiments of the present application. In this regard, each block in the flowchart or block diagram can represent a module, a program segment or a part of code containing one or more executable instructions for implementing the specified logic function. It should also be noted that in some alternative implementation manners, the functions noted in the blocks can occur in different order from that noted in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and they can also be executed in reverse order depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for executing the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0094] In addition, each functional module in the various embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0095] The functions, if implemented in the form of software functional modules and sold or used as independent products, can be stored in a computer readable storage medium. Based on such an understanding, the technical solutions of the present application can essentially or in part be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods according to the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and various other media that can store program codes. It should be noted that, in this document, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that processes, methods, articles or devices that include a series of elements do not only include those elements, but also include other elements not explicitly listed, or further include elements inherent to such processes, methods, articles or devices. Without more limitations, the element defined by the statement "comprises a" does not exclude the presence of additional identical elements in the process, method, article or device that includes the element.

[0096] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for recommending a song order background based on a tablet terminal, characterized by, The method comprises the following steps: obtaining target user preference information corresponding to a current user of a target tablet terminal, and obtaining historical user preference information and historical song ordering information, wherein the historical user preference information and the historical song ordering information correspond to a historical user corresponding to the target tablet terminal and / or a historical user corresponding to an associated tablet terminal of the target tablet terminal; respectively performing semantic mining on the target user preference information and the historical user preference information to form target preference semantic features and historical preference semantic features; performing semantic mining on the historical song ordering information to form historical song ordering semantic features; based on preference fusion semantic features of the target preference semantic features and the historical preference semantic features, performing semantic adjustment on the historical song ordering semantic features to form adjusted song ordering semantic features; based on the adjusted song ordering semantic features, performing semantic restoration to form target recommended song information, wherein the target recommended song information is used to display to the current user through the target tablet terminal to complete song ordering recommendation.

2. The flat-panel terminal-based song ordering background recommendation method according to claim 1, characterized by, The step of based on the preference fusion semantic features of the target preference semantic features and the historical preference semantic features, performing semantic adjustment on the historical song ordering semantic features to form adjusted song ordering semantic features comprises: performing semantic mapping on the historical preference semantic features to form historical preference mapping parameters, wherein the historical preference mapping parameters are used to represent the importance of each position; based on the historical preference mapping parameters, performing weighted mapping on the historical preference semantic features to realize semantic fusion of the target preference semantic features and the historical preference semantic features to form preference fusion semantic features; based on the preference fusion semantic features, performing semantic adjustment on the historical song ordering semantic features to form adjusted song ordering semantic features.

3. The flat-panel terminal-based song ordering background recommendation method according to claim 2, characterized by, The step of based on the preference fusion semantic features, performing semantic adjustment on the historical song ordering semantic features to form adjusted song ordering semantic features comprises: in a first time step, based on the preference fusion semantic features, performing semantic adjustment on the historical song ordering semantic features to form an intermediate adjustment semantic feature of the first time step; in a second and each subsequent time step, based on the preference fusion semantic features, performing semantic adjustment on an intermediate adjustment semantic feature of a previous time step to form an intermediate adjustment semantic feature of a current time step; based on the intermediate adjustment semantic feature of the last time step, determining the adjusted song ordering semantic features.

4. The flat-panel terminal-based song ordering background recommendation method according to claim 3, characterized by, The step of in a second and each subsequent time step, based on the preference fusion semantic features, performing semantic adjustment on an intermediate adjustment semantic feature of a previous time step to form an intermediate adjustment semantic feature of a current time step comprises: in a second and each subsequent time step, after linear mapping of the preference fusion semantic features and the intermediate adjustment semantic feature of the previous time step, multiplying the two linear mapping features obtained to form a first association parameter distribution; obtaining an association weight distribution corresponding to a current time step, and weighting the first association parameter distribution based on the association weight distribution to form a second association parameter distribution, wherein the association weight distribution is obtained by multiplying an initial weight distribution and a gating adjustment parameter determined based on an intermediate adjustment semantic feature of a previous time step, and the initial weight distribution is obtained by learning sample data and corresponding labels; performing normalization mapping on the second association parameter distribution to form a third association parameter distribution, and performing association adjustment on a linear mapping feature of the intermediate adjustment semantic feature of the previous time step based on the third association parameter distribution to form an intermediate adjustment semantic feature of the current time step.

5. The flat-panel terminal-based song ordering background recommendation method according to claim 1, characterized by, The steps of respectively performing semantic mining on the target user preference information and the historical user preference information to form target preference semantic features and historical preference semantic features include: performing semantic mining on user input preference information and environment information included in the target user preference information to form first preference semantic features and target environment semantic features; performing semantic mining on user input preference information and environment information included in the historical user preference information to form second preference semantic features and historical environment semantic features; performing semantic fusion on the first preference semantic features and the target environment semantic features to form target preference semantic features; performing semantic fusion on the second preference semantic features and the historical environment semantic features to form historical preference semantic features.

6. The flat-panel terminal-based song ordering background recommendation method according to claim 5, characterized by, The step of performing semantic fusion on the first preference semantic features and the target environment semantic features to form target preference semantic features includes: in each forward diffusion stage, gradually applying a noise semantic feature to the target environment semantic feature to form a noise environment semantic feature corresponding to each forward diffusion stage; in each backward diffusion stage, performing noise suppression on the noise environment semantic feature corresponding to the corresponding forward diffusion stage based on the first preference semantic feature to form a noise suppression semantic feature corresponding to each backward diffusion stage; based on the noise suppression semantic feature corresponding to the last backward diffusion stage, determining the target preference semantic feature.

7. The flat-panel terminal-based song ordering background recommendation method according to claim 6, characterized by, The step of, in each backward diffusion stage, performing noise suppression on the noise environment semantic feature corresponding to the corresponding forward diffusion stage based on the first preference semantic feature to form a noise suppression semantic feature corresponding to each backward diffusion stage includes: in the first backward diffusion stage, based on the first preference semantic feature, performing cross-attention mining on the noise environment semantic feature corresponding to the last forward diffusion stage to realize noise suppression, to form a noise suppression semantic feature corresponding to the first backward diffusion stage; in each of the second and subsequent backward diffusion stages, based on the noise suppression semantic feature corresponding to the previous backward diffusion stage, performing cross-attention mining on the noise environment semantic feature corresponding to the corresponding forward diffusion stage to realize noise suppression, to form a noise suppression semantic feature corresponding to the current backward diffusion stage.

8. The tablet terminal-based song ordering background recommendation method according to any one of claims 1 to 7, characterized by, The step of performing semantic mining on the historical karaoke information to form historical karaoke semantic features comprises: For each historical song in the historical karaoke information, a plurality of convolutional processing is performed on the song spectrogram of the historical song based on a local mask mechanism, to form a plurality of song convolutional semantic features, wherein the local frequency regions masked in the song spectrogram are at least partially different for any two convolutional processing through the local mask mechanism; The plurality of song convolutional semantic features are added and the result of the addition is normalized to form historical karaoke semantic features.

9. A flat panel terminal-based song ordering background recommendation apparatus, characterized by comprising: Comprise: A karaoke preference acquisition module is configured to acquire target user preference information of a current user corresponding to a target tablet terminal, and to acquire historical user preference information and historical karaoke information, wherein the historical user preference information and the historical karaoke information correspond to a historical user corresponding to the target tablet terminal and / or a historical user corresponding to an associated tablet terminal of the target tablet terminal; A preference semantic mining module is configured to perform semantic mining on the target user preference information and the historical user preference information respectively to form target preference semantic features and historical preference semantic features; A karaoke semantic mining module is configured to perform semantic mining on the historical karaoke information to form historical karaoke semantic features; A semantic fusion adjustment module is configured to perform semantic adjustment on the historical karaoke semantic features based on preference fusion semantic features of the target preference semantic features and the historical preference semantic features to form adjusted karaoke semantic features; A karaoke semantic restoration module is configured to perform semantic restoration based on the adjusted karaoke semantic features to form target recommended song information, wherein the target recommended song information is used to display to the current user through the target tablet terminal to complete karaoke recommendation.

10. An electronic device, comprising: Comprise: A memory is configured to store a computer program; A processor connected with the memory is configured to execute the computer program stored in the memory to realize the karaoke background recommendation method based on a tablet terminal according to any one of claims 1-8.

Images