Detailed Description
The technical solutions of the embodiments of the present invention will be clearly described below with reference to the drawings in the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Currently, in a scenario that a user listens to songs through music software, the user can select the listened songs from the song listening function of the music software, the music software generally has two functions of a song collection list and a song recommendation list (i.e. a recommended song set), songs in the song collection list are added by the user, and songs in the song recommendation list are recommended by the music software according to the preference of the user for listening to songs. In some embodiments, the preferences of the user to listen to songs may be determined by the music software based on historical songs the user listened to. The user may listen to songs in the song collection list of the music software and the user may also listen to songs recommended by the song recommendation list of the music software.
However, during actual use, the user sometimes listens to different types of songs over time, e.g., the user prefers to listen to songs in the song recommendation list at 7-9 a.m. and to listen to songs in the song collection list at 7-9 a.m. It can be stated that the user sometimes prefers to listen to the collection of songs and sometimes prefers to listen to the recommended songs in the song recommendation list, so that the user needs to both the collection of songs and the recommendation list of songs.
The embodiment of the invention provides a music recommendation method aiming at the situation that the requirements of the user on a song collection list and a song recommendation list exist.
In the process of the music recommendation method, a user can obtain the number of user assets of the user in the music software by clicking a music recommendation mode in the music software in the user terminal, generate a music recommendation request according to the number of user assets, and send the music recommendation request to the server. After receiving the music recommendation request, the server may determine whether the number of user assets in the music recommendation request is greater than a preset number threshold, and if the determination result is that the number of user assets is greater than the preset number threshold, calculate an intention to listen to songs probability value of the user through the target calculation model, where in some embodiments, the intention to listen to songs probability value may be a proportional value according to a first number of songs collected in the song collection list and a second number of songs recommended in the recommended song list. After calculating the meaning probability value of the user's meaning of listening to songs, the first set of songs may be screened from the song collection list by a shuffling algorithm, which may be a fiher-yates shuffle algorithm in some embodiments, according to the meaning probability value of listening to songs, and the user characteristic information, the meaning information of listening to songs, the first set of songs, and the characteristic information of songs to be recommended may be input into a pre-trained song recommendation model to obtain the second set of songs. And finally, fusing the first song set and the second song set to obtain a target song set. The server may send the target song set to the user terminal, and recommend the target song set to a user corresponding to the user terminal. By combining the favorite song list and the recommended song set, the embodiment of the invention can more effectively recommend songs to the user.
The music recommendation method provided by the embodiment of the invention can be applied to a music recommendation system, the system comprises music recommendation equipment and a user terminal, the music recommendation equipment can be arranged in a server, and in some embodiments, the server can comprise intelligent terminal equipment such as a smart phone, a tablet computer, a notebook computer, a desktop computer, a vehicle-mounted intelligent terminal, a smart watch and the like, but not limited to the intelligent terminal equipment. In some embodiments, the server includes one or more databases therein, which may be used to store content such as audio files, songs, and the like. In some embodiments, the server may be a cloud server. In some embodiments, the server may be a single server or may be one or more clusters of servers that are a series of servers, and in some embodiments, the server may also be other computing-capable devices. In some embodiments, the user terminal may include, but is not limited to, smart terminal devices such as smart phones, tablet computers, notebook computers, desktop computers, car-mounted smart terminals, smart watches, etc., and in some embodiments, a music client, such as music software, may be installed on the user terminal.
The music recommendation system provided by the embodiment of the invention is schematically described below with reference to fig. 1.
Referring to fig. 1, fig. 1 is a schematic structural diagram of a music recommendation system provided by an embodiment of the present invention, where the system includes a user terminal 11 and a server 12, and in some embodiments, the user terminal 11 and the server 12 may establish a communication connection through a wireless communication manner, and in some scenarios, the user terminal 11 and the server 12 may also establish a communication connection through a wired communication manner.
In the embodiment of the invention, the server 12 can acquire user characteristic information and song listening characteristic information of a user, which are sent by the user terminal 11, wherein the user characteristic information comprises user identification information and/or user portrait characteristic information, the song listening characteristic information comprises one or more of song listening type information, song listening time information, song cutting information and song listening quantity information, the song listening type information comprises one or more of collection song information and recommendation song information, the song listening time information comprises one or more of song listening time length, current time, song listening starting time and current time, the server 12 can input the user characteristic information and the song listening characteristic information into a pre-trained target calculation model to obtain song intention probability value of the user, the server 12 can screen a first song set from a song collection list of the user according to the song listening intention probability value, extract song characteristic information to be recommended from each song to be recommended in a recommendation song set, input the user characteristic information, the song listening characteristic information, the first song set and the song to be recommended characteristic information into a pre-trained song model to obtain a second song set, the server 12 can input the user characteristic information, the song listening characteristic information and the song to be recommended to the first song characteristic information into the pre-trained song model, and the second song model can be processed by the server 12 to obtain a target song set, namely the target song set can be pushed to the target set 11, and the target song set can be processed to the target set is pushed to the target set, and the target set is obtained. In some embodiments, the output mode may be an interface displayed on the user terminal 11, in some embodiments, the output mode may also be a voice prompt, in other embodiments, the output mode may also be other modes, and the embodiment of the present invention does not specifically limit the output mode.
Therefore, the embodiment of the invention can more accurately and effectively recommend the songs preferred by the user to the user by combining the song collection mode with the song recommendation mode, and improves the accuracy and the effectiveness of song recommendation.
The music recommendation method provided by the embodiment of the invention is schematically described below with reference to fig. 2.
Referring specifically to fig. 2, fig. 2 is a flowchart of a music recommendation method provided by an embodiment of the present invention, where the music recommendation method of the embodiment of the present invention may be executed by a music recommendation device, and the music recommendation device is disposed in a server, where the server is specifically explained as before. Specifically, the method of the embodiment of the invention comprises the following steps.
S201, user characteristic information and song listening characteristic information of a user are obtained.
In the embodiment of the invention, the music recommendation device can acquire user characteristic information and song listening characteristic information of a user, wherein the user characteristic information comprises user identification information and/or user portrait characteristic information, in some embodiments, the song listening characteristic information comprises one or more of song listening type information, song listening time information, song cutting information and song listening quantity information, the song listening type information comprises collection song information and recommended song information, and the song listening time information comprises one or more of song listening time, current time, song listening starting time and current time interval. In some embodiments, the user characteristic information and the listen to song characteristic information are both vector information.
In some embodiments, the user identification information includes, but is not limited to, a user name, and the user portrayal characteristic information includes, but is not limited to, one or more of a user age, gender, user listening preference, and the like. In some embodiments, the song-listening type information is 1 or 0, and the song-listening type information is determined according to a historical song-listening record of the user within a preset time range from the current time, and is used for representing whether the song is in the song collection list. If the history listening record is a song in the song collection list, the listening type information is collection song information and takes 1, and if the history listening record is not a song in the song collection list, the listening type information is recommended song information and takes 0. In some embodiments, the listening to songs time information may be obtained via a time stamp. In some embodiments, the song-listening feature information may also include song base information, such as song base information may include, but is not limited to, one or more of song identification, language identification, singer identification, and the like.
In one embodiment, before obtaining the user characteristic information and the song listening characteristic information, the music recommendation device may obtain the number of songs in the song collection list of the user when receiving a music recommendation request sent by the user terminal corresponding to the user, and determine whether the number of songs in the song collection list of the user is greater than a preset number threshold, if so, may execute the step of obtaining the user characteristic information and the song listening characteristic information, and if not, may output prompt information for instructing the user to add songs to the song collection list until the number of songs in the song collection list is greater than the preset number threshold, and then send the music recommendation request.
For example, assuming that the preset number threshold is 5, when the server receives a music recommendation request sent by a user terminal corresponding to a user, it may be determined whether the number of songs in the song collection list of the user is greater than 5, and if the number of songs in the song collection list is greater than 5, it may be determined to perform the step of acquiring the user feature information and the song listening feature information. If the number of songs in the song collection list is less than or equal to 5, a prompt message may be output for instructing the user to add songs to the song collection list until the number of songs in the song collection list is greater than 5.
In one embodiment, the music recommendation request carries user information and user song listening information, and the music recommendation device can extract user vector information and song listening vector information from the user information and the user song listening information carried by the music recommendation request when acquiring the user characteristic information and the user song listening information, namely, determine the user vector information as the user characteristic information and determine the song listening vector information as the song listening characteristic information.
In some embodiments, the music recommendation device may use a preset feature extraction model to obtain the user vector information and the song listening vector information when extracting the user vector information and the song listening vector information from the user information and the song listening information carried by the music recommendation request. In some embodiments, the preset feature extraction model may be obtained through training a neural network model, and the embodiment of the present invention does not specifically limit the feature extraction model. In some embodiments, the neural network model used in the embodiments of the present invention may be a deep neural network (Deep Neural Networks, DNN), the structure of which may be illustrated in fig. 3, fig. 3 is a schematic structural diagram of a DNN, and fig. 7 is a schematic structural diagram of a three-layer DNN, where the neural network layers inside the DNN may be divided into three types, i.e., an input layer, a hidden layer and an output layer, generally, the first layer is an input layer, the last layer is an output layer, the middle layers are all hidden layers, and the layers are all connected, that is, any neuron of the i layer is connected to any neuron of the i+1 layer.
S202, inputting the user characteristic information and the song listening characteristic information into a pre-trained target calculation model to obtain a song listening intention probability value of the user, and screening a first song set from a song collection list of the user according to the song listening intention probability value.
In the embodiment of the invention, the music recommendation equipment can input the user characteristic information and the song listening characteristic information into the pre-trained target calculation model to obtain the song listening intention probability value of the user, and screen the first song set from the song collection list of the user according to the song listening intention probability value. In some embodiments, the pre-trained target computing model may be trained from a neural network model. In some embodiments, the neural network model employed by embodiments of the present invention may include, but is not limited to, DNN.
In one embodiment, when the music recommendation device inputs the user characteristic information and the song listening characteristic information into the pre-trained target calculation model to obtain the song listening intention probability value of the user, the music recommendation device can input the user characteristic information and the song listening characteristic information into the pre-trained target calculation model to predict and obtain the first song number of the collected songs in the user song listening collection list and the second song number of the recommended songs in the user song listening recommendation list, and the proportional value of the first song number and the second song number is determined to be the song listening intention probability value of the user. In some embodiments, the listen to song intent probability value is used to indicate a probability value that the user would like to listen to songs of the song collection list.
For example, the music recommendation device inputs the user characteristic information and the listen song characteristic information acquired from the music recommendation request sent by the user 1 into the pre-trained target calculation model, predicts that the first song number of the collection songs in the listen song collection list of the user 1 is 10, and the second song number of the recommendation songs in the listen recommendation song list of the user 1 is 20, and can determine that the ratio value of the first song number 10 to the second song number 20 is 1/2, so that the listen song intention probability value of the user 1 can be determined to be 0.5.
In one embodiment, when the meaning probability value of the user for listening to songs is greater than a first preset threshold, the meaning probability value of listening to songs is confirmed to take the first threshold, and when the meaning probability value of listening to songs of the user is less than a second preset threshold, the meaning probability value of listening to songs is confirmed to take the second threshold, wherein the first preset probability is greater than the second preset probability. For example, assuming that the probability value of the intention to listen to songs is P, the first preset threshold value is 0.8, the second preset threshold value is 0.2, when P is greater than 0.8, the P value is determined to be 0.8, and when P is less than 0.2, the P value is determined to be 0.2.
In one embodiment, the music recommendation device may acquire a first training data set before inputting user feature information and listening feature information into a pre-trained target calculation model to obtain a listening intention probability value of a user, the first training data set including a plurality of first training data, the plurality of first training data including first training user feature information and first training listening feature information, input the plurality of first training data in the first training data set into a preset first deep neural network model to perform training to obtain a first loss function value, adjust a first model parameter of the first deep neural network model according to the first loss function value, input the plurality of first training data in the first training data set into the first deep neural network model after adjusting the first model parameter to perform retraining, and determine to obtain the target calculation model when the retraining obtained first loss function value is smaller than a first preset threshold.
Specifically, fig. 4 is a schematic training diagram of a target computing model provided by the embodiment of the present invention, where a first training data set 41 is input into a preset first feature extraction model 42 to obtain first training user feature information and first training song listening feature information, the first training data set 41 includes information such as user identification information, user portrait information, song listening type information, song listening time information, song cutting information, and song listening number information, the first user feature information and the first training song listening feature information are input into the first deep neural network model 43 to perform training to obtain a first loss function value 44, a first model parameter of the first deep neural network model 43 is adjusted according to the first loss function value 44, a plurality of first training data are input into the first deep neural network model 43 after the first model parameter is adjusted to perform retraining, and when the first loss function value 44 obtained by retraining is smaller than a first preset threshold, the target computing model is determined.
In one embodiment, when the music recommendation device screens the first song set from the song collection list of the user according to the listening intention probability value, a shuffling algorithm may be utilized to randomly select a song priority list from the song collection list, determine the number to be screened according to the listening intention probability value and the number of songs collected in the song priority list, and screen the first song set from the song priority list according to the number to be screened.
In one embodiment, the music recommendation device may exchange (or not exchange) the last song with one of any n-1 songs before and then the next-to-last song with one of any n-2 songs before when randomly selecting a song priority list from the song collection list using a shuffling algorithm that ensures that the probability of each element at each location is equal until the last element (i.e., song) is completed.
For example, assuming that the initial ordering of songs in the song collection is Song 0, song 1, song 2, song 3, song 4, indicated by the numeral 01234, the step of randomly selecting a list of song priorities from the song collection is:
Randomly selecting a number (e.g., 3) and a number 4 from among the 5 positions [0,4] (e.g., 0 and 4) to obtain 01243, randomly selecting a number (e.g., 0) and a number 3 from among the 4 positions [0,3] (e.g., 0 and 3) to obtain 41203, randomly selecting a number (e.g., 0) and a number 3 from among the 4 positions [0,4] (e.g., 0 and 4) =1/5), randomly selecting a number (e.g., 0) and a number 2 from among the 3 positions [0,2] (e.g., 0 and 2) to obtain 21403, randomly selecting a number (e.g., 0) and a number 1/5 from among the 2 positions [0,1] (e.g., 0 and 1) to obtain 12403, and a number (e.g., 0/5) and a number (e.g., 0/3) =1/5 from among the 2 positions [0,2] (e.g., 0 and 3) to obtain 21403, and a number (e.g., 0/5) from among the 2 positions [0,2] to obtain a number (e.g., 0/3) and a number (0/3) =1/5 from among the 2 positions [0,2] to obtain 12403, and a number (e.g., 0/3) from among the number 2 from among the 3 positions (e.g., 0 and 2) from among the 3 (e.g., 0 and 3) and 3. Thus, song priority lists are available as Song 1, song 2, song 4, song 0, song 3.
In one embodiment, when determining the number to be filtered according to the listening intention probability value and the number of collected songs in the song priority list, the music recommendation device may determine the number to be filtered according to a product of the listening intention probability value and the number of collected songs in the song priority list.
And S203, extracting the characteristic information of the songs to be recommended from each song to be recommended in the recommended song set, and inputting the user characteristic information, the song listening characteristic information, the first song set and the characteristic information of the songs to be recommended into a pre-trained song recommendation model to obtain a second song set.
In the embodiment of the invention, the music recommendation device can acquire a recommended song set, extract the characteristic information of the song to be recommended from each song to be recommended in the recommended song set, and input the user characteristic information, the song listening characteristic information, the first song set and the characteristic information of the song to be recommended into a pre-trained song recommendation model to obtain the second song set. In some embodiments, the pre-trained song recommendation model may include, but is not limited to, DNN.
In one embodiment, when inputting user characteristic information, song listening characteristic information, a first song set and song characteristic information to be recommended into a pre-trained song recommendation model to obtain a second song set, the music recommendation device may input the user characteristic information, the song listening characteristic information, the first song set and the song characteristic information to be recommended into the pre-trained song recommendation model to obtain a recommendation score value corresponding to each song to be recommended in the recommended song set, sort the songs to be recommended according to the recommendation score value corresponding to each song to be recommended in order from large to small, and determine that the songs to be recommended arranged in the front M bits form the second song set, where M is a positive integer.
For example, assuming that the value of M is 5, the music recommendation device may input the user characteristic information, the listening characteristic information, the first song set of the user 1, and the to-be-recommended song characteristic information of the user 1 into the pre-trained song recommendation model, to obtain a recommendation score value of 80 corresponding to song 1, 82 corresponding to song 2, 83 corresponding to song 3, 81 corresponding to song 4, 79 corresponding to song 5, 84 corresponding to song 6, 78 corresponding to song 7, 85 corresponding to song 8, 86 corresponding to song 9, and 88 corresponding to song 10 in the to-be-recommended song set. And sorting the songs to be recommended according to the corresponding recommendation score value of each song to be recommended from big to small into a song 10, a song 9, a song 8, a song 6, a song 3, a song 2, a song 4, a song 1, a song 5 and a song 7, and determining that the song 10, the song 9, the song 8, the song 6 and the song 3 to be recommended which are arranged in the first 5 bits are the second song set.
In one embodiment, the music recommendation device may acquire a second training data set before inputting the user feature information, the song listening feature information, the first song set and the song feature information to be recommended into the pre-trained song recommendation model to obtain a second song set, the second training data set includes a plurality of second training data, the plurality of second training data includes one or more of second training user feature information, second training song listening feature information and song information of a song listening intention, the plurality of second training data in the second training data set are input into the second deep neural network model to perform training to obtain a second loss function value, a second model parameter of the second deep neural network model is adjusted according to the second loss function value, the second deep neural network model after the plurality of second training data in the second training data set are input into the second model parameter is retrained, and when the second loss function value obtained by retrained is smaller than a second preset threshold value, the song recommendation model is determined to be obtained. In some embodiments, the training listen to the meaning song information may include, but is not limited to, a listen to meaning probability value of the training user.
Specifically, fig. 5 is a schematic diagram of a song recommendation model according to an embodiment of the present invention, where the second training data set 51 includes user identification information, user portrait information, song listening type information, song listening time information, song cutting information, song listening quantity information, and song listening intention information, the second user feature information, the second training song listening feature information, and the song listening intention probability value are input into the second deep neural network model 53 to train, a second loss function value 54 is obtained, a second model parameter of the second deep neural network model 53 is adjusted according to the second loss function value 54, the second deep neural network model 53 after the second model parameter is adjusted is input into the plurality of second training data sets to retrain, and when the second loss function value 54 obtained by retraining is smaller than a second preset threshold, the song recommendation model is determined.
S204, carrying out fusion processing on the first song set and the second song set to obtain a target song set, and pushing the target song set to a user terminal corresponding to the user.
In the embodiment of the invention, the music recommendation device can perform fusion processing on the first song set and the second song set to obtain the target song set, and push the target song set to the user terminal corresponding to the user.
In one embodiment, when the music recommendation device performs fusion processing on the first song set and the second song set to obtain the target song set, the fusion processing may be performed on the first song set and the second song set according to a first preset weight of the first song set and a second preset weight of the second song set to obtain the target song set. In other embodiments, the music recommendation device may further perform fusion processing on the first song set and the second song set in other manners, and the manner of performing fusion processing on the first song set and the second song set in the embodiment of the present invention is not specifically limited.
The embodiment of the invention can input the user characteristic information and the song listening characteristic information into a pre-trained target calculation model to obtain a song listening intention probability value of a user, screen a first song set from a song collection list of the user according to the song listening intention probability value, input the user characteristic information, the song listening characteristic information, the first song set and the to-be-recommended song characteristic information into a pre-trained song recommendation model to obtain a second song set, and perform fusion processing on the first song set and the second song set to obtain a target song set and push the target song set to a user terminal corresponding to the user. By adopting a mode of combining the song listening intention of the user at different time, the song collection of the user and the song recommendation, songs meeting the user preference at different time periods can be more accurately and effectively recommended to the user, the song listening requirements of the user at different time periods are met, and the accuracy and the effectiveness of song recommendation are improved.
Referring to fig. 6, fig. 6 is a schematic structural diagram of a music recommendation device according to an embodiment of the present invention. Specifically, the music recommendation device is arranged in a server, and comprises an acquisition unit 601, a screening unit 602, a determination unit 603 and a pushing unit 604;
an obtaining unit 601, configured to obtain user feature information and song listening feature information of a user;
The screening unit 602 is configured to input the user feature information and the song listening feature information into a pre-trained target calculation model, obtain a user's listening intention probability value, and screen a first song set from the user's song collection list according to the listening intention probability value;
The determining unit 603 is configured to extract to-be-recommended song feature information from each to-be-recommended song in the recommended song set, and input the user feature information, the song listening feature information, the first song set and the to-be-recommended song feature information into a pre-trained song recommendation model to obtain a second song set;
and the pushing unit 604 is configured to perform fusion processing on the first song set and the second song set to obtain a target song set, and push the target song set to a user terminal corresponding to the user.
Further, the filtering unit 602 inputs the user feature information and the listening feature information into a pre-trained target calculation model, and is specifically configured to:
Inputting the user characteristic information and the song listening characteristic information into a pre-trained target calculation model, and predicting to obtain the first song number of the collected songs in the user song listening collection list and the second song number of the recommended songs in the user song listening recommendation list;
And determining the proportional value of the first song quantity and the second song quantity as the listening intention probability value of the user.
Further, when the screening unit 602 screens the first album from the song collection list of the user according to the meaning probability value of listening to songs, the screening unit is specifically configured to:
randomly selecting a song priority list from the song collection list by using a shuffling algorithm;
Determining the quantity to be screened according to the meaning probability value of the song and the quantity of the collected songs in the song priority list;
and screening the first song set from the song priority list according to the quantity to be screened.
Further, when the determining unit 603 inputs the user feature information, the song listening feature information, the first song set, and the song feature information to be recommended into a pre-trained song recommendation model to obtain a second song set, the determining unit is specifically configured to:
Inputting the user characteristic information, the song listening characteristic information, the first song set and the song to be recommended characteristic information into a pre-trained song recommendation model to obtain recommendation score values corresponding to each song to be recommended in the song set to be recommended;
And sequencing the songs to be recommended according to the recommendation score value corresponding to each song to be recommended in order from big to small, and determining that the songs to be recommended ranked in the first M bits form the second song set, wherein M is a positive integer.
Further, when the pushing unit 604 performs fusion processing on the first song set and the second song set to obtain a target song set, the pushing unit is specifically configured to:
And carrying out fusion processing on the first song set and the second song set according to the first preset weight of the first song set and the second preset weight of the second song set to obtain the target song set.
Further, before the filtering unit 602 inputs the user feature information and the listening feature information into the pre-trained target calculation model to obtain the listening intention probability value of the user, the filtering unit is further configured to:
acquiring a first training data set, wherein the first training data set comprises a plurality of first training data, and the plurality of first training data comprises first training user characteristic information and first training song listening characteristic training information;
Inputting the plurality of first training data in the first training data set into a preset first deep neural network model for training to obtain a first loss function value;
Adjusting a first model parameter of the first deep neural network model according to the first loss function value;
Inputting the plurality of first training data in the first training data set into a first deep neural network model with the first model parameters adjusted for retraining;
And when the first loss function value obtained through retraining is smaller than a first preset threshold value, determining to obtain the target calculation model.
Further, before the determining unit 603 inputs the user feature information, the song listening feature information, the first song set, and the song feature information to be recommended into a pre-trained song recommendation model to obtain a second song set, the determining unit is further configured to:
Acquiring a second training data set, wherein the second training data set comprises a plurality of second training data, and the plurality of second training data comprises one or more of second training user characteristic information, second training song listening characteristic information and training song listening intention song information;
Inputting the plurality of second training data in the second training data set into a second deep neural network model for training to obtain a second loss function value;
Adjusting a second model parameter of the second deep neural network model according to the second loss function value;
inputting the plurality of second training data in the second training data set into a second deep neural network model with the second model parameters adjusted for retraining;
and when the second loss function value obtained through retraining is smaller than a second preset threshold value, determining to obtain the song recommendation model.
Further, before the acquiring unit 601 acquires the user feature information and the song listening feature information of the user, the acquiring unit is further configured to:
When a music recommendation request sent by a user terminal corresponding to the user is received, acquiring the number of songs in a song collection list of the user, and judging whether the number of songs in the song collection list of the user is larger than a preset number threshold;
if yes, executing the steps of acquiring the user characteristic information and the song listening characteristic information;
If the judgment result is negative, outputting prompt information, wherein the prompt information is used for indicating the user to add songs to the song collection list until the number of the songs in the song collection list is greater than the preset number threshold value, and then sending the music recommendation request.
The embodiment of the invention can input the user characteristic information and the song listening characteristic information into a pre-trained target calculation model to obtain a song listening intention probability value of a user, screen a first song set from a song collection list of the user according to the song listening intention probability value, input the user characteristic information, the song listening characteristic information, the first song set and the song characteristic information to be recommended into a pre-trained song recommendation model to obtain a second song set, and perform fusion processing on the first song set and the second song set to obtain a target song set and push the target song set to a user terminal corresponding to the user. By the method, the requirement of users on listening songs in different environments is met, and songs can be recommended to the users more effectively.
Referring to fig. 7, fig. 7 is a schematic structural diagram of a server according to an embodiment of the present invention. Specifically, the server includes a memory 701 and a processor 702.
In one embodiment, the server further comprises a data interface 703, the data interface 703 is used for transferring data information between the server and other devices.
The memory 701 may include a volatile memory (volatile memory), the memory 701 may include a non-volatile memory (nonvolatile memory), and the memory 701 may include a combination of the above types of memories. The processor 702 may be a central processing unit (central processing unit, CPU). The processor 702 may further comprise a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (programmable logic device, PLD), or a combination thereof. The PLD may be a complex programmable logic device (complex programmable logic device, CPLD), a field-programmable gate array (FPGA) or any combination thereof.
The memory 701 is used for storing a program, and the processor 702 may call the program stored in the memory 701, for performing the following steps:
acquiring user characteristic information and song listening characteristic information of a user;
Inputting the user characteristic information and the song listening characteristic information into a pre-trained target calculation model to obtain a song listening intention probability value of a user, and screening a first song set from a song collection list of the user according to the song listening intention probability value;
extracting to-be-recommended song characteristic information from each to-be-recommended song of a recommended song set, and inputting the user characteristic information, the song listening characteristic information, the first song set and the to-be-recommended song characteristic information into a pre-trained song recommendation model to obtain a second song set;
And carrying out fusion processing on the first song set and the second song set to obtain a target song set, and pushing the target song set to a user terminal corresponding to the user.
Further, the processor 702 inputs the user feature information and the listening feature information into a pre-trained target calculation model, and is specifically configured to:
Inputting the user characteristic information and the song listening characteristic information into a pre-trained target calculation model, and predicting to obtain the first song number of the collected songs in the user song listening collection list and the second song number of the recommended songs in the user song listening recommendation list;
And determining the proportional value of the first song quantity and the second song quantity as the listening intention probability value of the user.
Further, when the processor 702 screens the first album from the song collection of the user according to the listening intention probability value, the processor is specifically configured to:
randomly selecting a song priority list from the song collection list by using a shuffling algorithm;
Determining the quantity to be screened according to the meaning probability value of the song and the quantity of the collected songs in the song priority list;
and screening the first song set from the song priority list according to the quantity to be screened.
Further, when the processor 702 inputs the user feature information, the song listening feature information, the first song set, and the song feature information to be recommended into a pre-trained song recommendation model to obtain a second song set, the method is specifically used for:
Inputting the user characteristic information, the song listening characteristic information, the first song set and the song to be recommended characteristic information into a pre-trained song recommendation model to obtain recommendation score values corresponding to each song to be recommended in the song set to be recommended;
And sequencing the songs to be recommended according to the recommendation score value corresponding to each song to be recommended in order from big to small, and determining that the songs to be recommended ranked in the first M bits form the second song set, wherein M is a positive integer.
Further, when the processor 702 performs fusion processing on the first song set and the second song set to obtain a target song set, the method is specifically used for:
And carrying out fusion processing on the first song set and the second song set according to the first preset weight of the first song set and the second preset weight of the second song set to obtain the target song set.
Further, before inputting the user feature information and the listening feature information into the pre-trained target calculation model to obtain the listening intention probability value of the user, the processor 702 is further configured to:
acquiring a first training data set, wherein the first training data set comprises a plurality of first training data, and the plurality of first training data comprises first training user characteristic information and first training song listening characteristic training information;
Inputting the plurality of first training data in the first training data set into a preset first deep neural network model for training to obtain a first loss function value;
Adjusting a first model parameter of the first deep neural network model according to the first loss function value;
Inputting the plurality of first training data in the first training data set into a first deep neural network model with the first model parameters adjusted for retraining;
And when the first loss function value obtained through retraining is smaller than a first preset threshold value, determining to obtain the target calculation model.
Further, before inputting the user feature information, the song listening feature information, the first song set, and the song feature information to be recommended into the pre-trained song recommendation model, the processor 702 is further configured to:
Acquiring a second training data set, wherein the second training data set comprises a plurality of second training data, and the plurality of second training data comprises one or more of second training user characteristic information, second training song listening characteristic information and training song listening intention song information;
Inputting the plurality of second training data in the second training data set into a second deep neural network model for training to obtain a second loss function value;
Adjusting a second model parameter of the second deep neural network model according to the second loss function value;
inputting the plurality of second training data in the second training data set into a second deep neural network model with the second model parameters adjusted for retraining;
and when the second loss function value obtained through retraining is smaller than a second preset threshold value, determining to obtain the song recommendation model.
Further, before the processor 702 obtains the user feature information and the song-listening feature information of the user, the processor is further configured to:
When a music recommendation request sent by a user terminal corresponding to the user is received, acquiring the number of songs in a song collection list of the user, and judging whether the number of songs in the song collection list of the user is larger than a preset number threshold;
if yes, executing the steps of acquiring the user characteristic information and the song listening characteristic information;
If the judgment result is negative, outputting prompt information, wherein the prompt information is used for indicating the user to add songs to the song collection list until the number of the songs in the song collection list is greater than the preset number threshold value, and then sending the music recommendation request.
The embodiment of the invention can input the user characteristic information and the song listening characteristic information into a pre-trained target calculation model to obtain a song listening intention probability value of a user, screen a first song set from a song collection list of the user according to the song listening intention probability value, input the user characteristic information, the song listening characteristic information, the first song set and the song characteristic information to be recommended into a pre-trained song recommendation model to obtain a second song set, and perform fusion processing on the first song set and the second song set to obtain a target song set and push the target song set to a user terminal corresponding to the user. By the method, the requirement of users on listening songs in different environments is met, and songs can be recommended to the users more effectively.
The embodiment of the present invention further provides a computer readable storage medium, where a computer program is stored, where the computer program when executed by a processor implements the method described in the embodiment corresponding to fig. 2 of the present invention, and may also implement the apparatus according to the embodiment corresponding to fig. 6 of the present invention, which is not described herein again.
The computer readable storage medium may be an internal storage unit of the device according to any of the foregoing embodiments, for example, a hard disk or a memory of the device. The computer readable storage medium may also be an external storage device of the device, such as a plug-in hard disk, a smart memory card (SMART MEDIA CARD, SMC), a Secure Digital (SD) card, a flash memory card (FLASH CARD), etc. that are provided on the device. Further, the computer readable storage medium may also include both internal storage units and external storage devices of the device. The computer-readable storage medium is used to store the computer program and other programs and data required by the terminal. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.
Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program stored on a computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a random-access Memory (Random Access Memory, RAM), or the like.
The above disclosure is only a few examples of the present invention, and it is not intended to limit the scope of the present invention, but it is understood by those skilled in the art that all or a part of the above embodiments may be implemented and equivalents thereof may be modified according to the scope of the present invention.