Object recommendation method and device, electronic equipment and storage medium

Abstract

The invention relates to an object recommendation method and device, electronic equipment and a storage medium, and the method comprises the steps that historical object sequences and interaction features corresponding to different behaviors executed by a user are obtained, the historical object sequence corresponding to each behavior comprises historical objects which are sequentially arranged according to behavior execution time, and the interaction features are obtained; the interaction feature corresponding to each behavior is used for representing the preference degree of the user to the behavior; obtaining preference feature representation of the user based on an attention mechanism according to the historical object sequences and the interaction features corresponding to all behaviors; and according to the to-be-recommended object and the preference feature representation, determining whether to recommend the to-be-recommended object to the user, and improving the feature expression ability of the preference feature representation, so that the recommendation effect of recommendation based on the preference feature representation can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of electronic information technology, and in particular to an object recommendation method, device, electronic device and storage medium. Background Art

[0002] With the rapid popularization of Internet technology, the amount of information has shown explosive growth. While it brings more choices to users, it also causes serious information overload problems. Users need time and energy to find the information they are interested in, and the traditional method of recommending objects to users has a low accuracy rate. Summary of the invention

[0003] In order to overcome the problems existing in the related art, the present disclosure provides an object recommendation method, device and storage medium.

[0004] According to a first aspect of an embodiment of the present disclosure, there is provided an object recommendation method, comprising:

[0005] Obtaining historical object sequences and interaction features corresponding to different behaviors performed by the user, wherein the historical object sequence corresponding to each behavior includes historical objects arranged in sequence according to the execution time of the behavior, and the interaction feature corresponding to each behavior is used to indicate the user's preference for the behavior;

[0006] According to the historical object sequences and the interaction features corresponding to all behaviors, obtaining the preference feature representation of the user based on the attention mechanism;

[0007] Determine whether to recommend the object to be recommended to the user based on the object to be recommended and the preference feature representation.

[0008] Optionally, obtaining the user's preference feature representation based on an attention mechanism according to the historical object sequences and the interaction features corresponding to all behaviors includes:

[0009] Encoding each of the historical object sequences to obtain a historical object sequence vector corresponding to each of the historical object sequences;

[0010] generating a plurality of virtual view vectors respectively corresponding to each of the historical object sequences;

[0011] Generate a plurality of target virtual view vectors corresponding to each of the historical object sequences based on an attention mechanism according to the historical object sequence vectors and the plurality of virtual view vectors respectively corresponding to each of the historical object sequences;

[0012] According to all the target virtual view vectors, all the virtual view vectors and the interaction features, a preference feature representation of the user is obtained based on an attention mechanism.

[0013] Optionally, generating a plurality of target virtual view vectors corresponding to each of the historical object sequences based on an attention mechanism according to the historical object sequence vectors and the plurality of virtual view vectors respectively corresponding to each of the historical object sequences includes:

[0014] For each of the historical object sequences, multiple virtual view vectors corresponding to the historical object sequence and the historical object sequence vector are input into the first attention layer to obtain a weighted representation corresponding to each virtual view vector corresponding to the historical object sequence. The weighted representation corresponding to each of the virtual view vectors is used to represent the target virtual view vector corresponding to the virtual view vector.

[0015] Optionally, for each of the historical object sequences, inputting a plurality of virtual view vectors corresponding to the historical object sequence and the historical object sequence vector into the first attention layer to obtain a weighted representation corresponding to each virtual view vector corresponding to the historical object sequence, including:

[0016] The weighted representation corresponding to each virtual view vector is calculated by the following formula:

[0017]

[0018]

[0019] K 1 =W K *X;

[0020] V 1 =W V *X;

[0021] Among them, out 1 A weighted representation for representing each virtual view vector corresponding to the historical object sequence, is used to represent multiple virtual view vectors corresponding to the historical object sequence, X is used to represent the historical object sequence vector, d k Used to represent the scaled factor, T represents the transposed symbol, W Q Used to represent the first weight matrix, W K represents the second weight matrix, W V represents the third weight matrix, softmax() is the normalized exponential function, K 1 is used to represent the weighted result of weighting the historical object sequence vector corresponding to the historical object sequence based on the second weight matrix, V 1 Used to represent a weighted result of weighting the historical object sequence vector corresponding to the historical object sequence based on the third weight matrix.

[0022] Optionally, obtaining the user's preference feature representation based on an attention mechanism according to all the target virtual view vectors, all the virtual view vectors and the interaction features includes:

[0023] All the target virtual view vectors, all the virtual view vectors and the interaction features are input into a second attention layer to obtain a preference feature representation of the user.

[0024] Optionally, inputting all the target virtual view vectors, all the virtual view vectors and the interaction features into a second attention layer to obtain the user's preference feature representation includes:

[0025] The user's preference feature representation is calculated by the following formula:

[0026]

[0027] Among them, out 2 The feature representation used to characterize the user’s preference, Q 2 Used to characterize the interaction features, K 2 The virtual view vector V is used to represent all the targets. 2 Used to represent all the virtual view vectors, T represents the transposed symbol, d k Used to characterize the scaled factor, softmax() is a normalized exponential function.

[0028] Optionally, determining whether to recommend the object to be recommended to the user according to the object to be recommended and the preference feature representation includes:

[0029] Determining an estimated click rate of the user for the object to be recommended based on the object to be recommended and the preference feature representation;

[0030] Determine whether to recommend the object to be recommended to the user based on the estimated click rate of the object to be recommended by the user.

[0031] Optionally, the different behaviors include at least two of clicking behaviors, liking behaviors, browsing behaviors and purchasing behaviors.

[0032] According to a second aspect of an embodiment of the present disclosure, there is provided an object recommendation device, comprising:

[0033] An acquisition module, used to acquire historical object sequences and interaction features corresponding to different behaviors performed by the user, wherein the historical object sequence corresponding to each behavior includes historical objects arranged in sequence according to the execution time of the behavior, and the interaction feature corresponding to each behavior is used to indicate the user's preference for the behavior;

[0034] A first determination module is used to obtain the user's preference feature representation based on the attention mechanism according to the historical object sequences and the interaction features corresponding to all behaviors;

[0035] The recommendation module is used to determine whether to recommend the object to be recommended to the user based on the object to be recommended and the preference feature representation.

[0036] Optionally, the first determining module includes:

[0037] An encoding submodule, used for encoding each of the historical object sequences to obtain a historical object sequence vector corresponding to each of the historical object sequences;

[0038] A generating submodule, used for generating a plurality of virtual view vectors corresponding to each of the historical object sequences;

[0039] A first determination submodule is configured to generate a plurality of target virtual view vectors corresponding to each of the historical object sequences based on an attention mechanism according to the historical object sequence vectors and the plurality of virtual view vectors respectively corresponding to each of the historical object sequences;

[0040] The second determination submodule is used to obtain the user's preference feature representation based on the attention mechanism according to all the target virtual view vectors, all the virtual view vectors and the interaction features.

[0041] Optionally, the first determination submodule is specifically used to input, for each of the historical object sequences, multiple virtual view vectors corresponding to the historical object sequence and the historical object sequence vector into the first attention layer, to obtain a weighted representation corresponding to each virtual view vector corresponding to the historical object sequence, and the weighted representation corresponding to each virtual view vector is used to characterize the target virtual view vector corresponding to the virtual view vector.

[0042] Optionally, the first determining submodule is specifically configured to calculate a weighted representation corresponding to each virtual view vector by using the following formula:

[0043]

[0044]

[0045] K 1 =W K *X;

[0046] V 1 =W V *X;

[0047] Among them, out 1A weighted representation for representing each virtual view vector corresponding to the historical object sequence, is used to represent multiple virtual view vectors corresponding to the historical object sequence, X is used to represent the historical object sequence vector, d k Used to represent the scaled factor, T represents the transposed symbol, W Q Used to represent the first weight matrix, W K represents the second weight matrix, W V represents the third weight matrix, softmax() is the normalized exponential function, K 1 is used to represent the weighted result of weighting the historical object sequence vector corresponding to the historical object sequence based on the second weight matrix, V 1 Used to represent a weighted result of weighting the historical object sequence vector corresponding to the historical object sequence based on the third weight matrix.

[0048] Optionally, the second determination submodule is specifically used to input all the target virtual view vectors, all the virtual view vectors and the interaction features into a second attention layer to obtain a preference feature representation of the user.

[0049] Optionally, the second determination submodule is specifically configured to calculate the user's preference characteristic representation by the following formula:

[0050]

[0051] Among them, out 2 The feature representation used to characterize the user’s preference, Q 2 Used to characterize the interaction features, K 2 The virtual view vector V is used to represent all the targets. 2 Used to represent all the virtual view vectors, T represents the transposed symbol, d k Used to characterize the scaled factor, softmax() is a normalized exponential function.

[0052] Optionally, the recommendation module includes:

[0053] A third determination submodule is used to determine the estimated click rate of the user for the object to be recommended based on the object to be recommended and the preference feature representation;

[0054] The fourth recommendation submodule is used to determine whether to recommend the object to be recommended to the user according to the estimated click rate of the user for the object to be recommended.

[0055] Optionally, the different behaviors include at least two of clicking behaviors, liking behaviors, browsing behaviors and purchasing behaviors.

[0056] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device, including:

[0057] processor;

[0058] a memory for storing processor-executable instructions;

[0059] The processor is configured to execute the steps of the object recommendation method provided in the first aspect of the present disclosure.

[0060] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which computer program instructions are stored. When the program instructions are executed by a processor, the steps of the object recommendation method provided in the first aspect of the present disclosure are implemented.

[0061] The technical solution provided by the embodiments of the present disclosure may include the following beneficial effects: since the user's preference feature representation is determined based on the historical object sequences and interaction features corresponding to different behaviors, and since the interaction features corresponding to each behavior are used to represent the user's preference for the behavior, the obtained user's preference features focus on the user's preference for different behaviors, which can improve the feature expression ability of the preference feature representation, thereby improving the recommendation effect based on the preference feature representation.

[0062] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0064] Figure 1 The figure is a flowchart of an object recommendation method according to an exemplary embodiment.

[0065] Figure 2 The diagram is a model architecture diagram for modeling user preference feature representation according to an exemplary embodiment.

[0066] Figure 3 The figure is a block diagram of a device for recommending an object according to an exemplary embodiment.

[0067] Figure 4 It is a block diagram of an electronic device according to an exemplary embodiment.

[0068] Figure 5 is a block diagram of another electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0069] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0070] It should be noted that all actions of acquiring signals, information or data in the present disclosure are carried out in compliance with the relevant data protection laws and policies of the country where the device is located and with the authorization given by the owner of the corresponding device.

[0071] In the related art, the multi-sequence modeling method is to add the modeling results of each single sequence and then take the average of the added results. Although this method is simple, a lot of information is lost in the process of upper-level feature extraction. In addition, users have certain preferences for various behaviors, and the above multi-sequence modeling method treats all behaviors equally, which is obviously unreasonable and will also affect the accuracy of object recommendation based on modeling results.

[0072] In view of this, the embodiments of the present disclosure provide an object recommendation method, device, electronic device and storage medium to solve the problem of low recommendation accuracy.

[0073] The technical solution of the present disclosure is described in detail below with reference to embodiments.

[0074] Figure 1 is a flowchart of an object recommendation method according to an exemplary embodiment. Figure 1 As shown, the object recommendation method can be used in electronic devices, and the object recommendation method can be applied to algorithm recommendation scenarios such as APP, shopping cart, product station, etc. Figure 1 , the object recommendation method may include the following steps.

[0075] In step S11, historical object sequences and interaction features corresponding to different behaviors performed by the user are obtained, wherein the historical object sequence corresponding to each behavior includes historical objects arranged in sequence according to the execution time of the behavior, and the interaction feature corresponding to each behavior is used to indicate the user's preference for the behavior.

[0076] The different behaviors may include at least two of click behaviors, like behaviors, browsing behaviors, and purchase behaviors. Correspondingly, the historical object sequence may be a historical object sequence corresponding to the click behavior, and the historical objects correspond to the clicked objects; the historical object sequence may be a historical object sequence corresponding to the like behavior, and the historical objects correspond to the liked objects; the historical object sequence may be a historical object sequence corresponding to the browsing behavior, and the historical objects correspond to the browsed objects; the historical object sequence may be a historical object sequence corresponding to the purchase behavior, and the historical objects correspond to the purchased objects.

[0077] Among them, historical objects can be commodities, advertisements, videos, audios, etc.

[0078] The interaction feature may be the number of times the user performs an action on a historical object. The greater the number of times, the more the user prefers such an action.

[0079] In step S12, according to the historical object sequences and interaction features corresponding to all behaviors, the user's preference feature representation is obtained based on the attention mechanism.

[0080] First, the encoding result obtained by encoding the historical object sequence corresponding to all behaviors can be regarded as the preference representation of each user behavior. The encoding operation refers to extracting feature information from the original data. These feature information are represented by embedded vectors. The encoding operation can also be understood as obtaining a low-dimensional embedded vector through neural network mapping. Then, the attention mechanism is introduced in the calculation of the user's preference feature representation. That is, through the attention mechanism, different weights are assigned to each behavior representation. For the specific implementation method of obtaining the user's preference feature representation based on the attention mechanism, please refer to the following embodiment, which will not be described in detail in this embodiment.

[0081] In step S13, it is determined whether to recommend the object to be recommended to the user based on the object to be recommended and the preference feature representation.

[0082] As an example, the above-mentioned step of determining whether to recommend the object to be recommended to the user based on the object to be recommended and the preference feature representation may include: determining the user's estimated click rate for the recommended object based on the object to be recommended and the preference feature representation; determining whether to recommend the object to be recommended to the user based on the user's estimated click rate for the recommended object. Wherein, in the scenario of predicting the click rate, the object to be recommended and the preference feature representation are input into the fully connected layer, so that the user's estimated click probability for the recommended object can be obtained. When the estimated click probability is greater than a preset probability threshold, it is determined to recommend the object to be recommended to the user; when the estimated click probability is less than or equal to the preset probability threshold, it is determined not to recommend the object to be recommended to the user.

[0083] Through the above method, since the user's preference feature representation is determined based on the historical object sequences and interaction features corresponding to different behaviors, and since the interaction features corresponding to each behavior are used to represent the user's preference for the behavior, the user's preference features obtained focus on the user's preference for different behaviors, which can improve the feature expression ability of the preference feature representation, thereby improving the recommendation effect based on the preference feature representation.

[0084] In some embodiments, the implementation method of the above-mentioned step S12 may include: encoding each historical object sequence to obtain a historical object sequence vector corresponding to each historical object sequence; generating multiple virtual view vectors corresponding to each historical object sequence; generating multiple target virtual view vectors corresponding to each historical object sequence based on an attention mechanism according to the historical object sequence vector and multiple virtual view vectors corresponding to each historical object sequence; and obtaining a user's preference feature representation based on an attention mechanism according to all target virtual view vectors, all virtual view vectors and interaction features.

[0085] The historical object sequence vector can be obtained by encoding the historical object sequence.

[0086] For each historical object sequence, the historical object sequence can be depicted from different perspectives, thereby obtaining different virtual view vectors, and the perspectives of the virtual view vectors depicted by different historical object sequences can be the same. As an example, the multiple virtual view vectors can be three.

[0087] Among them, for the historical object sequence, the target virtual view vectors corresponding to different virtual view vectors are virtual view vectors after the corresponding virtual view vectors are weighted. Taking the multiple virtual view vectors as virtual view vector 1, virtual view vector 2 and virtual view vector 3 as an example, the target virtual view vectors corresponding to virtual view vector 1, virtual view vector 2 and virtual view vector 3 are target virtual view vector 1, target virtual view vector 2 and target virtual view vector 3 respectively, and the weighted virtual view vector, that is, target virtual view vector 1, can reflect the relationship between virtual view vector 2 and virtual view vector 3 and virtual view vector 1.

[0088] As an example, the above-mentioned step of generating multiple target virtual view vectors corresponding to each historical object sequence based on the attention mechanism according to the historical object sequence vectors and multiple virtual view vectors corresponding to each historical object sequence can be implemented in the following manner: for each historical object sequence, the multiple virtual view vectors and the historical object sequence vector corresponding to the historical object sequence are input into the first attention layer to obtain the weighted representation corresponding to each virtual view vector corresponding to the historical object sequence, and the weighted representation corresponding to each virtual view vector is used to characterize the target virtual view vector corresponding to the virtual view vector. The network structure of the first attention layer can refer to the relevant technology, and this embodiment will not be described in detail here.

[0089] Specifically, the weighted representation corresponding to each virtual view vector can be calculated by the following formula (1):

[0090]

[0091]

[0092] K 1 =W K *X;

[0093] V 1 =W V *X;

[0094] Among them, in the above formula (1), out 1 It is used to represent the target virtual view vector corresponding to each virtual view vector corresponding to the historical object sequence. It is used to represent multiple virtual view vectors corresponding to the historical object sequence, X is used to represent the historical object sequence vector, d k It is used to represent the scaled factor, which is used to make the data conform to the distribution with mean 0 and variance 1. T represents the transposed symbol, and W Q Used to represent the first weight matrix, W K represents the second weight matrix, W V represents the third weight matrix, softmax() is a normalized exponential function, where k is greater than or equal to 2, where Q 1 It is used to represent the weighted result of weighting multiple virtual view vectors corresponding to the historical object sequence based on the first weight matrix, K 1 It is used to represent the weighted result of weighting the historical object sequence vector corresponding to the historical object sequence based on the second weight matrix, V 1 Used to represent a weighted result of weighting the historical object sequence vector corresponding to the historical object sequence based on the third weight matrix.

[0095] As an example, the above step of obtaining the user's preference feature representation based on the attention mechanism according to all target virtual view vectors, all virtual view vectors and interaction features can be implemented in the following manner: all target virtual view vectors, all virtual view vectors and interaction features are input into the second attention layer to obtain the user's preference feature representation. The network structure of the second attention layer can refer to the relevant technology, and this embodiment will not be described in detail here.

[0096] Specifically, the user's preference feature representation can be calculated by the following formula (2):

[0097]

[0098] Among them, out 2 It is used to characterize the user’s preference features, Q 2 Used to characterize the interaction features, K 2 Used to represent all target virtual view vectors, V 2 Used to represent all virtual view vectors, T represents the transposed symbol, d k Used to represent the scaled factor. The scaled factor is used to make the data conform to the distribution of 0 mean and variance 1. Softmax() is a normalized exponential function.

[0099] It is worth noting that the virtual view vector is used when calculating each target virtual view vector and when calculating the user's preference feature representation. Therefore, the virtual view vector can better represent the relationship between the user, historical objects and the interaction features between the user and historical objects, reduce information loss, and help improve the feature expression ability of the user's preference feature representation, thereby helping to improve the recommendation effect based on the user's preference feature representation.

[0100] Figure 2 is a model architecture diagram for modeling user preference feature representation according to an exemplary embodiment. Figure 2 , behavior sequence 1 to behavior sequence N are historical object sequences corresponding to different behaviors, N is greater than or equal to 3, and each behavior sequence includes M items, and item can be understood as a historical object. Figure 2 View in Figure 1 ,See Figure 2 And Vision Figure 3 They represent different target virtual view vectors respectively, and the view latent vector is used to represent the virtual view vector.

[0101] First, the behavior sequence vector in the input layer is encoded using the behavior encoding module to obtain an encoded vector, namely, a historical object sequence vector. The behavior sequence vector in the input layer is obtained by encoding the historical object sequence.

[0102] Next, the multi-behavior view extraction module is used to generate three virtual view vectors corresponding to the historical object sequence, and the multi-behavior view extraction module is used to generate three target virtual view vectors corresponding to each historical object sequence based on the attention mechanism according to the historical object sequence vector and the three virtual view vectors corresponding to each historical object sequence.

[0103] Next, the view aggregation module is used to obtain the user's preference feature representation based on the attention mechanism according to the three target virtual view vectors corresponding to all historical object sequences, all interaction features and the three virtual view vectors corresponding to all historical object sequences.

[0104] Finally, the fully connected layer is used to process based on the user's preference feature representation and the object features of the object to be recommended, so as to obtain the estimated click rate of the object to be recommended predicted by the fully connected layer. From the above, it can be seen that the step of obtaining the user's preference feature representation based on the attention mechanism according to the historical object sequence and interaction features corresponding to all behaviors, and determining whether to recommend the object to be recommended to the user according to the object to be recommended and the preference feature representation can be implemented through the input layer, the behavior encoding module, the multi-behavior view extraction module, the view aggregation module and the fully connected layer. The architecture composed of the input layer, the behavior encoding module, the multi-behavior view extraction module, the view aggregation module and the fully connected layer is referred to as the recommendation model below, that is, the historical object sequences and interaction features corresponding to different behaviors performed by the user can be input into the recommendation model to obtain the estimated click rate of the object to be recommended.

[0105] When training the recommendation model, the sample historical object sequences and sample interaction features corresponding to different behaviors performed by historical users are input into the recommendation model to be trained to obtain the preference feature representation of historical users. Then, based on the recommendation model to be trained and the preference feature representation of historical users, the estimated click-through rate of the sample recommended objects is obtained. Based on the sample click-through rate and the estimated click-through rate, the network parameters of the recommendation model are adjusted until the training conditions are met to obtain a trained recommendation model.

[0106] Figure 3 is a block diagram of an object recommendation device according to an exemplary embodiment. Figure 3 , the device 300 comprises:

[0107] The acquisition module 301 is used to acquire the historical object sequences and interaction features corresponding to different behaviors performed by the user, wherein the historical object sequence corresponding to each behavior includes historical objects arranged in sequence according to the execution time of the behavior, and the interaction feature corresponding to each behavior is used to indicate the user's preference for the behavior;

[0108] A first determination module 302 is used to obtain the user's preference feature representation based on the attention mechanism according to the historical object sequence and the interaction features corresponding to all behaviors;

[0109] The recommendation module 303 is used to determine whether to recommend the object to be recommended to the user according to the object to be recommended and the preference feature representation.

[0110] Optionally, the first determining module 302 includes:

[0111] An encoding submodule, used for encoding each of the historical object sequences to obtain a historical object sequence vector corresponding to each of the historical object sequences;

[0112] A generating submodule, used for generating a plurality of virtual view vectors corresponding to each of the historical object sequences;

[0113] A first determination submodule is configured to generate a plurality of target virtual view vectors corresponding to each of the historical object sequences based on an attention mechanism according to the historical object sequence vectors and the plurality of virtual view vectors respectively corresponding to each of the historical object sequences;

[0114] The second determination submodule is used to obtain the user's preference feature representation based on the attention mechanism according to all the target virtual view vectors, all the virtual view vectors and the interaction features.

[0115] Optionally, the first determination submodule is specifically used to input, for each of the historical object sequences, multiple virtual view vectors corresponding to the historical object sequence and the historical object sequence vector into the first attention layer, to obtain a weighted representation corresponding to each virtual view vector corresponding to the historical object sequence, and the weighted representation corresponding to each virtual view vector is used to characterize the target virtual view vector corresponding to the virtual view vector.

[0116] Optionally, the first determining submodule is specifically configured to calculate the weighted representation corresponding to each virtual view vector by using the following formula:

[0117]

[0118]

[0119] K 1 =W K *X;

[0120] V 1 =W V *X;

[0121] Among them, out 1A weighted representation for representing each virtual view vector corresponding to the historical object sequence, is used to represent multiple virtual view vectors corresponding to the historical object sequence, X is used to represent the historical object sequence vector, d k Used to represent the scaled factor, T represents the transposed symbol, W Q Used to represent the first weight matrix, W K represents the second weight matrix, W V represents the third weight matrix, softmax() is the normalized exponential function, K 1 is used to represent the weighted result of weighting the historical object sequence vector corresponding to the historical object sequence based on the second weight matrix, V 1 Used to represent a weighted result of weighting the historical object sequence vector corresponding to the historical object sequence based on the third weight matrix.

[0122] Optionally, the second determination submodule is specifically used to input all the target virtual view vectors, all the virtual view vectors and the interaction features into a second attention layer to obtain a preference feature representation of the user.

[0123] Optionally, the second determination submodule is specifically configured to calculate the user's preference characteristic representation by the following formula:

[0124]

[0125] Among them, out 2 The feature representation used to characterize the user’s preference, Q 2 Used to characterize the interaction features, K 2 The virtual view vector V is used to represent all the targets. 2 Used to represent all the virtual view vectors, T represents the transposed symbol, d k Used to characterize the scaled factor, softmax() is a normalized exponential function.

[0126] Optionally, the recommendation module 303 includes:

[0127] A third determination submodule is used to determine the estimated click rate of the user for the object to be recommended based on the object to be recommended and the preference feature representation;

[0128] The fourth recommendation submodule is used to determine whether to recommend the object to be recommended to the user according to the estimated click rate of the user for the object to be recommended.

[0129] Optionally, the different behaviors include at least two of clicking behaviors, liking behaviors, browsing behaviors and purchasing behaviors.

[0130] Regarding the device 300 in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0131] Based on the same inventive concept, according to an embodiment of the present disclosure, a computer-readable storage medium is further provided, on which computer program instructions are stored. When the program instructions are executed by a processor, the steps of the object recommendation method provided in the present disclosure are implemented.

[0132] Based on the same inventive concept, the present disclosure also provides an electronic device, including:

[0133] processor;

[0134] a memory for storing processor-executable instructions;

[0135] The processor is configured to execute the steps of the object recommendation method provided above in the present disclosure.

[0136] Figure 4 4 is a block diagram of an electronic device according to an exemplary embodiment. For example, the electronic device 400 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, a server, etc.

[0137] Reference Figure 4 , the electronic device 400 may include one or more of the following components: a processing component 402 , a memory 404 , a power component 406 , a multimedia component 408 , an audio component 410 , an input / output interface 412 , a sensor component 414 , and a communication component 416 .

[0138] The processing component 402 generally controls the overall operation of the electronic device 400, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 402 may include one or more processors 420 to execute instructions to complete all or part of the steps of the above-mentioned object recommendation method. In addition, the processing component 402 may include one or more modules to facilitate the interaction between the processing component 402 and other components. For example, the processing component 402 may include a multimedia module to facilitate the interaction between the multimedia component 408 and the processing component 402.

[0139] The memory 404 is configured to store various types of data to support operations on the electronic device 400. Examples of such data include instructions for any application or method operating on the electronic device 400, contact data, phone book data, messages, pictures, videos, etc. The memory 404 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0140] The power supply component 406 provides power to the various components of the electronic device 400. The power supply component 406 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the electronic device 400.

[0141] The multimedia component 408 includes a screen that provides an output interface between the electronic device 400 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 408 includes a front camera and / or a rear camera. When the electronic device 400 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and the rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0142] The audio component 410 is configured to output and / or input audio signals. For example, the audio component 410 includes a microphone (MIC), and when the electronic device 400 is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 404 or sent via the communication component 416. In some embodiments, the audio component 410 also includes a speaker for outputting audio signals.

[0143] The input / output interface 412 provides an interface between the processing component 402 and the peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: a home button, a volume button, a start button, and a lock button.

[0144] The sensor assembly 414 includes one or more sensors for providing various aspects of status assessment for the electronic device 400. For example, the sensor assembly 414 can detect the open / closed state of the electronic device 400, the relative positioning of the components, such as the display and keypad of the electronic device 400, and the sensor assembly 414 can also detect the position change of the electronic device 400 or a component of the electronic device 400, the presence or absence of contact between the user and the electronic device 400, the orientation or acceleration / deceleration of the electronic device 400, and the temperature change of the electronic device 400. The sensor assembly 414 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 414 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 414 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0145] The communication component 416 is configured to facilitate wired or wireless communication between the device 400 and other devices. The electronic device 400 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 416 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 416 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0146] In an exemplary embodiment, the electronic device 400 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the steps of the above-mentioned object recommendation method.

[0147] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 404 including instructions, and the instructions can be executed by a processor 420 of an electronic device 400 to complete the object recommendation method described above. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

[0148] Figure 55 is a block diagram of another electronic device according to an exemplary embodiment. For example, the electronic device 500 may be provided as a server. Figure 5 The electronic device 500 includes a processing component 522, which further includes one or more processors, and a memory resource represented by a memory 532 for storing instructions executable by the processing component 522, such as an application. The application stored in the memory 532 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 522 is configured to execute instructions to perform all steps of the above-mentioned object recommendation method.

[0149] The electronic device 500 may also include a power supply component 526 configured to perform power management of the electronic device 500, a wired or wireless network interface 550 configured to connect the electronic device 500 to a network, and an input / output interface 558. The electronic device 500 may operate based on an operating system stored in the memory 532, such as Windows Server 2003. TM , Mac OS X TM , Unix TM , Linux TM , FreeBSD TM or similar.

[0150] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the present disclosure. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered as exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0151] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. An object recommendation method, characterized in that: include: Obtaining historical object sequences and interaction features corresponding to different behaviors performed by the user, wherein the historical object sequence corresponding to each behavior includes historical objects arranged in sequence according to the execution time of the behavior, and the interaction feature corresponding to each behavior is used to indicate the user's preference for the behavior; According to the historical object sequences and the interaction features corresponding to all behaviors, obtaining the preference feature representation of the user based on the attention mechanism; Determine whether to recommend the object to be recommended to the user based on the object to be recommended and the preference feature representation.

2. The method according to claim 1, characterized in that The step of obtaining the user's preference feature representation based on the attention mechanism according to the historical object sequences and the interaction features corresponding to all behaviors includes: Encoding each of the historical object sequences to obtain a historical object sequence vector corresponding to each of the historical object sequences; generating a plurality of virtual view vectors respectively corresponding to each of the historical object sequences; Generate a plurality of target virtual view vectors corresponding to each of the historical object sequences based on an attention mechanism according to the historical object sequence vectors and the plurality of virtual view vectors respectively corresponding to each of the historical object sequences; According to all the target virtual view vectors, all the virtual view vectors and the interaction features, a preference feature representation of the user is obtained based on an attention mechanism.

3. The method according to claim 2, characterized in that The step of generating a plurality of target virtual view vectors corresponding to each of the historical object sequences based on an attention mechanism according to the historical object sequence vectors respectively corresponding to each of the historical object sequences and the plurality of virtual view vectors comprises: For each of the historical object sequences, multiple virtual view vectors corresponding to the historical object sequence and the historical object sequence vector are input into the first attention layer to obtain a weighted representation corresponding to each virtual view vector corresponding to the historical object sequence. The weighted representation corresponding to each of the virtual view vectors is used to represent the target virtual view vector corresponding to the virtual view vector.

4. The method according to claim 3, characterized in that For each of the historical object sequences, a plurality of virtual view vectors corresponding to the historical object sequence and the historical object sequence vector are input into the first attention layer to obtain a weighted representation corresponding to each virtual view vector corresponding to the historical object sequence, including: The weighted representation corresponding to each virtual view vector is calculated by the following formula: K1=W K *X; V1=W V *X; Among them, out1 is used to represent the weighted representation corresponding to each virtual view vector corresponding to the historical object sequence. is used to represent multiple virtual view vectors corresponding to the historical object sequence, X is used to represent the historical object sequence vector, d k Used to represent the scaled factor, T represents the transposed symbol, W Q Used to represent the first weight matrix, W K represents the second weight matrix, W V represents the third weight matrix, softmax() is a normalized exponential function, K1 is used to characterize the weighted result of weighting the historical object sequence vector corresponding to the historical object sequence based on the second weight matrix, and V1 is used to characterize the weighted result of weighting the historical object sequence vector corresponding to the historical object sequence based on the third weight matrix.

5. The method according to claim 4, characterized in that The step of obtaining the user's preference feature representation based on an attention mechanism according to all the target virtual view vectors, all the virtual view vectors and the interaction features includes: All the target virtual view vectors, all the virtual view vectors and the interaction features are input into a second attention layer to obtain a preference feature representation of the user.

6. The method according to claim 5, characterized in that The step of inputting all the target virtual view vectors, all the virtual view vectors and the interaction features into the second attention layer to obtain the user's preference feature representation includes: The user's preference feature representation is calculated by the following formula: Among them, out2 is used to represent the user's preference feature representation, Q2 is used to represent the interaction feature, K2 is used to represent all the target virtual view vectors, V2 is used to represent all the virtual view vectors, T represents the transposition symbol, d k Used to characterize the scaled factor, softmax() is a normalized exponential function.

7. The method according to claim 1, characterized in that The step of determining whether to recommend the object to be recommended to the user based on the object to be recommended and the preference feature representation includes: Determining an estimated click rate of the user for the object to be recommended based on the object to be recommended and the preference feature representation; Determine whether to recommend the object to be recommended to the user based on the estimated click rate of the object to be recommended by the user.

8. The method according to claim 1, characterized in that The different behaviors include at least two of clicking behaviors, liking behaviors, browsing behaviors, and purchasing behaviors.

9. An object recommendation device, characterized in that: include: An acquisition module, used to acquire historical object sequences and interaction features corresponding to different behaviors performed by the user, wherein the historical object sequence corresponding to each behavior includes historical objects arranged in sequence according to the execution time of the behavior, and the interaction feature corresponding to each behavior is used to indicate the user's preference for the behavior; A first determination module is used to obtain the user's preference feature representation based on the attention mechanism according to the historical object sequences and the interaction features corresponding to all behaviors; The recommendation module is used to determine whether to recommend the object to be recommended to the user based on the object to be recommended and the preference feature representation.

10. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to execute the steps of any one of the methods described in claims 1-8.

11. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the steps of the method described in any one of claims 1 to 8 are implemented.

Images