Resource recommendation method and device, equipment, storage medium and product
By using public domain and specific domain models in the recommendation system, combining the activity level and attribute information of the target object, the problem of inaccurate content recommendation in the recommendation system is solved, more accurate resource recommendation is achieved, and user consumption time and penetration rate are improved.
Patent Information
- Application Number
- CN202510263074.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-06
- Publication Date
- 2025-07-04
AI Technical Summary
The existing recommendation system cannot effectively learn the rules of data distribution of different groups, resulting in inaccurate content recommendations.
By inputting data into the common domain and specific domain models based on the activity level and attribute information of the target object, the first and second predicted values are obtained respectively, and the target predicted values are determined after fusing, thereby recommending network resources that meet the needs of the target object.
It improves the accuracy of recommended content and enhances the consumption duration, reading volume and consumption penetration of the object.
Smart Images
Figure CN120256718A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of computer technologies, and particularly to a resource recommendation method, apparatus, device, storage medium, and product. Background Art
[0002] Currently, recommendation systems on the market screen out the content needed by users from a vast amount of candidate content according to user interests. To provide accurate content recommendations to users, the classification of user groups will be optimized, different candidate content will be explored for different types of groups, and then target information will be recommended based on these candidate content.
[0003] Currently, it is impossible to train and learn the laws of data distribution for different groups, resulting in inaccurate final content recommendations. Summary of the Invention
[0004] In view of this, the present disclosure provides a resource recommendation method, apparatus, device, storage medium, and product to solve the problem of inaccurate content recommendation in related technologies.
[0005] In a first aspect, the present disclosure provides a resource recommendation method, and the method includes:
[0006] Determine target data based on the target activity level of a target object corresponding to a first network resource and the target attribute information of the target object;
[0007] Input the target data into the common domain of a target model to obtain a first predicted value corresponding to the target activity level, where the common domain is constructed based on the attribute information of a first object and the corresponding activity level;
[0008] Input the target data into the specific domain of the target model to obtain a second predicted value corresponding to the target activity level, where the specific domain is constructed based on the attribute information of a second object and the corresponding activity level, and the first object includes the second object;
[0009] Determine the target predicted value of the target object corresponding to the first network resource based on the first predicted value and the second predicted value;
[0010] Obtain a recommended second network resource from the first network resource based on the target predicted value.
[0011] In an embodiment of the present disclosure, based on the target activity level of the target object corresponding to the first network resource and the target attribute information of the target object, these data are input into the common domain to obtain a first prediction value corresponding to the target activity level, and these data are input into the specific domain to obtain a second prediction value corresponding to the target activity level. According to the first prediction value and the second prediction value, a target prediction value of the target object corresponding to the first network resource is obtained. Then, based on the target prediction value, a recommended second network resource is obtained from the first network resource. In this way, the embodiment of the present disclosure can learn the rules of the data distribution of group objects with different activity levels corresponding to network resources (these rules are reflected by the target prediction value), and obtain the recommended second network resource based on these rules, so that the recommended second network resource meets the real needs of the target object, improves the accuracy of the recommended content, and improves the consumption duration, reading volume, consumption days, and consumption penetration rate of objects in the recommendation scenario.
[0012] In a second aspect, the present disclosure provides a resource recommendation device, which includes:
[0013] A first determination module, configured to determine target data based on the target activity level of the target object corresponding to the first network resource and the target attribute information of the target object;
[0014] A first obtaining module, configured to input the target data into the common domain of the target model to obtain a first prediction value corresponding to the target activity level, where the common domain is constructed based on the attribute information of the first object and the corresponding activity level;
[0015] A second obtaining module, configured to input the target data into the specific domain of the target model to obtain a second prediction value corresponding to the target activity level, where the specific domain is constructed based on the attribute information of the second object and the corresponding activity level, and the first object includes the second object;
[0016] A second determination module, configured to determine a target prediction value of the target object corresponding to the first network resource based on the first prediction value and the second prediction value;
[0017] An obtaining module, configured to obtain a recommended second network resource from the first network resource based on the target prediction value.
[0018] In a third aspect, the present disclosure provides a computer device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the resource recommendation method in the first aspect or any corresponding embodiment thereof.
[0019] Fourthly, the present disclosure provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to execute the resource recommendation method according to the first aspect or any corresponding embodiment thereof.
[0020] Fifthly, the present disclosure provides a computer program product, including computer instructions, and the computer instructions are used to cause a computer to execute the resource recommendation method according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present disclosure. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.
[0022] Figure 1 is a flowchart of the resource recommendation method according to an embodiment of the present disclosure;
[0023] Figure 2 is a schematic diagram of the optimized model network structure of the resource recommendation method according to an embodiment of the present disclosure;
[0024] Figure 3 is a flowchart of the resource recommendation method according to another embodiment of the present disclosure;
[0025] Figure 4 is a block diagram of the structure of the resource recommendation device according to an embodiment of the present disclosure;
[0026] Figure 5 is a schematic diagram of the hardware structure of the computer device according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0027] In order to make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present disclosure with reference to the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are some, but not all, of the embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present disclosure.
[0028] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not used to limit the protection scope of the present disclosure.
[0029] In the description of the embodiments of the present disclosure, the term "including" and its like should be understood as an open inclusion, that is, "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The term "some embodiments" should be understood as "at least some embodiments". There may also be other explicit and implicit definitions hereinafter.
[0030] In the present disclosure, unless expressly stated, performing a step "in response to A" does not mean that the step is immediately performed after "A", but may include one or more intermediate steps.
[0031] It can be understood that the data involved in the technical solution (including but not limited to the data itself, the acquisition, use, storage, or deletion of the data) should comply with the requirements of the corresponding laws, regulations, and related provisions.
[0032] It can be understood that before using the technical solutions disclosed in the embodiments of the present disclosure, the type, scope of use, usage scenarios, etc. of the information involved in the present disclosure should be informed to the relevant users and the authorization of the relevant users should be obtained through appropriate means in accordance with the relevant laws and regulations. Among them, the relevant users may include any type of right subject, such as an individual, an enterprise, or a group.
[0033] For example, when a user's active request is received, a prompt message is sent to the relevant user to clearly prompt the relevant user that the operation requested by the user will require obtaining and using the information of the relevant user, so that the relevant user can autonomously choose whether to provide information to software or hardware such as an electronic device, an application program, a server, or a storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.
[0034] As an optional but non-limiting implementation manner, the manner of sending a prompt message to the relevant user in response to receiving the active request of the relevant user may be, for example, in the form of a pop-up window, and the prompt message may be presented in text in the pop-up window. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide information to the electronic device.
[0035] It should be understood that the above-mentioned notification and the process of obtaining user authorization are only illustrative and do not limit the implementation manner of the present disclosure. Other manners that comply with relevant laws and regulations can also be applied to the implementation manner of the present disclosure.
[0036] Currently, on the market, a recommendation system is used to distribute resource content that meets the personal interests of users, such as pushing military resources, entertainment resources, shopping resources, etc.
[0037] The currently common recommendation technology is to use a ranking model (especially a fine-ranking model) to calculate the interest scores of content for users. The recommendation system screens out a batch of content with the highest probability of user interest from a large number of content candidates according to the interest scores and presents it to users. Currently, in order to give users accurate content recommendations, the classification of user groups will be optimized, but the existing exploration directions mainly focus on the application of strategies. The population classification at the strategy application level cannot train and learn the laws of data distribution, resulting in inaccurate content recommendations. To solve the above problems, according to an embodiment of the present disclosure, an embodiment of a resource recommendation method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.
[0038] In this embodiment, a resource recommendation method is provided. Figure 1 It is a flowchart of a resource recommendation method according to an embodiment of the present disclosure, as Figure 1 shown. This method can be applied to the server side, and the method process includes the following steps:
[0039] Step S101, determine target data based on the target activity level of the target object corresponding to the first network resource and the target attribute information of the target object.
[0040] Optionally, in an embodiment of the present disclosure, the server can obtain in real time the target activity level of the target object corresponding to the first network resource that is currently to be pushed with resources. For example, the target activity level corresponding to when the target object consumes (such as browses, clicks, comments, likes, etc.) the first network resource. Here, the first network resource is any consumable information, such as news content, text content, etc., and there can be multiple such first network resources. Here, the target object is the object for which resources are recommended in the current resource push scenario, and this object can also be understood as a user.
[0041] In addition, activity is used to represent the operation frequency or degree of interest of an object when consuming a certain network resource, and it can include multiple levels, such as a high-activity level, a low-activity level, etc. Since the activity level corresponding to an object when consuming a certain network resource can only be one type, the target activity level corresponding to the target object when consuming the first network resource can be a high-activity level or a low-activity level.
[0042] In addition to obtaining the target activity level of the above-mentioned target object, it is also necessary to obtain the target attribute information of the target object. Here, the target attribute information can be some attribute information representing the personal characteristics of the target object. These attribute information of personal characteristics are usually fixed values related to the target object and can represent the identity of the target object.
[0043] Then, combine the target activity level of the target object and the target attribute information to obtain target data. It should be noted that while obtaining the target activity level and target attribute information of the target object, environmental information can also be obtained, such as the currently used network (including 2.4G, 5G, wifi), the current network speed, etc.
[0044] Step S102: Input the target data into the common domain of the target model to obtain a first predicted value corresponding to the target activity level. Here, the common domain is constructed based on the attribute information of the first object and the corresponding activity level.
[0045] Optionally, as Figure 2 shown, a target model is set up. This target model is composed of a three-structure method of a common domain + a high-activity domain + a low-activity domain. Currently, the high-activity domain or the low-activity domain is called the specific domain of the target model.
[0046] It should be understood that for the common domain: all data of the first object (i.e., all objects) (i.e., the attribute information of the first object and all activity levels corresponding to the first object when consuming network resources) are used for training and updating, so that the common domain model has the ability to capture the information of common interest of all users, such as being interested in certain real-time hotspots.
[0047] If the current target activity level is a high-activity level, the server needs to input the target data of the target object into the common domain for forward calculation to obtain a first predicted value corresponding to the target activity level: pa0. If the current target activity level is a low-activity level, the server needs to input the target data of the target object into the common domain for forward calculation to obtain a first predicted value corresponding to the target activity level: pa1.
[0048] Step S103: Input the target data into the specific domain of the target model to obtain a second predicted value corresponding to the target activity level. The specific domain is constructed based on the attribute information of the second object and the corresponding activity level, and the first object includes the second object.
[0049] Optionally, for the high-activity domain: Only use the data of objects with a high activity level (i.e., the attribute information of objects with a high activity level and the high activity level corresponding when objects with a high activity level consume network resources) for training and updating, so that the high-activity domain model has the ability to capture the unique interesting information of the high-activity group beyond the common interesting information, such as certain special information, information with scattered focuses, information with a wide range of interests, etc. For the low-activity domain: Only use the data of objects with a low activity level (i.e., the attribute information of objects with a low activity level and the low activity level corresponding when objects with a low activity level consume network resources) for training and updating, so that the low-activity domain model has the ability to capture the unique interesting information of the low-activity group beyond the common interesting information, such as certain niche information, information with a relatively single interest, etc.
[0050] If the current target activity level is a high activity level, the server needs to input the target data of the target object into the specific domain (high-activity domain) corresponding to the target activity level for forward calculation to obtain a second predicted value corresponding to the target activity level: pb0. If the current target activity level is a low activity level, the server needs to input the target data of the target object into the specific domain (low-activity domain) corresponding to the target activity level for forward calculation to obtain a second predicted value corresponding to the target activity level: pb1.
[0051] It can be understood that after determining the target activity level of the target object when consuming the first network resource, the specific type of the specific domain (high-activity domain or low-activity domain) is also determined, that is, the activity level corresponding when the second object constructing the specific domain consumes network resources is the target activity level. The second object here is included in the first object, and the second object only refers to the object with the target activity level. The specific domain is constructed based on the attribute information of the second object and the activity level corresponding when the second object consumes network resources.
[0052] Step S104: Determine the target predicted value of the target object corresponding to the first network resource based on the first predicted value and the second predicted value.
[0053] Optionally, the server fuses the obtained first predicted value and the second predicted value. The fusion methods here include addition, multiplication, etc. After fusion, the target predicted value will be obtained, and this target predicted value is the final interest situation of the target object for the first network resource.
[0054] Step S104: Obtain the recommended second network resource from the first network resources based on the target prediction value.
[0055] Optionally, the currently obtained target prediction value corresponds to the final interest situation of the target object when consuming a certain first network resource. Since there can be multiple first network resources, the number of obtained target prediction values is also multiple. At this time, by comparing multiple target prediction values, the second network resource to be recommended is selected from multiple first network resources, and then the second network resource can be recommended to the target object.
[0056] In the embodiments of the present disclosure, based on the target activity level of the target object corresponding to the first network resource and the target attribute information of the target object, these data are input into the common domain to obtain a first prediction value corresponding to the target activity level, and these data are input into the specific domain to obtain a second prediction value corresponding to the target activity level. According to the first prediction value and the second prediction value, the target prediction value of the target object corresponding to the first network resource is obtained. Then, based on the target prediction value, the recommended second network resource is obtained from the first network resources. In this way, the embodiments of the present disclosure can learn the rules of the data distribution of group objects with different activity levels corresponding to network resources (these rules are reflected by the target prediction value), and obtain the recommended second network resource based on these rules, so that the recommended second network resource meets the real needs of the target object, improves the accuracy of the recommended content, and improves the consumption duration, reading volume, consumption days, and consumption penetration rate of the object in the recommendation scenario.
[0057] In some alternative embodiments, the method further includes:
[0058] Step a1: Obtain the log data of the target object within a preset time interval.
[0059] Optionally, the server obtains the log data of the target object within a preset time interval, such as the log data within the most recent 15 days, generates the activity level information based on the log data, and the update frequency is daily. That is, the target activity level of the target object on date T is generated by the log data within the date interval [T - 15, T - 1] according to a certain algorithm.
[0060] Step a2: Obtain the active features of the target object based on the log data.
[0061] Optionally, after obtaining the log data, the active features of the target object within the date range of [T-15, T-1] can be collected, including: consumption duration (such as the duration of browsing and clicking on the first network resource), action days (such as the total number of days of consuming the first network resource within 15 days), active days (such as the total number of days of opening but not actually consuming the first network resource within 15 days), and active days ratio (such as the ratio of active days to action days).
[0062] Step a3: Determine the corresponding target activity level based on the active features.
[0063] Optionally, after the extraction of the active features of the target object is completed, the corresponding target activity level can be obtained based on these active features. It should be understood that the active features are a kind of parameter for deriving the activity level, and there is a corresponding relationship between the active features and the activity level. Therefore, after the active features of the target object are determined, its corresponding target activity level is also determined.
[0064] In the embodiments of the present disclosure, by obtaining the log data of the target object within the preset time range, and then obtaining the active features based on the log data, the corresponding target activity level can be quickly determined based on the active features, which is convenient for subsequent learning and training of the data distribution rules of group objects with different activity levels when consuming network resources.
[0065] In some optional implementation manners, the above step a3 includes:
[0066] Determine the active value corresponding to the first network resource based on the active features;
[0067] Match the target active interval according to the active value to determine the corresponding target cluster;
[0068] Determine the corresponding target activity level according to the target cluster.
[0069] Optionally, before determining the corresponding target activity level based on the active features of the target object, the embodiments of the present disclosure need to first divide multiple activity level types based on the active features of all collected objects. Specifically, currently, a suitable clustering algorithm (such as KNN (K-Nearest Neighbor)) can be used to calculate the respective consumption values based on the active features of all objects. For example, weighted summation of each active feature can be performed, and clustering is carried out based on these consumption values. Each cluster corresponds to an activity level. For example, it is divided into 3 clusters. Currently, there are a total of: 1 to 3 activity levels. The larger the activity level number, the higher the activity of the object. It can be set that those with a consumption value above 10 are set as the high activity level with a level value of 3, those with a consumption value in the range of 1 - 9 are set as the low activity level with a level value of 2, and those who are daily active objects in the past 15 days but have a consumption value of 0 (no consumption value) are set as the level with a level value of 1. The specific activity level classification is as follows in the table:
[0070]
[0071] Therefore, after obtaining the active features of the target object, the corresponding consumption value is obtained according to the active features of the target object. According to the consumption value, the consumption interval corresponding to the above 3 clustering clusters is matched. After determining the target consumption interval where the consumption value is located, the target cluster where the consumption value is located can be determined, and then the corresponding target activity level can be determined based on the target cluster.
[0072] In the embodiments of the present disclosure, the consumption value is obtained through the activity features of the target object, then the consumption interval corresponding to the clustering cluster is matched based on the consumption value to obtain the target cluster, and then the corresponding target activity level is found based on the target cluster, so as to achieve the purpose of quickly obtaining the target activity level of the target object.
[0073] In some alternative embodiments, step S102 includes:
[0074] Step S1021, converting the target data into a feature vector.
[0075] Optionally, in the embodiments of the present disclosure, the target data can be preprocessed first, such as normalization processing, standardization processing, missing value processing, etc. Then the preprocessed data is converted into a feature vector.
[0076] Step S1022, mapping the feature vector to the latent vector space of the common domain to obtain the first feature latent vector.
[0077] Optionally, using the predefined feature sharing latent vector layer, the feature vector is further mapped to a latent vector space with a specific dimension. Here, it is mapped to the latent vector space of the common domain to obtain its corresponding representation in the latent vector space as the first feature latent vector.
[0078] Step S1023: Input the first feature hidden vector into the deep network structure layer of the common domain to obtain a first prediction value.
[0079] Optionally, as Figure 2 shown, take the first feature hidden vector obtained by the projection feature embedding module as the input and pass it to Figure 2 the input layer of the deep network structure layer module of the common domain in . The dimension of this input vector needs to match the dimension expected by the deep network structure layer. If not, dimension adjustment operations (such as fully connected layers for increasing or decreasing dimensions) may be required. Then, the deep network structure layer module performs forward calculations of the deep network based on these hidden vectors, and finally obtains the prediction output of the model to obtain a first prediction value.
[0080] In some alternative embodiments, the above step S103 includes: Step S1031: Convert the target data into a feature vector.
[0081] Optionally, in the embodiments of the present disclosure, the target data can be preprocessed first, such as normalization processing, standardization processing, missing value processing, etc. Then, the preprocessed data is converted into a feature vector.
[0082] Step S1032: Map the feature vector to the hidden vector space of a specific domain to obtain a second feature hidden vector.
[0083] Optionally, the process of mapping the feature vector to the hidden vector space of a specific domain to obtain a second feature hidden vector is the same as the previous step S1022, except that the hidden vector space of the common domain is changed to the hidden vector space of the specific domain, and then through the processing of the hidden vector space of the specific domain, a second feature hidden vector is obtained. See Figure 2 .
[0084] Step S1033: Input the second feature hidden vector into the deep network structure layer of the specific domain to obtain a second prediction value.
[0085] Optionally, the process of inputting the second feature hidden vector into the deep network structure layer of the specific domain to obtain a second prediction value is the same as the previous step S1023, except that the deep network structure layer of the common domain is changed to the deep network structure layer of the specific domain to obtain a second prediction value. See Figure 2 .
[0086] In some alternative embodiments, before step S1032, the method further includes:
[0087] Perform a matrix multiplication operation on the first feature hidden vector and the migration matrix to perform a projection transformation on the first feature hidden vector to obtain the hidden vector space of the specific domain, where the migration matrix is determined based on the dimension of the first feature hidden vector and the dimension of the hidden vector space of the specific domain.
[0088] Optionally, as Figure 2 , migration matrices are provided in both the high-activity domain and the low-activity domain. The role of the migration matrix is to perform matrix multiplication operations on the first eigen latent vector in the common domain with the two migration matrices respectively, so as to project and transform them into the high-activity domain latent vector space and the low-activity domain latent vector space respectively, making the latent vectors participating in the calculations in the high-activity domain and the low-activity domain after projection more adaptable to the vector spaces of their corresponding domains.
[0089] For example, if the target activity level of the target object is a high activity level, then only need to perform matrix multiplication operation on the migration matrix in the high-activity domain and the first eigen latent vector, so as to achieve the projection transformation of the first eigen latent vector and obtain the latent vector space of the high-activity domain.
[0090] The current migration matrix is determined based on the dimension of the first eigen latent vector and the dimension of the latent vector space in the high-activity domain. For example, if the dimension of the first eigen latent vector is n and the dimension of the latent vector space in the high-activity domain is m, then the migration matrix is an m×n matrix.
[0091] If the target activity level of the target object is a low activity level, then only need to perform matrix multiplication operation on the migration matrix in the low-activity domain and the first eigen latent vector, so as to achieve the projection transformation of the first eigen latent vector and obtain the latent vector space of the low-activity domain.
[0092] In the embodiments of the present disclosure, the projection transformation into the high-activity domain and the low-activity domain latent vector spaces is realized by performing matrix multiplication operations on the first eigen latent vector with the two migration matrices respectively, so that the information in the common domain can be utilized in different domains. While ensuring the effect of the three-structure model of the common domain + high-activity domain + low-activity domain, the model can share all the underlying latent vector parameters without additionally increasing the scale of the model parameters.
[0093] In some optional implementation manners, the above step S103 includes:
[0094] Obtain model hyperparameters, where the model hyperparameters are used to control the influence ratios of the common domain and the specific domain;
[0095] Fuse the model hyperparameters, the first prediction value, and the second prediction value to obtain a target prediction value.
[0096] Optionally, set model hyperparameters to control the proportion of the influence of the common domain and the specific domain (high-activity domain or low-activity domain). For example, when the target activity level is the high-activity level, for the high-activity domain and the common domain, the model hyperparameter can be set to alpha (the value range is (0, 1), which can be optimized according to the actual business, preferably 0.5). At this time, fuse the first predicted value pa0 and the second predicted value pb0 obtained from the above embodiment with the model hyperparameter alpha to obtain the final target predicted value p0 = alpha × pa0 + (1 - alpha) × pb0;
[0097] When the target activity level is the low-activity level, for the low-activity domain and the common domain, the model hyperparameter can be set to beta (the value range is (0, 1), which can be optimized according to the actual business, preferably 0.5). At this time, fuse the first predicted value pa1 and the second predicted value pb1 obtained from the above embodiment with the model hyperparameter beta to obtain the final target predicted value p1 = beta × pa1 + (1 - beta) × pb1.
[0098] In the embodiments of the present disclosure, by setting model hyperparameters to adjust the proportion of the influence of the common domain and the specific domain, the model can better complete the corresponding tasks and improve the quality of task completion.
[0099] In some alternative embodiments, the above step S104 includes:
[0100] Sort the first network resources according to the target predicted value to obtain a sorted queue;
[0101] Select a preset number of the first network resources in the sorted queue as the recommended second network resources.
[0102] Optionally, each first network resource corresponds to a target predicted value. When the number of first network resources is multiple, the corresponding target predicted values are also multiple. Since the target predicted value represents the degree of interest of the target object in the first network resource, in order to obtain resources with a relatively high degree of interest of the target object as recommended content, the target predicted values can be sorted according to the numerical size at this time. It can be arranged in descending order or ascending order to obtain a sorted queue.
[0103] Select a preset number of the first network resources from the sorted queue as the recommended second network resources. Specifically, when selecting, the first N first network resources can be selected, for example, the first N first network resources, or the last N first network resources can be selected, or N first network resources can be selected according to the sorting situation. The selected first network resources can be recommended to the target object as the recommended second network resources. The triggering condition for sending can be real-time sending or conditional sending, such as sending according to a set fixed time point, etc.
[0104] In an embodiment of the present disclosure, according to the sorting queue, a preset number of first network resources are selected as recommended content, so that it is more likely to select resource content that meets the personal interests of the target object, and improve the consumption penetration rate and consumption days of the target object.
[0105] As a specific application embodiment of the embodiment of the present disclosure, as Figure 3 shown, taking the first network resource as entertainment information and the target object as user A as an example, the specific resource recommendation method process is as follows:
[0106] Step S301, obtain the log data when user A consumes the first network resource in the most recent 15 days;
[0107] Step S302, obtain the consumption active characteristics of user A from the log data, such as consumption duration, action days, active days, and active days ratio;
[0108] Step S303, determine the consumption value of user A when consuming the first network resource according to the consumption active characteristics;
[0109] Step S304, determine the clustering cluster divided according to the consumption value, and then obtain that the activity level of user A is a high activity level;
[0110] Step S305, input the attribute information and high activity level of user A into the public domain to obtain a first predicted value, and input the attribute information and high activity level of user A into the high-activity domain to obtain a second predicted value;
[0111] Step S306, obtain the model hyperparameters that control the influence ratio of the public domain and the high-activity domain;
[0112] Step S307, substitute the model hyperparameters into the first predicted value and the second predicted value to obtain the target predicted value of the degree of interest of user A in the first network resource;
[0113] Taking the first network resource as military information, dance videos, etc. as examples, repeatedly execute steps S301 - S307 to obtain multiple target predicted values;
[0114] Step S308, based on the numerical magnitudes of the multiple target predicted values, determine the top N recommended second network resources from the multiple first network resources.
[0115] In this embodiment, a resource recommendation device is further provided. This device is used to implement the above-mentioned embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that realizes a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.
[0116] This embodiment provides a resource recommendation device, as Figure 4 shown, including:
[0117] A first determination module 401, configured to determine target data based on the target activity level of the target object corresponding to the first network resource and the target attribute information of the target object;
[0118] A first obtaining module 402, configured to input the target data into the common domain of the target model to obtain a first predicted value corresponding to the target activity level, where the common domain is constructed based on the attribute information of the first object and the corresponding activity level;
[0119] A second obtaining module 403, configured to input the target data into the specific domain of the target model to obtain a second predicted value corresponding to the target activity level, where the specific domain is constructed based on the attribute information of the second object and the corresponding activity level, and the first object includes the second object;
[0120] A second determination module 404, configured to determine the target predicted value of the target object corresponding to the first network resource based on the first predicted value and the second predicted value;
[0121] An obtaining module 405, configured to obtain a recommended second network resource from the first network resource based on the target predicted value.
[0122] In some alternative implementation manners, the method further includes:
[0123] A first obtaining module, configured to obtain the log data of the target object within a preset time interval;
[0124] A second obtaining module, configured to obtain the active characteristics of the target object based on the log data;
[0125] A third determination module, configured to determine the corresponding target activity level based on the active characteristics.
[0126] In some alternative implementation manners, the third determination module includes:
[0127] A first determination sub-module, configured to determine the active value corresponding to the first network resource based on the active characteristics;
[0128] A matching sub-module, configured to match a target active range according to an active value and determine a corresponding target class cluster;
[0129] A second determining sub-module, configured to determine a corresponding target activity level according to the target class cluster.
[0130] In some alternative embodiments, the first obtaining module 402 includes:
[0131] A first conversion sub-module, configured to convert target data into a feature vector;
[0132] A first mapping sub-module, configured to map the feature vector to a latent vector space in a common domain to obtain a first feature latent vector;
[0133] A first input sub-module, configured to input the first feature latent vector into a deep network structure layer in the common domain to obtain a first prediction value.
[0134] In some alternative embodiments, the second obtaining module 403 includes:
[0135] A second conversion sub-module, configured to convert target data into a feature vector;
[0136] A second mapping sub-module, configured to map the feature vector to a latent vector space in a specific domain to obtain a second feature latent vector;
[0137] A second input sub-module, configured to input the second feature latent vector into a deep network structure layer in the specific domain to obtain a second prediction value.
[0138] In some alternative embodiments, the apparatus further includes:
[0139] An operation sub-module, configured to perform a matrix multiplication operation on the first feature latent vector and a transfer matrix before mapping the feature vector to the latent vector space in the specific domain to obtain the second feature latent vector, and perform a projection conversion on the first feature latent vector to obtain the latent vector space in the specific domain, where the transfer matrix is determined based on the dimension of the first feature latent vector and the dimension of the latent vector space in the specific domain.
[0140] In some alternative embodiments, the second determining module 404 includes:
[0141] An obtaining sub-module, configured to obtain model hyperparameters, where the model hyperparameters are used to control the influence ratio of the common domain and the specific domain;
[0142] A fusion sub-module, configured to fuse the model hyperparameters, the first prediction value, and the second prediction value to obtain a target prediction value.
[0143] In some alternative embodiments, the obtaining module 405 includes:
[0144] A sorting sub-module, configured to sort first network resources according to target prediction values to obtain a sorted queue;
[0145] A selection sub-module, configured to select a preset number of first network resources from the sorted queue as the recommended second network resources.
[0146] The further function descriptions of the above-mentioned various modules and units are the same as those in the corresponding embodiments above, and will not be elaborated here.
[0147] The resource recommendation device in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.
[0148] The embodiments of the present disclosure also provide a computer device having the above-mentioned Figure 4 shown resource recommendation device.
[0149] Please refer to Figure 5 , Figure 5 which is a schematic structural diagram of a computer device provided by an alternative embodiment of the present disclosure. As Figure 5 shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a set of blade servers, or a multi-processor system). Figure 5 In
[0150] a single processor 10 is taken as an example.
[0151] Among them, the memory 20 stores instructions that can be executed by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiments.
[0152] The memory 20 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 may optionally include a memory remotely provided with respect to the processor 10, and these remote memories can be connected to the computer device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.
[0153] The memory 20 may include a volatile memory, for example, a random access memory; the memory may also include a non-volatile memory, for example, a flash memory, a hard disk, or a solid-state drive; the memory 20 may further include a combination of the above types of memories.
[0154] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.
[0155] The embodiments of the present disclosure further provide a computer-readable storage medium. The method according to the embodiments of the present disclosure can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be processed by such software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium may further include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.
[0156] A part of the present disclosure can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the present disclosure through the operations of the computer. Those skilled in the art should understand that the forms of existence of computer program instructions in a computer-readable medium include but are not limited to source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include but are not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Herein, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to the computer.
[0157] Although the embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations all fall within the scope defined by the appended claims.
Claims
1. A resource recommendation method, characterized in that, The method includes: Determining target data based on the target activity level of the target object corresponding to the first network resource and the target attribute information of the target object; Inputting the target data into the common domain of the target model to obtain a first predicted value corresponding to the target activity level, where the common domain is constructed based on the attribute information of the first object and the corresponding activity level; Inputting the target data into the specific domain of the target model to obtain a second predicted value corresponding to the target activity level, where the specific domain is constructed based on the attribute information of the second object and the corresponding activity level, and the first object includes the second object; Determining the target predicted value of the target object corresponding to the first network resource based on the first predicted value and the second predicted value; Obtaining a recommended second network resource from the first network resource based on the target predicted value.
2. The method according to claim 1, characterized in that, The method further includes: Obtaining the log data of the target object within a preset time interval; Obtaining the activity characteristics of the target object based on the log data; Determining the corresponding target activity level based on the activity characteristics.
3. The method according to claim 2, wherein The determining the corresponding target activity level based on the activity characteristics includes: Determining the activity value corresponding to the first network resource based on the activity characteristics; Matching a target activity interval according to the activity value to determine a corresponding target cluster; Determining the corresponding target activity level according to the target cluster.
4. The method according to claim 1, wherein The inputting the target data into the common domain of the target model to obtain a first predicted value corresponding to the target activity level includes: Converting the target data into a feature vector; Mapping the feature vector to the hidden vector space of the common domain to obtain a first feature hidden vector; Inputting the first feature hidden vector into the deep network structure layer of the common domain to obtain the first predicted value.
5. The method according to claim 4, wherein The inputting the target data into the specific domain of the target model to obtain a second predicted value corresponding to the target activity level includes: Converting the target data into a feature vector; Mapping the feature vector to the hidden vector space of the specific domain to obtain a second feature hidden vector; Inputting the second feature hidden vector into the deep network structure layer of the specific domain to obtain the second predicted value.
6. The method according to claim 5, characterized in that, Before the mapping the feature vector to the hidden vector space of the specific domain to obtain a second feature hidden vector, the method further includes: Performing a matrix multiplication operation on the first feature hidden vector and a transfer matrix to perform a projection transformation on the first feature hidden vector to obtain the hidden vector space of the specific domain, where the transfer matrix is determined based on the dimension of the first feature hidden vector and the dimension of the hidden vector space of the specific domain.
7. The method according to claim 1, characterized in that The determining the target predicted value of the target object corresponding to the first network resource based on the first predicted value and the second predicted value includes: Obtaining model hyperparameters, where the model hyperparameters are used to control the influence ratio of the common domain and the specific domain; Fusing the model hyperparameters, the first predicted value, and the second predicted value to obtain the target predicted value.
8. The method according to claim 1, characterized in that, Obtaining the recommended second network resources from the first network resources based on the target prediction value includes: Sorting the first network resources according to the target prediction value to obtain a sorted queue; Selecting a preset number of the first network resources in the sorted queue as the recommended second network resources.
9. A resource recommendation device, characterized in that, The device includes: A first determination module, configured to determine target data based on the target activity level of the target object corresponding to the first network resources and the target attribute information of the target object; A first obtaining module, configured to input the target data into the common domain of the target model to obtain a first prediction value corresponding to the target activity level, where the common domain is constructed based on the attribute information of the first object and the corresponding activity level; A second obtaining module, configured to input the target data into the specific domain of the target model to obtain a second prediction value corresponding to the target activity level, where the specific domain is constructed based on the attribute information of the second object and the corresponding activity level, and the first object includes the second object; A second determination module, configured to determine the target prediction value of the target object corresponding to the first network resources based on the first prediction value and the second prediction value; An obtaining module, configured to obtain the recommended second network resources from the first network resources based on the target prediction value.
10. A computer device, characterized in that, including: A memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the resource recommendation method according to any one of claims 1 to 8.
11. A computer-readable storage medium, characterized in that, Computer instructions are stored on the computer-readable storage medium, and the computer instructions are used to cause a computer to execute the resource recommendation method according to any one of claims 1 to 8.
12. A computer program product, characterized in that, including computer instructions, and the computer instructions are used to cause a computer to execute the resource recommendation method according to any one of claims 1 to 8.