Recommendation method, device and computer storage medium

Abstract

Embodiments of the present application provide a recommendation method and device, electronic equipment and computer storage medium. The recommendation method comprises: obtaining, as a target object, a digital object preferred by a physical object and obtaining a preference degree of the corresponding target object based on digital object historical interaction data generated by the physical object interacting with the digital object; obtaining, as a candidate object, a digital object that has an association relationship with the target object and has not been interacted with historically from a database for storing digital objects and obtaining a matching degree between the candidate object and the target object; screening the candidate object to be recommended to the physical object based on an index of the candidate object and the matching degree between the target object and the candidate object; and inputting the matching degree between the candidate object to be recommended and the target object, the preference degree of the physical object for the target object and the candidate object to be recommended into a first supervised model to predict the interest degree of the physical object for the candidate object, and determining the candidate object to be recommended.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the computer technical field, and particularly, to a recommendation method and device and computer storage medium. BACKGROUND

[0002] With the continuous development of the Internet, network information is more and more complex, in order to enable users to obtain information from the network to see the information they need conveniently, recommendation technology is applied.

[0003] The current recommendation technology is mostly pushed by the recommendation platform to the target user according to the demand of the recommender, and the information or product of the recommender is pushed to the target user, but this recommendation method does not consider the actual demand or possible demand of the target user, thereby leading to low recommendation efficiency. SUMMARY

[0004] Therefore, embodiments of the present application provide a recommendation scheme to at least partially solve the above problems.

[0005] According to a first aspect of embodiments of the present application, a recommendation method is provided, comprising: obtaining, as a target object, a digital object preferred by a physical object and obtaining a preference degree of the corresponding target object based on digital object historical interaction data generated by the physical object interacting with the digital object; obtaining, as a candidate object, a digital object having an association relationship with the target object and not having a history of interaction from a database for storing digital objects and obtaining a matching degree between the candidate object and the target object; screening a candidate object to be recommended to the physical object based on an index of the candidate object and the matching degree between the target object and the candidate object; inputting the matching degree between the candidate object to be recommended and the target object, the preference degree of the physical object to the target object, and the candidate object to be recommended into a first supervised model, and predicting, by the first supervised model, an interest degree of the physical object to the candidate object; and determining a candidate object to be recommended to the physical object according to the interest degree of the candidate object.

[0006] According to a second aspect of the embodiments of the present application, a recommendation method is provided, including: obtaining, as a target product, a product preferred by a user and obtaining a preference degree of the corresponding target product based on product historical interaction data generated by the user interacting with the product from products interacted with by the user; obtaining, as a candidate product, a product having an association relationship with the target product and not historically interacted with from a database for storing products and obtaining a matching degree of the candidate product and the target product; screening a candidate product to be recommended to the user based on an index of the candidate product and the matching degree of the target product and the candidate product; inputting the matching degree of the target product and the candidate product to be recommended, the preference degree of the target product by the user, and the candidate product to be recommended into a first supervised model to predict an interest degree of the user for the candidate product by the first supervised model; and determining a candidate product to be recommended to the user according to the interest degree.

[0007] According to a third aspect of the embodiments of the present application, a recommendation method is provided, including: obtaining, as a target advertisement, an advertisement preferred by a user and obtaining a preference degree of the corresponding target advertisement based on advertisement historical interaction data generated by the user interacting with the advertisement from advertisements interacted with by the user; obtaining, as a candidate advertisement, an advertisement having an association relationship with the target advertisement and not historically interacted with from a database for storing advertisements and obtaining a matching degree of the candidate advertisement and the target advertisement; screening a candidate advertisement to be recommended to the user based on an index of the candidate advertisement and the matching degree of the target advertisement and the candidate advertisement; inputting the matching degree of the target advertisement and the candidate advertisement to be recommended, the preference degree of the target advertisement by the user, and the candidate advertisement to be recommended into a first supervised model to predict an interest degree of the user for the candidate advertisement by the first supervised model; and determining an advertisement to be recommended to the user according to the interest degree.

[0008] According to a fourth aspect of the embodiments of the present application, a recommendation device is provided, comprising: a preference determining module configured to obtain, as a target object, a digital object preferred by a physical object from digital objects that have been interacted with by the physical object based on digital object historical interaction data generated by the physical object interacting with the digital object, and obtain a preference degree of the corresponding target object; an association determining module configured to obtain, as a candidate object, a digital object that has an association relationship with the target object and has not been interacted with historically from a database for storing digital objects, and obtain a matching degree of the candidate object and the target object; a screening module configured to screen a candidate object to be recommended to the physical object based on an index of the candidate object and the matching degree of the target object and the candidate object; an input module configured to input the matching degree of the target object and the candidate object to be recommended, the preference degree of the target object by the physical object, and the candidate object to be recommended into a first supervised model, and predict, by the first supervised model, an interest degree of the physical object in the candidate object; and a determining module configured to determine the candidate object to be recommended to the physical object according to the interest degree of the candidate object.

[0009] According to a fifth aspect of the embodiments of the present application, a recommendation device is provided, comprising: a preference determining module configured to obtain, as a target object, a commodity preferred by a user from commodities that have been interacted with by the user based on commodity historical interaction data generated by the user interacting with the commodity, and obtain a preference degree of the corresponding target object; an association determining module configured to obtain, as a candidate object, a commodity that has an association relationship with the target object and has not been interacted with historically from a database for storing commodities, and obtain a matching degree of the candidate object and the target object; a screening module configured to screen a candidate object to be recommended to the user based on an index of the candidate object and the matching degree of the target object and the candidate object; an input module configured to input the matching degree of the target object and the candidate object to be recommended, the preference degree of the target object by the user, and the candidate object to be recommended into a first supervised model, and predict, by the first supervised model, an interest degree of the user in the candidate object; and a determining module configured to determine the candidate object to be recommended to the user according to the interest degree.

[0010] According to a sixth aspect of the embodiments of the present application, a recommendation device is provided, comprising: a preference determining module, configured to obtain, based on advertisement historical interaction data generated by user interaction with an advertisement, a user preferred advertisement from the advertisements that the user has interacted with as a target advertisement and a preference degree of the corresponding target advertisement; an association determining module, configured to obtain, from a database storing advertisements, an advertisement that has a correlation relationship with the target advertisement and has not been interacted with historically as a candidate advertisement and a matching degree of the candidate advertisement and the target advertisement; a screening module, configured to screen a candidate advertisement to be recommended to the user based on an index of the candidate advertisement and the matching degree of the target advertisement and the candidate advertisement; an input module, configured to input the matching degree of the candidate advertisement to be recommended and the target advertisement, the preference degree of the target advertisement by the user, and the candidate advertisement to be recommended into a first supervised model, and predict an interest degree of the user for the candidate advertisement by the first supervised model; and a determining module, configured to determine an advertisement to be recommended to the user according to the interest degree.

[0011] According to a seventh aspect of the embodiments of the present application, an electronic device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, the processor, the memory, and the communication interface complete communication with each other through the communication bus; the memory is configured to store at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to the recommendation method as described above.

[0012] According to an eighth aspect of the embodiments of the present application, a computer storage medium is provided, which stores a computer program, and the program is executed by a processor to implement the recommendation method as described above.

[0013] According to the recommendation scheme provided by the embodiments of the present application, when recommending to a physical object such as a user, not only the digital objects that the physical object has interacted with are considered, but also the preferences of the physical object for the digital objects that have been interacted with are fully considered, that is, the target object preferred by the physical object is obtained; and based on this, a digital object that has a certain correlation relationship with the target object preferred by the physical object but has not been interacted with by the physical object, that is, a candidate object, is obtained, and the candidate object to be recommended is screened from the candidate objects; further, the interest of the physical object is predicted based on the matching degree of the candidate object to be recommended and the target object and the preference degree of the physical object for the target object by the first supervised model, so as to obtain the interest degree of the physical object for the candidate object to be recommended, and the candidate object finally recommended to the physical object is determined based on the interest degree. It can be seen that the recommendation scheme of the embodiments of the present application fully considers the actual needs or possible needs of the physical object such as the target user, so that the recommended object is more in line with the interests and needs of the user, and the recommendation efficiency is improved. BRIEF DESCRIPTION OF DRAWINGS

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0015] Figure 1A This is a flowchart illustrating the steps of a recommended method according to Embodiment 1 of this application;

[0016] Figure 1B for Figure 1A A schematic diagram of a scenario example in the illustrated embodiment;

[0017] Figure 2A This is a flowchart illustrating the steps of a recommended method according to Embodiment 2 of this application;

[0018] Figure 2B for Figure 2A The illustration shows a schematic diagram of the matching relationship between non-interactive objects and target objects after aggregation processing in one embodiment.

[0019] Figure 2C This is a flowchart illustrating the steps of a method for constructing a candidate object set according to Embodiment 2 of this application;

[0020] Figure 3 This is a schematic diagram of the neural network architecture used in a recommended method according to Embodiment 3 of this application;

[0021] Figure 4 This is a flowchart illustrating a recommendation method in an e-commerce scenario according to Embodiment 4 of this application;

[0022] Figure 5 This is a flowchart illustrating a recommendation method for advertising recommendation according to Embodiment 5 of this application;

[0023] Figure 6 This is a structural block diagram of a recommended device according to Embodiment Six of this application;

[0024] Figure 7 This is a schematic diagram of the structure of an electronic device according to Embodiment 7 of this application. Detailed Implementation

[0025] In the following, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art should fall within the scope of protection of the present application.

[0026] The specific implementation of the embodiments of the present application will be further described below in conjunction with the drawings of the embodiments of the present application.

[0027] Embodiment One

[0028] Figure 1A A step flow chart of a recommendation method according to Embodiment One of the present application; as shown in FIG. 1, it includes: Figure 1A

[0029] S101, based on the digital object historical interaction data generated by the physical object interacting with the digital object, obtaining the digital object preferred by the physical object from the digital objects that the physical object has historically interacted with as a target object and obtaining the preference degree of the corresponding target object.

[0030] In the present embodiment, the physical object means an actually existing object, including but not limited to a user and / or a device used by the user such as a client and the like; the digital object means an object presented in a digital form, including but not limited to a commodity object, an audio / video object, an electronic reading object and the like displayed in a digital form.

[0031] The digital object historical interaction data generated by the physical object interacting with the digital object can include the interaction time, specific interaction actions and the like of the physical object interacting with the digital object, which describe the interaction. The specific form of the interaction can be different due to different scenarios, including but not limited to browsing, ordering, collecting, sharing, commenting and the like. Similarly, in different scenarios, the digital objects that the physical object has historically interacted with can also be different. For example, in an e-commerce scenario, the digital objects that the physical object has historically interacted with can be items added to a shopping cart or items that have been purchased; in an advertising scenario, the digital objects that the physical object has historically interacted with can be advertisements that have been watched or collected, and the like.

[0032] ​According to the digital object historical interaction data, the preference degree of the physical object to each digital object that the physical object has interacted with can be determined, and then the digital object preferred by the physical object can be taken as a target object, and the preference degree of the physical object to the target object can be determined. The implementation of the process can be implemented by a person skilled in the art in an appropriate manner. In an optional manner, the process can be implemented by a data model for predicting the preference behavior of the physical object, such as a supervised model. However, the application is not limited thereto, and other manners such as data statistical manner are also applicable.

[0033] In S102, a digital object that has a correlation relationship with the target object and has not been interacted with historically is obtained as a candidate object from a database for storing the digital object, and a matching degree of the candidate object to the target object is obtained.

[0034] In the embodiment of the application, the digital object that has a correlation relationship with the target object is usually a digital object that has a high similarity to the target object (for example, both are portable mobile terminal products, etc.), or a digital object that has a high correlation degree (for example, both are often selected by the same user at the same time or in sequence, etc.). In addition, the target object is a preferred object of the physical object, and therefore, the object that has a correlation relationship with the target object also has a relatively high possibility of being preferred by the physical object.

[0035] Further, in the embodiment of the application, the object that has not been interacted with by the physical object, that is, the digital object that has not been interacted with historically, is determined from the object that has a correlation relationship with the target object, and the digital object is taken as a candidate object. Unlike the conventional recommendation scenario in which the recommendation is made to the physical object without distinction, the scheme of the embodiment of the application pays more attention to the digital object that has not been interacted with by the physical object. On the one hand, for the digital object that has been interacted with by the physical object, the physical object has known the information of the digital object, and the information of the digital object can also be easily obtained, and the continuous recommendation will affect the experience of the physical object. On the other hand, for the digital object that has a correlation relationship with the target object preferred by the physical object and has not been interacted with by the physical object, the degree of interest of the physical object to the digital object is high, and the way and channel of obtaining the information of the digital object by the physical object can be expanded, so that the physical object can obtain the information at a low cost, and the experience of the physical object is improved.

[0036] The matching degree of the candidate object to the target object can effectively predict the degree of interest or preference of the physical object to the candidate object.

[0037] In a feasible manner, the candidate object that has a correlation relationship with the target object and has not been interacted with historically and the matching degree of the candidate object to the target object can be obtained by a graph model.

[0038] For example, a relationship graph corresponding to the digital objects stored in the database can be established in advance, which can be represented as G=(N, E), where N represents a set of nodes, and E represents a set of edges, and the edges are used to connect the nodes. It should be noted that the digital objects stored in the database are digital objects collected and counted through a large amount of related data, which include digital objects that have interacted with the physical object.

[0039] Specifically, one digital object can correspond to one node N, and if two nodes are connected by an edge, it can be identified that the two digital objects have an association relationship. Further, the type, closeness, and the like of the association between the two digital objects can be represented by the type, weight, length, and the like of the edge. Thus, the position of the target object in the relationship graph can be located through the pre-established relationship graph, and the digital objects having an association relationship with the target object can be determined. Of course, the above is only an example and does not limit the present application.

[0040] Based on this, the digital objects having an association relationship with the target object can be, for example, the digital objects corresponding to the other nodes directly connected to the node corresponding to the target object through the edge; or the digital objects corresponding to the other nodes connected to the node corresponding to the target object through n edges, where n can be set by a person skilled in the art, and the present embodiment does not limit this. The smaller n is, the closer the association relationship is.

[0041] After determining the digital objects having an association relationship with the target object, the digital objects that have interacted with the physical object in history can be removed, and the remaining digital objects that have not interacted in history are taken as candidate objects.

[0042] The matching degree of the candidate object and the target object can be used to represent the association degree of the candidate object and the target object. For example, if the relationship graph corresponding to the digital objects stored in the database is established in advance, the matching degree of the candidate object and the target object can be determined based on the edge between the node corresponding to the candidate object and the node corresponding to the target object and various parameters corresponding to the edge. Of course, the above is only an example and does not limit the present application. Based on the preference degree of the physical object to the target object and the association degree, the degree of interest or preference of the physical object to the candidate object can be effectively predicted.

[0043] S103, based on the index of the candidate object and the matching degree of the target object and the candidate object, screening the candidate object to be recommended to the physical object.

[0044] Based on the index of the candidate object and the matching degree of the target object and the candidate object, the candidate object is screened.

[0045] The index of the candidate object can be used to identify the candidate object. In addition, since the target object is a digital object preferred by the physical object, the matching degree between the target object and the candidate object can indirectly represent the preference degree of the physical object for the candidate object. The filtering based on the index of the candidate object and the matching degree between the target object and the candidate object can make the retained candidate objects different from each other, and can ensure that the filtered candidate objects meet the needs of the physical object as much as possible. In addition, after filtering, the number of candidate objects can be further reduced, so as to reduce the calculation amount of the first supervised model in the subsequent step and improve the recommendation speed.

[0046] In S104, the matching degree between the candidate object to be recommended and the target object, the preference degree of the physical object for the target object, and the candidate object to be recommended are input into the first supervised model, and the first supervised model is used to predict the interest degree of the physical object for the candidate object.

[0047] In this embodiment, the first supervised model can be any appropriate supervised data model. Through training of the first supervised model, the first supervised model can have the function of predicting the preference data and the preference degree of the physical object according to the input data. Specifically, in this embodiment, the first supervised model has the function of predicting the interest degree of the physical object for the candidate object.

[0048] In this embodiment, by inputting the matching degree between the candidate object to be recommended and the target object, the preference degree of the physical object for the target object, and the candidate object to be recommended into the first supervised model, the preference degree obtained for representing the preference of the physical object and the matching degree obtained for representing the association between the target object and the candidate object can be effectively transmitted to the first supervised model. Therefore, the first supervised model does not need to learn the preference of the physical object and the association between objects again, and the accuracy and efficiency of predicting the interest degree of the physical object for the candidate object are improved.

[0049] In S105, the candidate object to be recommended to the physical object is determined according to the interest degree of the candidate object.

[0050] After the interest degree of the physical object for the candidate object is determined, the candidate object with a higher interest degree can be recommended to the physical object. The specific recommendation method can be any appropriate method according to actual needs, and this embodiment is not limited thereto.

[0051] The scheme of this embodiment will be exemplarily described below through a specific use scenario.

[0052] Reference Figure 1BIn the example, the target object preferred by the physical object (e.g., a user) and the preference degree of the target object can be obtained from the digital objects that have been interacted with by the physical object (e.g., the goods purchased by the user) based on the digital object historical interaction data (e.g., the goods purchased by the user). The target object can include multiple target objects, i.e., historical digital object 1, historical digital object 2, and historical digital object n. For example, the preferred goods i1, i2, and in purchased by the user.

[0053] After the target object is determined, the digital objects that have not been interacted with by the physical object and that have an association relationship with the target object can be obtained from the database as candidate objects. The candidate objects can be uninteracted object 1, uninteracted object 2, and uninteracted object m. The matching degree between the candidate objects and the target object can also be obtained. For example, the candidate goods j1, j2, and jm that have an association relationship with the preferred goods i1, i2, and in purchased by the user and that have not been purchased by the user. It should be noted that the candidate goods j1, j2, and jm can have an association relationship with some or all of the preferred goods i1, i2, and in. It is assumed that the goods that have an association relationship with the preferred goods i1, i2, and in include goods j1, j2, and jX, which are composed of the association goods corresponding to each of the goods i1, i2, and in. The candidate goods j1, j2, and jm are obtained by removing the goods purchased by the user from the goods j1, j2, and jX. Each of the candidate goods j1, j2, and jm has an association relationship with one of the goods i1, i2, and in, but it is possible that the association goods corresponding to a certain good in i1, i2, and in belong to the goods purchased by the user and thus are not in j1, j2, and jm. However, each of j1, j2, and jm has a matching degree with the preferred goods of the user corresponding thereto. However, this is not limited thereto. In actual applications, a certain candidate good can also have a matching degree with multiple preferred goods. For example, the electronic wristwatch A has a certain matching degree with the mobile phone B and a certain matching degree with the watch C.

[0054] Afterwards, the un-interacted digital objects 1, 2, …, m can be filtered based on the indexes of the candidate objects and the matching degrees between the target object and the candidate objects to form candidate objects to be recommended. For example, candidate goods j1, j2, …, jp are filtered from the candidate goods j1, j2, …, jm to be recommended. Then, the matching degrees between the candidate objects to be recommended and the target object, the preference degrees of the physical object to the target object, and the candidate objects to be recommended are input into the first supervised model, and the interest degrees of the physical object to the candidate objects are predicted by the first supervised model. The candidate objects to be recommended are determined according to the interest degrees of the candidate objects. For example, the candidate goods j1, j2, …, jp to be recommended are part of the candidate goods j1, j2, …, jm, and therefore the matching degrees between the candidate goods to be recommended and the preferred goods are data in the matching degrees between all the candidate goods and the preferred goods. Based on this, the matching degrees between the candidate goods j1, j2, …, jp to be recommended and the preferred goods i1, i2, …, in, the preference degrees of the user to the preferred goods i1, i2, …, in, and the information of the candidate goods j1, j2, …, jp to be recommended are input into the first supervised model, and the interest degrees of the user to the candidate goods j1, j2, …, jp to be recommended are predicted by the first supervised model. Then, the candidate goods to be finally recommended to the user, such as the candidate goods j1, j2, j3, etc., are determined.

[0055] According to the scheme provided in the embodiments of the present application, when recommending to a physical object such as a user, not only the digital objects interacted by the physical object are considered, but also the preference of the physical object to the interacted digital objects is fully considered to obtain the target object preferred by the physical object. Then, based on this, the candidate objects having a certain correlation with the target object preferred by the physical object but not yet interacted by the physical object are obtained, and the candidate objects to be recommended are filtered from the candidate objects. Then, the interest of the physical object is predicted by the first supervised model based on the matching degrees between the candidate objects to be recommended and the target object and the preference degrees of the physical object to the target object, so that the interest degrees of the physical object to the candidate objects to be recommended are obtained. Finally, the candidate objects to be finally recommended to the physical object are determined based on the interest degrees. It can be seen that the recommendation scheme of the embodiments of the present application fully considers the actual needs or possible needs of the physical object such as the target user, so that the recommended objects are more in line with the interests and needs of the user, and the recommendation efficiency is improved.

[0056] The recommendation method of the embodiments can be executed by any appropriate electronic device having data processing capability, including but not limited to: a server, a mobile terminal (such as a mobile phone, a PAD, etc.), and a PC, etc.

[0057] Embodiment Two

[0058] Figure 2AA flowchart of steps of a recommendation method according to Embodiment Two of the present application; as shown in Figure 2A

[0059] S201, inputting digital object historical interaction data generated based on interaction between a physical object and a digital object into a second supervised model, performing preference behavior prediction of the physical object by the second supervised model, obtaining a digital object preferred by the physical object and taking the digital object as a target object, and obtaining a preference degree of the physical object to the target object.

[0060] The digital object historical interaction data includes information of the digital object that has been interacted with by the physical object and corresponding historical interaction behavior information.

[0061] In this embodiment, the second supervised model is used to determine the target object and the preference degree of the physical object to the target object. The second supervised model can be any appropriate supervised data model, and the type or implementation of the second supervised model is not limited in this embodiment.

[0062] The second supervised model can be trained by using a large amount of historical interaction data of non-specific physical objects as training samples. For example, in an e-commerce scenario, the historical interaction data can be browsing data and purchase data of all users on an e-commerce platform; or in an advertisement recommendation scenario, the historical interaction data can be historical records of users watching advertisements. In this way, the historical interaction data of users can be fully mined, and the accuracy of the determined target object can be improved.

[0063] By using the second supervised model, the target object preferred by the physical object and the preference degree of the corresponding target object can be more accurate.

[0064] S202, inputting the target object into a graph model, performing association relationship prediction between the pre-stored digital objects in the database and the target object based on the pre-stored digital objects in the database by the graph model, and obtaining pre-stored digital objects having an association relationship with the target object and a matching degree between the pre-stored digital objects and the target object.

[0065] In this embodiment, the digital objects stored in the database are referred to as pre-stored digital objects for ease of distinction. As described above, the pre-stored digital objects include digital objects that have been interacted with by the physical object. Based on the pre-stored digital objects stored in the database, the graph model is used to determine a plurality of pre-stored digital objects having an association relationship with the target object and a matching degree between the pre-stored digital objects and the target object. The graph model simplifies the relationship description and calculation between a plurality of digital objects, and improves the efficiency of determining the digital objects and the matching degree.

[0066] ​For example, the plurality of target objects can be respectively input into the graph model, and the graph model can output, for each target object, a pre-stored digital object having an association relationship with the target object and a corresponding matching degree. Of course, the plurality of target objects can also be simultaneously input into the graph model, and the present application does not limit this.

[0067] Since the graph model is a graph structure established based on data in the database, its calculation can also be implemented based on the graph structure. Specifically, after the target object is input into the graph model, the graph model can calculate the information of the target object to determine a node in the graph structure corresponding to the target object; then based on the edges of the graph structure, other nodes directly or indirectly connected to the node through the edges can be determined, and based on this, the pre-stored digital objects having an association relationship with the target object can be determined. In addition, among the pre-stored digital objects corresponding to the other nodes directly or indirectly connected to the node through the edges, there can also be nodes of pre-stored digital objects that have historically interacted and nodes of pre-stored digital objects that have not historically interacted.

[0068] S203, from the pre-stored digital objects having an association relationship with the target object, determine a pre-stored digital object that the physical object has not historically interacted with as a candidate object, and determine a matching degree of the candidate object and the target object.

[0069] In this embodiment, the candidate object that the physical object has not historically interacted with can be determined according to whether the pre-stored digital object having an association relationship with the target object is a historically non-interacted object, and the matching degree of the candidate object and the target object can be determined.

[0070] In one possible way, after the historically non-interacted object is determined from the pre-stored digital objects, it can be further judged whether there is an advertising object among them, and if there is, the advertising object can also be removed therefrom. This is because, although the advertising object is not targeted at the current physical object for promotion, the physical object is likely to have known the information of the advertising object, and there is no need to recommend it to the physical object again, so it can be removed from the non-interacted digital objects of the physical object to further improve the recommendation efficiency and avoid repeated recommendations. However, it is not limited thereto, and the matching degree of such a candidate object and the target object can be lowered, or the matching score corresponding to such a candidate object can be lowered in the subsequent to reduce the possibility of recommendation to the physical object.

[0071] S204, based on the index of the candidate object and the matching degree of the target object and the candidate object, screening the candidate object to be recommended to the physical object.

[0072] In this embodiment, the index information can include identification information of the candidate object or a category of the candidate object.

[0073] The identification information can be a unique identification of the candidate object. For example, if the candidate object is a specific commodity, the identification information is a commodity ID; if the candidate object is an advertisement, the identification information is an advertisement ID.

[0074] The category of the candidate object can be used to identify the classification or category of the candidate object. For example, if the candidate object is a commodity, the category is specifically the category of the commodity, such as a mobile phone, a dress, etc.

[0075] In this embodiment, step S204 can include the following a or b:

[0076] a) When the index of the candidate object includes the identification information of the candidate object, the candidate object is filtered based on the matching degree between the target object and the candidate object to obtain a filtered candidate object; the filtered candidate object is aggregated based on the identification information of the filtered candidate object to obtain a candidate object to be recommended to the physical object.

[0077] In this embodiment, when the candidate object is filtered based on the matching degree between the target object and the candidate object, m candidate objects with higher matching degrees can be retained, so that the candidate object to be recommended is more in line with the recommendation requirements of the physical object, and then the filtered candidate object is aggregated based on the identification information of the candidate object, so that the identification information of the candidate object to be recommended to the physical object is different, avoiding repeated calculation. In this embodiment, the aggregation of the filtered candidate object is used to concentrate the candidate objects with the same identification information together, and one of them is taken as a representative, so as to avoid recommending duplicate candidate objects.

[0078] For example, for each target object, m candidate objects with higher matching degrees can be retained, and then the m candidate objects can be aggregated according to the identification information to obtain a candidate object to be recommended to the physical object.

[0079] Suppose there are n target objects, for each target object, the candidate object is filtered based on the matching degree between the target object and the candidate object, and m candidate objects with higher matching degrees are retained, then for n target objects, m*n candidate objects are retained in total. Then the m*n candidate objects can be aggregated according to the identification information of the candidate object to obtain a candidate object to be recommended to the physical object.

[0080] For example, if the physical object is user, the target object is item, and the candidate object is No_Item, the relationship graph between user-target object-uninteracted object after aggregation can refer to Figure 2B , Figure 2BIn the figure, the left node u1 represents the physical object user, the middle nodes I1, I2, I3, I4, Im represent the target objects determined from the digital objects Item interacted with by the physical object in history, and the right nodes N1, N2, Nk represent the candidate objects No_Item obtained after screening. The connection edge between the physical object and the target object represents that the target object is the preferred target object of the physical object, and the connection edge between the target object and the candidate object represents that the target object and the candidate object have an association relationship.

[0081] Through the identification information of the candidate object, accurate recommendation of the digital object can be realized.

[0082] B) When the index of the candidate object includes the category of the candidate object, based on the category of the candidate object and the matching degree between the target object and the candidate object, the candidate object to be recommended to the physical object is screened.

[0083] In this embodiment, when the screening is based on the category of the candidate object and the matching degree between the target object and the candidate object, the category of the candidate object can be screened according to the matching degree between the target object and the candidate object, the category corresponding to the m candidate objects with higher matching degrees is retained, and the candidate objects under the category are recommended to the physical object. For example, the category includes tablet computers and smart wearable devices, and then multiple goods under the tablet computers can be recommended to the target user, and / or multiple goods under the smart wearable devices can be recommended to the user. Of course, the above is only an example and does not limit the embodiments of the present application. Through the category of the candidate object, the recommendation range can be appropriately expanded based on the preference of the physical object, so that the physical object obtains more information.

[0084] S205, constructing a candidate object set containing the candidate object to be recommended according to the matching degree between the candidate object to be recommended and the target object, the preference degree of the physical object to the target object, and the association relationship between the target object and the candidate object to be recommended.

[0085] In this embodiment, step S205 can be implemented as shown in Figure 2C

[0086] S2051, determining the matching score of the candidate object to be recommended according to the matching degree between the candidate object to be recommended and the target object and the association relationship between the candidate object to be recommended and the target object.

[0087] Optionally, in this embodiment, the matching degree between the candidate object to be recommended and the target object can be aggregated according to the association relationship between the candidate object to be recommended and the target object, to obtain the matching score of the candidate object to be recommended.

[0088] ​The aggregation processing can be implemented by an aggregation function. For example, the matching degrees of the same candidate object and multiple target objects can be input into the aggregation function, and the matching score of the candidate object is calculated by the aggregation function to comprehensively consider the matching degrees of the candidate object and the target objects. In an implementation, the aggregation function can be implemented as an Aggregation() function, but is not limited thereto. In actual applications, those skilled in the art can also use other forms of aggregation functions, which are not limited in the embodiment.

[0089] In the embodiment, since the matching score can be used to represent the matching degrees between the candidate object and all the corresponding target objects, after the matching score is calculated, the matching score can be used as a parameter of the candidate object to construct the candidate object set of the physical object, so as to simplify the data complexity in the candidate object set, and facilitate the first supervision model to predict the interest degree of the physical object to the candidate object according to the candidate object set.

[0090] However, the matching score can be further processed in an implementation to further improve the accuracy and efficiency of the recommendation. For example, candidate objects with low or high interaction frequencies of the physical object are determined according to big data statistics, etc. (wherein those skilled in the art can appropriately set a threshold value to determine whether the interaction frequency is low or high according to actual conditions). For example, commodities such as televisions, refrigerators and computers are usually low-frequency interaction commodities because users do not frequently purchase them, and commodities such as daily necessities and snacks are usually high-frequency interaction commodities because users frequently purchase them. Accordingly, the matching score of the candidate object to be recommended can be adjusted. For example, if the candidate object is a high-frequency interaction candidate object, the matching score is adjusted according to certain rules; otherwise, the matching score is adjusted. The rules can be appropriately set by those skilled in the art according to actual needs, and the embodiment of the application is not limited thereto.

[0091] In addition, as described above, for the advertising object in the candidate object to be recommended, the matching score can also be adjusted in this step to reduce the recommendation degree and possibility.

[0092] S2052, determining a preference score of the physical object to the candidate object to be recommended according to the preference degrees of the target objects and the association relationships between the target objects and the candidate object to be recommended.

[0093] Optionally, in the embodiment, the preference degrees of the target objects can be aggregated according to the association relationships between the target objects and the candidate object to be recommended to obtain the preference score of the physical object to the candidate object to be recommended.

[0094] The aggregation process in this step can also be implemented by an aggregation function such as the Aggregation() function. For example, the preference degrees of multiple target objects corresponding to the same candidate object can be input into the aggregation function, and the preference score of the candidate object can be calculated by the aggregation function.

[0095] In this embodiment, the preference score can be used to represent the preference degree of the physical object to the candidate object. Therefore, after the preference score is calculated, it can be used as a parameter of the candidate object to construct the candidate object set for the physical object, which simplifies the data complexity in the candidate object set, improves the connection between the data in the candidate object set and the recommended object, and facilitates the subsequent first supervised model to predict the interest degree of the physical object to the candidate object according to the candidate object set.

[0096] It should be noted that in actual application, the execution of steps S2051 and S2052 can not be in a certain order, or can be performed in parallel.

[0097] S2053, constructing a candidate object set containing the candidate object to be recommended according to at least the matching score of the candidate object to be recommended and the preference score of the candidate object to be recommended.

[0098] In this embodiment, by establishing the candidate object set for the physical object, the preference degree of the physical object to the target object obtained by the second supervised model and the matching degree of the target object to the candidate object obtained by the graph model can be conveniently transmitted to the first supervised model, and the utilization rate of information is improved.

[0099] In addition, optionally, before step S2053, the following steps can also be performed in this embodiment:

[0100] Determining the number of target objects having an association relationship with the candidate object to be recommended to improve the prediction accuracy of the subsequent first supervised model.

[0101] Based on this, step S2053 can include: constructing a candidate object set containing the candidate object to be recommended according to at least the matching score of the candidate object to be recommended, the preference score of the candidate object to be recommended, and the number of target objects having an association relationship with the candidate object to be recommended. In this way, the data included in the candidate object set can be enriched, information loss can be minimized, and the number of target objects having an association relationship with the candidate object can be input into the first supervised model as supplementary information to assist in prediction, thereby improving the accuracy of the predicted interest degree of the physical object to the candidate object.

[0102] In this embodiment, the number of target objects having a correlation relationship with the candidate object can be determined by an aggregation function, or can be determined by other manners, which will not be described herein again.

[0103] In S206, the first supervised model is input with the candidate object set, and the first supervised model is used to predict the interest degree of the physical object to the candidate object.

[0104] As described above, the first supervised model in this embodiment can be any appropriate supervised data model. Optionally, the first supervised model can be implemented as a Deep Interest Evolution Network (DIEN) model.

[0105] The DIEN model focuses on the process of user interest evolution, so as to more accurately express the user interest and bring higher CTR prediction accuracy. However, the DIEN model is not limited thereto, and other supervised data models, such as a DeepMCP model, can also be applicable to the scheme of the embodiments of the present application.

[0106] In S207, the candidate object to be recommended to the physical object is determined according to the interest degree of the physical object to the candidate object.

[0107] For example, the candidate object can be recommended to the physical object in the order from high to low of the interest degree, or the candidate object whose interest degree is greater than a certain threshold is determined first, and the candidate object is recommended to the physical object, and the like, which is not limited in the embodiments of the present application.

[0108] According to the scheme provided in the embodiments of the present application, when the physical object such as a user is recommended, not only the digital object interacted by the physical object is considered, but also the preference of the physical object to the interacted digital object is fully considered to obtain the target object which is the digital object preferred by the physical object; and on this basis, the candidate object which has a certain correlation relationship with the target object preferred by the physical object but has not been interacted by the physical object is obtained, and the candidate object to be recommended is screened out from the candidate object; then, the interest of the physical object is predicted based on the matching degree of the candidate object to be recommended and the target object and the preference degree of the physical object to the target object by the first supervised model, so as to obtain the interest degree of the physical object to the candidate object to be recommended, and the candidate object finally recommended to the physical object is determined based on the interest degree. It can be seen that the recommendation scheme of the embodiments of the present application fully considers the actual demand or possible demand of the physical object such as the target user, so that the recommended object is more in line with the user interest and demand, and the recommendation efficiency is improved.

[0109] The recommendation method of the embodiments of the present application can be executed by any appropriate electronic device with data processing capability, including but not limited to: a server, a mobile terminal (such as a mobile phone, a PAD, and the like), a PC, and the like.

[0110] Embodiment Three

[0111] Figure 3 An architecture diagram of a neural network used in a recommendation method provided by an embodiment of the present application; as shown in Figure 3 which includes a second supervised model F1, a graph model G2, and a first supervised model F3.

[0112] In use, the digital object historical interaction data generated by the interaction of the physical object with the digital object can be input to the second supervised model F1, and the target object preferred by the physical object and the preference degree of the target object are obtained through the second supervised model F1.

[0113] Figure 3 The input of the second supervised model shown in the above is the historical interaction data of the physical object, i.e., the user User, and the digital object Item, including interaction entity information Entity info (such as the information of the purchased goods), interaction time information Time info, and interaction party information Side Info. Among them, the interaction entity information Entity info can specifically include item, brand information brand of the item, at least one category information of the item, Figure 3 Two types of category information cate and cate_1 are shown in the above; the interaction time information Time info can include time window information window, time-based object access sequence sequence, and multimodal information multimodal. The interaction party information Side Info can include user information user side and object information item info.

[0114] In addition, Figure 3 The model type information Model that can be adopted by the second supervised model F1 is also shown in the above, including an inference-based model (inference based model) or a parameter-based model (parameter based model), etc., such as a DIEN model or a DeepMCP model, etc.

[0115] The second supervised model F1 predicts the preference of the user User based on the input historical interaction data, and outputs a prediction result, Figure 3 which is shown in the above as a triple <User, Item, pctr>, wherein User is used to represent the user, Item is used to represent the preferred object of the user, i.e., the target object, and pctr is used to represent the preference degree of the user to the target object.

[0116] Further, based on the output of F1, m digital objects with higher preference degree pctr are selected as target objects. Wherein, m is a positive integer, and its specific value can be appropriately set by those skilled in the art according to actual needs.

[0117] Then, the target objects output by F1 are input into the graph model G2.

[0118] Figure 3 Among them, the graph model G2 can be a model based on general attribute multiplex heterogeneous network embedding (GATNE, General Attributed Multiplex Heterogeneous Network Embedding). But not limited to this, other graph models that can obtain associated data associated with input data based on input data are also applicable.

[0119] In this embodiment, the target objects can include multiple, respectively Item 1, Item 2…Item m, for each target object, the graph model G2 can output the digital objects having an associated relationship with the target object, and the matching degree of the digital objects and the target object. Exemplarily, Item 1, Item 2…Item m respectively correspond to the digital objects {N1, N2…Nm}, N1 includes at least one digital object matching Item 1, N2 includes at least one digital object matching Item 2, and so on, Nm includes at least one digital object matching Item m.

[0120] It should be noted that the digital objects having an associated relationship with the target object predicted by the graph model G2 can not only indicate similar objects to the target object, but also indicate objects having a matching relationship with the target object, such as beer, diapers, etc.

[0121] The input of the graph model G2 is the target object Item output by F1, and the output includes digital objects having an associated relationship with the target object Item and the matching degree match_score between the target object Item and the digital objects having an associated relationship. Further, based on the historical interaction data obtained before, the digital objects that the user User has interacted with can be removed from the output digital objects to obtain digital objects No_Item that the user User has not interacted with, as shown in Figure 3

[0122] In addition, Figure 3 ​In the middle, the data part Data of the graph model G2 can include point data node, edge data edge, time data time, and the like. The model type information Model of the graph model G2 indicates that it can adopt a statistical based model, or can adopt an embedding based model, and the like.

[0123] For example, for the target object Item i, the graph model G2 can output the corresponding digital object Ni, which can include n digital objects associated with the target object Item i. Correspondingly, n matching degrees can also be output, respectively match_score i，j Wherein, i = [1, m], j = [1, n]. Further, the digital objects that have been interacted by the user are removed from the output digital objects, to obtain the digital objects No_Item that have not been interacted by the user and the corresponding match_score. Wherein, the digital object No_Item is the candidate object.

[0124] In an implementation manner, after obtaining the output of the graph model G2, the candidate objects No_Item corresponding to the target objects Item 1, Item 2…Item m can be screened based on the indexes of the candidate objects and the matching degrees of the target objects and the candidate objects, to obtain the candidate objects to be recommended.

[0125] Then, the candidate object set can be constructed based on the matching degree match_score of the candidate object to be recommended and the target object, the preference degree pctr of the physical object to the target object, and the candidate object No_Item, and input to the first supervised model F3.

[0126] In the construction process of the candidate object set, for a certain candidate object No_Item, the target objects Item associated with the candidate object No_Item can also be determined, and then the preference degree of the determined target objects Item can be calculated through an aggregation function to obtain the preference score corresponding to the candidate object No_Item. The matching degree between the candidate object No_Item and the determined target objects Item can also be calculated through an aggregation function to obtain the matching score corresponding to the candidate object No_Item. Then the candidate object set can be constructed based on the preference score, the matching score of the candidate object No_Item, and the candidate object No_Item, and input to the first supervised model F3.

[0127] Optionally, the number of target objects Item corresponding to the candidate object No_Item, connect_cnt, can also be counted as one of the parameters in the candidate object set.

[0128] Similar to the second supervised model F1, the first supervised model F3 can also adopt any appropriate supervised data model, including but not limited to DIEN model and DeepMCP model, etc. Figure 3 In this embodiment, the form of DeepMCP model is adopted, where the meaning of each node can refer to Figure 2B Part.

[0129] In another implementation manner, the above aggregation process can be directly performed by the first supervised model F3, for example, the target objects and the corresponding preference degrees of the second supervised model F1, and the candidate objects and the corresponding matching degrees output by the graph model G2 which have the association relationship with the target objects, can be directly input to the first supervised model F3, and the first supervised model F3 determines the above preference score and matching score.

[0130] In summary, the first supervised model F3 can calculate and output the interest degree of the user User to the candidate object No_Item based on the output of the second supervised model F1 and the graph model G2. Then, the user can be recommended according to the interest degree, for example, multiple candidate objects No_Item with high interest degree are recommended to the user.

[0131] According to the recommendation scheme provided in the embodiments of the present application, when recommending to the physical object such as the user, not only the digital objects interacted by the physical object are considered, but also the preference of the physical object to the interacted digital objects is fully considered to obtain the target object which is the digital object preferred by the physical object; and on this basis, the candidate object which has a certain association relationship with the target object preferred by the physical object but has not been interacted by the physical object is obtained, and the candidate object to be recommended is screened out; further, the interest of the physical object is predicted based on the matching degree of the candidate object to be recommended and the target object and the preference degree of the physical object to the target object by the first supervised model, so as to obtain the interest degree of the physical object to the candidate object to be recommended, and the candidate object finally recommended to the physical object is determined based on the interest degree. It can be seen that the recommendation scheme of the embodiments of the present application fully considers the actual demand or possible demand of the physical object such as the target user, so that the recommended object is more in line with the interest and demand of the user, and the recommendation efficiency is improved.

[0132] The recommendation method of the embodiments can be executed by any appropriate electronic device with data processing capability, including but not limited to: server, mobile terminal (such as mobile phone, PAD, etc.) and PC, etc.

[0133] Embodiment Four

[0134] Figure 4 Fig. 1 is a flowchart of a recommendation method according to an embodiment of the present application in an e-commerce scenario; as shown in the figure, the method comprises the following steps: Figure 4

[0135] S401, obtaining, as target goods and corresponding preference degrees, goods preferred by a user from goods interacted with by the user based on historical interaction data of the user and the goods.

[0136] S402, obtaining, as candidate goods, goods having an association relationship with the target goods and having not been interacted with historically from a database for storing the goods and obtaining a matching degree of the candidate goods and the target goods.

[0137] S403, screening candidate goods to be recommended to the user based on an index of the candidate goods and the matching degree of the target goods and the candidate goods.

[0138] S404, inputting the matching degree of the candidate goods to be recommended and the target goods, the preference degree of the user for the target goods and the candidate goods to be recommended into a first supervised model, and predicting, by the first supervised model, an interest degree of the user for the candidate goods.

[0139] S405, determining, according to the interest degree, candidate goods to be recommended to the user.

[0140] The aforementioned recommendation method is applied to an e-commerce scenario in this embodiment, which can effectively improve the recommendation efficiency of an e-commerce platform, meet the recommendation needs of users and expand the range of recommended goods. Moreover, the recommendation of goods not interacted with by the user is based on the preference of the user, which greatly improves the targeting and accuracy of the recommendation.

[0141] It should be noted that the description of each step in this embodiment is relatively brief, and the relevant parts can be referred to the relevant description in the aforementioned multiple embodiments, which will not be described here again.

[0142] The recommendation method of this embodiment can be executed by any appropriate electronic device with data processing capability, including but not limited to: a server, a mobile terminal (such as a mobile phone, a PAD, etc.) and a PC, etc.

[0143] Embodiment Five

[0144] Figure 5 Fig. 1 is a flowchart of a recommendation method according to an embodiment of the present application in an e-commerce scenario; as shown in the figure, the method comprises the following steps: Figure 5

[0145] ​​S501, obtaining, as target advertisements, the advertisements preferred by the user and obtaining the preference degrees of the target advertisements based on advertisement history interaction data generated by the user interacting with the advertisements.

[0146] S502, obtaining, as candidate advertisements, the advertisements that have a correlation relationship with the target advertisements and have not been interacted with historically from a database for storing advertisements and obtaining the matching degrees of the candidate advertisements and the target advertisements.

[0147] S503, screening the candidate advertisements to be recommended to the user based on the indexes of the candidate advertisements and the matching degrees of the target advertisements and the candidate advertisements.

[0148] S504, inputting the matching degrees of the target advertisements and the candidate advertisements to be recommended, the preference degrees of the target advertisements by the user, and the candidate advertisements to be recommended into a first supervised model, and predicting the interest degrees of the candidate advertisements by the user by the first supervised model.

[0149] S505, determining the advertisements to be recommended to the user according to the interest degrees.

[0150] The foregoing recommendation method is applied to an advertisement scenario in this embodiment, which can effectively improve the delivery efficiency of an advertisement delivery platform and expand the delivery range of advertisements. Moreover, the recommendation and delivery of advertisements that have not been interacted with by the user are based on the preferences of the user, which greatly improves the targeting and accuracy of the recommendation.

[0151] It should be noted that the description of each step in this embodiment is relatively brief, and the related parts can be referred to the related description in the foregoing multiple embodiments, which will not be described here again.

[0152] The recommendation method of this embodiment can be executed by any appropriate electronic device with data processing capability, including but not limited to: a server, a mobile terminal (such as a mobile phone, a PAD, etc.), and a PC, etc.

[0153] Embodiment Six

[0154] Figure 6 FIG. 6 is a structural block diagram of a recommendation device according to Embodiment Six of the present application; as shown in the figure, it includes: Figure 6

[0155] The preference determination module 601 is configured to obtain, as target objects, the digital objects preferred by the physical object and obtain the preference degrees of the target objects based on digital object history interaction data generated by the physical object interacting with the digital objects.

[0156] ​The association determining module 602 is configured to obtain, from a database for storing digital objects, a digital object that has an association relationship with the target object and has not been interacted with historically as a candidate object and obtain a matching degree of the candidate object and the target object;

[0157] The screening module 603 is configured to screen the candidate object to be recommended to the physical object based on the index of the candidate object and the matching degree of the target object and the candidate object.

[0158] The input module 604 is configured to input the matching degree of the candidate object to be recommended and the target object, the preference degree of the physical object for the target object, and the candidate object to be recommended into a first supervised model, and predict the interest degree of the physical object for the candidate object by the first supervised model.

[0159] The determining module 605 is configured to determine the candidate object to be recommended to the physical object according to the interest degree of the candidate object.

[0160] Optionally, in the embodiment of the present application, the input module 604 comprises a candidate set constructing module configured to construct a candidate object set containing the candidate object to be recommended according to the matching degree of the candidate object to be recommended and the target object, the preference degree of the physical object for the target object, and the association relationship between the target object and the candidate object to be recommended; and a prediction module configured to input the candidate object set into the first supervised model and predict the interest degree of the physical object for the candidate object by the first supervised model.

[0161] Optionally, in the embodiment of the present application, the candidate set constructing module is configured to determine a matching score of the candidate object to be recommended according to the matching degree of the candidate object to be recommended and the target object and the association relationship between the target object and the candidate object to be recommended; determine a preference score of the physical object for the candidate object to be recommended according to the preference degree of the physical object for the target object and the association relationship between the target object and the candidate object to be recommended; and construct a candidate object set containing the candidate object to be recommended according to at least the matching score of the candidate object to be recommended and the preference score of the candidate object to be recommended.

[0162] Optionally, in the embodiment of the present application, the device further comprises a quantity determining module configured to determine the number of target objects having an association relationship with the candidate object to be recommended; and a candidate set constructing module configured to construct a candidate object set containing the candidate object to be recommended according to at least the matching score of the candidate object to be recommended, the preference score of the candidate object to be recommended, and the number of target objects having an association relationship with the candidate object to be recommended.

[0163] Optionally, in the embodiments of the present application, when the candidate set construction module determines the preference score of the physical object for the candidate object to be recommended according to the preference degree of the physical object for the target object and the association relationship between the target object and the candidate object to be recommended, the candidate set construction module aggregates the preference degrees of the target object according to the association relationship between the target object and the candidate object to be recommended, and obtains the preference score of the physical object for the candidate object to be recommended.

[0164] Optionally, in the embodiments of the present application, when the candidate set construction module determines the matching score of the candidate object to be recommended according to the matching degree between the candidate object to be recommended and the target object, the candidate set construction module aggregates the matching degrees between the candidate object to be recommended and the target object according to the association relationship between the candidate object to be recommended and the target object, and obtains the matching score of the candidate object to be recommended.

[0165] Optionally, in the embodiments of the present application, the index of the candidate object includes identification information of the candidate object or a category of the candidate object; and the screening module is specifically configured to: when the index of the candidate object includes the identification information of the candidate object, screen the candidate object based on the matching degree between the target object and the candidate object, to obtain screened candidate objects; aggregate the candidate objects based on the identification information of the screened candidate objects, to obtain the candidate objects to be recommended to the physical object; or when the index of the candidate object includes the category of the candidate object, screen the candidate objects to be recommended to the physical object based on the category of the candidate object and the matching degree between the target object and the candidate object.

[0166] Optionally, in the embodiments of the present application, the preference determination module is specifically configured to: input the digital object historical interaction data into a second supervised model, perform preference behavior prediction for the physical object by the second supervised model, obtain a digital object preferred by the physical object and take the digital object as the target object, and obtain the preference degree of the physical object for the target object, wherein the digital object historical interaction data includes information of a digital object that has been historically interacted with by the physical object and corresponding historical interaction behavior information.

[0167] Optionally, in the embodiment of the present application, the association determining module is specifically configured to: input the target object into a graph model, and perform, by the graph model, prediction of an association relationship between a pre-stored digital object and the target object based on the pre-stored digital object stored in the database, to obtain a pre-stored digital object having an association relationship with the target object and a matching degree between the pre-stored digital object and the target object; and determine, from the pre-stored digital object having the association relationship with the target object, a pre-stored digital object that has not interacted with the physical object in history as the candidate object, and determine the matching degree between the candidate object and the target object.

[0168] Optionally, in the embodiment of the present application, the graph network model is a GATNE model; the first supervised model is a DIEN model or a DeepMCP model; and the second supervised model is a DIEN model or a DeepMCP model.

[0169] The recommendation device of the embodiment is used to implement the corresponding recommendation method in the foregoing plurality of method embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be described herein again. In addition, the function implementation of each module in the recommendation device of the embodiment can be referred to the description of the corresponding part in the foregoing method embodiments, which will not be described herein again either.

[0170] When the recommendation device of the embodiment is applied to an e-commerce scenario, the following is included in the recommendation device:

[0171] The preference determining module is configured to obtain, based on historical interaction data of a commodity generated by interaction between a user and the commodity, a commodity preferred by the user from commodities that have been interacted with by the user as a target commodity and obtain a preference degree of the corresponding target commodity;

[0172] The association determining module is configured to obtain, from a database for storing commodities, a commodity having an association relationship with the target commodity and not having been interacted with in history as a candidate commodity and obtain a matching degree between the candidate commodity and the target commodity;

[0173] The screening module is configured to screen the candidate commodity to be recommended to the user based on the index of the candidate commodity and the matching degree between the target commodity and the candidate commodity;

[0174] The input module is configured to input the matching degree between the candidate commodity to be recommended and the target commodity, the preference degree of the target commodity by the user, and the candidate commodity to be recommended into a first supervised model, and predict, by the first supervised model, an interest degree of the candidate commodity by the user;

[0175] The determining module is configured to determine the candidate commodity to be recommended to the user according to the interest degree of the candidate commodity.

[0176] When the recommendation device of the embodiment is applied to an advertising scenario, the recommendation device includes:

[0177] a preference determining module configured to obtain, as target advertisements, advertisements preferred by a user from advertisements interacted by the user based on advertisement historical interaction data generated by user interaction with the advertisements, and obtain a preference degree of the corresponding target advertisements;

[0178] an association determining module configured to obtain, as candidate advertisements, advertisements having an association relationship with the target advertisements and not interacted by the user in the past from a database for storing advertisers, and obtain a matching degree of the candidate advertisements and the target advertisements;

[0179] a screening module configured to screen the candidate advertisements to be recommended to the user based on the index of the candidate advertisements and the matching degree of the target advertisements and the candidate advertisements;

[0180] an input module configured to input the matching degree of the target advertisements and the candidate advertisements to be recommended, the preference degree of the target advertisements by the user, and the candidate advertisements to be recommended into a first supervised model, and predict an interest degree of the user for the candidate advertisements by the first supervised model;

[0181] a determining module configured to determine, according to the interest degree, an advertisement to be recommended to the user.

[0182] Embodiment Seven

[0183] Referring to Figure 7 , a structural schematic diagram of an electronic device according to an embodiment of the present application is shown, and the embodiments of the present application do not limit the specific implementation of the electronic device.

[0184] As Figure 7 shown, the electronic device can include a processor 702, a communications interface 704, a memory 706, and a communications bus 708.

[0185] Among them:

[0186] The processor 702, the communications interface 704, and the memory 706 complete mutual communication through the communications bus 708.

[0187] The communications interface 704 is configured to communicate with other electronic devices or servers.

[0188] The processor 702 is configured to execute the program 710, and specifically can execute related steps in the above-mentioned recommendation method embodiments.

[0189] Specifically, the program 710 can include program code including computer operation instructions.

[0190] The processor 702 can be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement one or more embodiments of the application. The one or more processors included in the smart device can be of the same type, such as one or more CPUs, or can be of different types, such as one or more CPUs and one or more ASICs.

[0191] The memory 706 is configured to store a program 710. The memory 706 can include a high-speed RAM memory, and can further include a non-volatile memory, such as at least one disk memory.

[0192] The program 710 can be specifically configured to cause the processor 702 to perform the following operations: based on digital object historical interaction data generated by a physical object interacting with a digital object, obtaining, from the digital object that has been historically interacted with by the physical object, a target object preferred by the physical object and a preference degree of the target object; obtaining, from a database configured to store digital objects, a candidate object that has not been historically interacted with and that has an association relationship with the target object, and obtaining a matching degree between the candidate object and the target object; based on an index of the candidate object and the matching degree between the candidate object and the target object, screening the candidate object to be recommended to the physical object; inputting the matching degree between the candidate object to be recommended and the target object, the preference degree of the physical object to the target object, and the candidate object to be recommended into a first supervised model, and predicting, by the first supervised model, an interest degree of the physical object to the candidate object; and determining, according to the interest degree of the candidate object, the candidate object to be recommended to the physical object.

[0193] In an optional implementation, the program 710 is further configured to cause the processor 702 to, when inputting the matching degree between the candidate object to be recommended and the target object, the preference degree of the physical object to the target object, and the candidate object to be recommended into the first supervised model, and predicting, by the first supervised model, the interest degree of the physical object to the candidate object: construct a candidate object set including the candidate object to be recommended according to the matching degree between the candidate object to be recommended and the target object, the preference degree of the physical object to the target object, and the association relationship between the target object and the candidate object to be recommended; and input the candidate object set into the first supervised model, and predict, by the first supervised model, the interest degree of the physical object to the candidate object.

[0194] In an optional implementation, the program 710 is further configured to cause the processor 702 to, when constructing the candidate object set containing the candidate object to be recommended according to the matching degree of the candidate object to be recommended and the target object, the preference degree of the physical object to the target object, and the association relationship between the target object and the candidate object to be recommended: determine a matching score of the candidate object to be recommended according to the matching degree of the candidate object to be recommended and the target object, and the association relationship between the target object and the candidate object to be recommended; determine a preference score of the physical object to the candidate object to be recommended according to the preference degree of the physical object to the target object, and the association relationship between the target object and the candidate object to be recommended; and construct the candidate object set containing the candidate object to be recommended according to at least the matching score of the candidate object to be recommended and the preference score of the physical object to the candidate object to be recommended.

[0195] In an optional implementation, the program 710 is further configured to cause the processor 702 to determine the number of target objects having an association relationship with the candidate object to be recommended; and when constructing the candidate object set containing the candidate object to be recommended according to at least the matching score of the candidate object to be recommended and the preference score of the physical object to the candidate object to be recommended: construct the candidate object set containing the candidate object to be recommended according to at least the matching score of the candidate object to be recommended, the preference score of the physical object to the candidate object to be recommended, and the number of target objects having an association relationship with the candidate object to be recommended.

[0196] In an optional implementation, the program 710 is further configured to cause the processor 702 to, when determining the preference score of the physical object to the candidate object to be recommended according to the preference degree of the physical object to the target object, and the association relationship between the target object and the candidate object to be recommended: aggregate the preference degrees of the target objects to obtain the preference score of the physical object to the candidate object to be recommended according to the association relationship between the target objects and the candidate object to be recommended.

[0197] In an optional implementation, the program 710 is further configured to cause the processor 702 to, when determining the matching score of the candidate object to be recommended according to the matching degree of the candidate object to be recommended and the target object: aggregate the matching degrees of the candidate object to be recommended and the target objects to obtain the matching score of the candidate object to be recommended according to the association relationship between the candidate object to be recommended and the target objects.

[0198] In an optional implementation, the index of the candidate object includes identification information of the candidate object; and the processor 702 is further caused to, when filtering the candidate objects to be recommended to the physical object based on the index of the candidate object and the matching degree between the target object and the candidate object: filter the candidate objects based on the matching degree between the target object and the candidate object, to obtain filtered candidate objects; and aggregate the filtered candidate objects based on the identification information of the filtered candidate objects, to obtain the candidate objects to be recommended to the physical object.

[0199] In an optional implementation, the index of the candidate object includes a category of the candidate object; and the processor 702 is further caused to, when filtering the candidate objects to be recommended to the physical object based on the index of the candidate object and the matching degree between the target object and the candidate object: filter the candidate objects to be recommended to the physical object based on the category of the candidate object and the matching degree between the target object and the candidate object.

[0200] In an optional implementation, the processor 702 is further caused to, when obtaining, based on digital object historical interaction data generated by the physical object interacting with the digital object, a digital object preferred by the physical object as the target object and a preference degree of the corresponding target object from the digital objects that have been interacted with by the physical object: input the digital object historical interaction data into a second supervised model, perform preference behavior prediction of the physical object by the second supervised model, obtain the digital object preferred by the physical object as the target object, and obtain the preference degree of the physical object to the target object, where the digital object historical interaction data includes information of the digital objects that have been interacted with by the physical object and corresponding historical interaction behavior information.

[0201] In an optional implementation, the processor 702 is further caused to, when obtaining, from a database for storing the digital objects, a digital object that has not been interacted with by the physical object and that has an association relationship with the target object as a candidate object and a matching degree between the candidate object and the target object: input the target object into a graph model, perform, by the graph model, association relationship prediction between a pre-stored digital object and the target object based on pre-stored digital objects stored in the database, obtain the pre-stored digital object that has an association relationship with the target object and a matching degree between the pre-stored digital object and the target object; and determine, from the pre-stored digital object that has an association relationship with the target object, the pre-stored digital object that has not been interacted with by the physical object as the candidate object, and determine the matching degree between the candidate object and the target object.

[0202] In an optional implementation, the graph network model is a GATNE model; the first supervised model is a DIEN model or a DeepMCP model; and the second supervised model is a DIEN model or a DeepMCP model.

[0203] The specific implementation of each step in the program 710 can refer to the corresponding description in the corresponding steps and units in the above-mentioned recommendation method embodiments, and will not be described here. Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working process of the above-mentioned devices and modules can refer to the corresponding process description in the foregoing method embodiments, and will not be described here.

[0204] It should be noted that, according to the needs of implementation, each component / step described in the embodiments of the present application can be split into more components / steps, or two or more components / steps or part of the operation of the components / steps can be combined into a new component / steps, to achieve the purpose of the embodiments of the present application.

[0205] The above-mentioned method according to the embodiments of the present application can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium such as a CD ROM, a RAM, a floppy disk, a hard disk or a magneto-optical disk, or be implemented by computer code originally stored in a remote recording medium or a non-transitory machine-readable medium downloaded through a network and stored in a local recording medium, so that the method described herein can be processed by such software on a recording medium using a general-purpose computer, a special-purpose processor or programmable or special-purpose hardware such as an ASIC or an FPGA. It can be understood that the computer, processor, microprocessor controller or programmable hardware includes a storage component (for example, RAM, ROM, flash memory, etc.) that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, processor or hardware, the recommendation method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the recommendation method shown herein, the execution of the code will convert the general-purpose computer into a special-purpose computer for executing the recommendation method shown herein.

[0206] Those of ordinary skill in the art can realize that the units and method steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of the present application.

[0207] The above embodiments are only used for describing the present application, and are not intended to limit the present application. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present application. Therefore, all equivalent technical solutions belong to the scope of the present application, and the patent protection scope of the present application should be defined by the claims.

Claims

1. A recommendation method, comprising: Based on the historical interaction data of digital objects generated by the interaction between physical objects and digital objects, the digital objects preferred by the physical object are obtained from the digital objects that the physical object has historically interacted with, and the preference degree of the corresponding target object is obtained. From the database used to store digital objects, obtain digital objects that are associated with the target object and have no historical interaction as candidate objects, and obtain the matching degree between the candidate objects and the target object; Based on the index of the candidate objects and the matching degree between the target object and the candidate objects, candidate objects to be recommended to the physical object are filtered; The matching degree between the candidate object to be recommended and the target object, the physical object's preference for the target object, and the candidate object to be recommended are input into the first supervised model, and the first supervised model predicts the physical object's interest in the candidate object; Based on the interest level of the candidate objects, candidate objects for recommendation to the physical object are determined.

2. The method according to claim 1, wherein, The step of inputting the matching degree between the candidate object to be recommended and the target object, the physical object's preference for the target object, and the candidate object to be recommended into a first supervised model, and having the first supervised model predict the physical object's interest in the candidate object, includes: Based on the matching degree between the candidate objects to be recommended and the target object, the physical object's preference for the target object, and the association between the target object and the candidate objects to be recommended, a candidate object set containing the candidate objects to be recommended is constructed. The candidate object set is input into the first supervised model, which then predicts the physical object's interest in the candidate objects.

3. The method according to claim 2, wherein, The step of constructing a candidate object set containing the candidate objects to be recommended based on the matching degree between the candidate objects to be recommended and the target object, the physical object's preference for the target object, and the association relationship between the target object and the candidate objects to be recommended includes: Based on the matching degree between the candidate object to be recommended and the target object, and the association relationship between the candidate object to be recommended and the target object, the matching score of the candidate object to be recommended is determined; Based on the physical object's preference for the target object and the association between the target object and the candidate object to be recommended, the preference score of the physical object for the candidate object to be recommended is determined. A candidate object set containing the candidate objects to be recommended is constructed based at least on the matching score and the preference score of the candidate objects to be recommended.

4. The method according to claim 3, wherein, The method further includes: Determine the number of target objects that are associated with the candidate objects to be recommended; The step of constructing a candidate object set containing the candidate objects to be recommended, based at least on the matching score and the preference score of the candidate objects to be recommended, includes: A candidate object set containing the candidate object to be recommended is constructed based at least on the matching score of the candidate object to be recommended, the preference score of the candidate object to be recommended, and the number of target objects that are associated with the candidate object to be recommended.

5. The method according to claim 3, wherein, The step of determining the preference score of the physical object for the candidate object to be recommended based on the physical object's preference for the target object and the association between the target object and the candidate object to be recommended includes: Based on the association between the target object and the candidate object to be recommended, the preference degree of the target object is aggregated to obtain the preference score of the physical object for the candidate object to be recommended.

6. The method according to claim 3, wherein, The step of determining the matching score of the candidate object to be recommended based on the matching degree between the candidate object to be recommended and the target object includes: Based on the association between the candidate objects to be recommended and the target objects, the matching degree between the candidate objects to be recommended and the target objects is aggregated to obtain the matching score of the candidate objects to be recommended.

7. The method according to claim 1, wherein, The index of the candidate object includes: the identification information of the candidate object; The process of filtering candidate objects to be recommended to the physical object based on the index of the candidate object and the matching degree between the target object and the candidate object includes: Based on the matching degree between the target object and the candidate object, the candidate object is filtered to obtain the filtered candidate object; Based on the identification information of the filtered candidate objects, the filtered candidate objects are aggregated to obtain candidate objects to be recommended to the physical object.

8. The method according to claim 1, wherein, The index of the candidate objects includes: the category of the candidate objects; The process of filtering candidate objects to be recommended to the physical object based on the index of the candidate object and the matching degree between the target object and the candidate object includes: Based on the category of the candidate objects and the matching degree between the target object and the candidate objects, candidate objects to be recommended to the physical object are selected.

9. The method according to claim 1, wherein, The historical interaction data of digital objects generated based on the interaction between physical objects and digital objects, from the digital objects that the physical object has historically interacted with, obtains the digital objects preferred by the physical object as target objects and obtains the preference degree of the corresponding target objects, including: The historical interaction data of the digital object is input into the second supervised model, which predicts the preference behavior of the physical object, obtains the digital object preferred by the physical object, and uses the digital object as the target object. Furthermore, the preference degree of the physical object for the target object is obtained. The historical interaction data of the digital object includes information on the digital objects that the physical object has interacted with in the past and the corresponding historical interaction behavior information.

10. The method according to claim 1, wherein, The step of obtaining candidate digital objects that are associated with the target object and have no historical interaction from the database used to store the digital objects, and obtaining the matching degree between the candidate objects and the target object, includes: The target object is input into the graph model, which predicts the association between the pre-stored digital objects and the target object based on the pre-stored digital objects stored in the database, thereby obtaining the pre-stored digital objects that are associated with the target object and the matching degree between the pre-stored digital objects and the target object; From the pre-stored digital objects that are associated with the target object, pre-stored digital objects that have not been interacted with in the history of the physical object are identified as candidate objects, and the matching degree between the candidate objects and the target object is determined.

11. A recommendation method, comprising: Based on the historical interaction data of products generated by user-product interaction, the products that the user has interacted with in the past are obtained as target products and the corresponding preference degree of the target products are obtained. From the database used to store goods, obtain goods that are related to the target product and have no historical interaction as candidate goods, and obtain the matching degree between the candidate goods and the target product; Based on the index of the candidate products and the matching degree between the target product and the candidate products, candidate products to be recommended to the user are selected; The matching degree between the candidate products to be recommended and the target product, the user's preference for the target product, and the candidate products to be recommended are input into the first supervised model, and the first supervised model predicts the user's interest in the candidate products. Based on the level of interest, candidate products to be recommended to the user are determined.

12. A recommendation method, comprising: Based on the historical interaction data of users and advertisements, the advertisements that the user has interacted with in the past are obtained as target advertisers and the preference degree of the corresponding target advertisements are obtained. From the database used to store advertisements, obtain advertisements that are related to the target advertisement and have no historical interaction as candidate advertisements, and obtain the matching degree between the candidate advertisements and the target advertisement; Based on the index of the candidate ads and the matching degree between the target ad and the candidate ads, candidate ads to be recommended to the user are filtered. The matching degree between the candidate advertisement to be recommended and the target advertisement, the user's preference for the target advertisement, and the candidate advertisement to be recommended are input into the first supervised model, and the first supervised model predicts the user's interest in the candidate advertisement. Based on the level of interest, the advertisements recommended to the user are determined.

13. A recommended device, comprising: The preference determination module is used to obtain the digital objects preferred by the physical object as target objects and obtain the preference degree of the corresponding target objects from the historical interaction data of digital objects generated by the interaction between physical objects and digital objects. The association determination module is used to obtain, from the database used to store digital objects, digital objects that are associated with the target object and have not been interacted with in the past as candidate objects, and to obtain the matching degree between the candidate objects and the target object; The filtering module is used to filter candidate objects to be recommended to the physical object based on the index of the candidate object and the matching degree between the target object and the candidate object; The input module is used to input the matching degree between the candidate object to be recommended and the target object, the physical object's preference for the target object, and the candidate object to be recommended into the first supervised model, and the first supervised model predicts the physical object's interest in the candidate object; The determination module is used to determine candidate objects to recommend to the physical object based on the interest level of the candidate objects.

14. A recommended device, comprising: The preference determination module is used to obtain the products that the user prefers as target products and obtain the preference degree of the corresponding target products from the products that the user has interacted with in the past, based on the historical interaction data of the products generated by the user's interaction with the products. The association determination module is used to obtain, from the database used to store goods, goods that are associated with the target goods and have not been interacted with in the past as candidate goods, and to obtain the matching degree between the candidate goods and the target goods; The filtering module is used to filter candidate products to be recommended to the user based on the index of the candidate products and the matching degree between the target product and the candidate products; The input module is used to input the matching degree between the candidate products to be recommended and the target product, the user's preference for the target product, and the candidate products to be recommended into the first supervised model, and the first supervised model predicts the user's interest in the candidate products; The determination module is used to determine candidate products to recommend to the user based on the degree of interest.

15. A recommended device, comprising: The preference determination module is used to obtain the ads that the user prefers as target ads and obtain the preference degree of the corresponding target ads from the ads that the user has interacted with in the past, based on the historical interaction data of the user and the ads generated by the user's interaction with the ads. The association determination module is used to obtain, from the database used for storing advertisements, advertisements that are associated with the target advertisement and have no historical interaction as candidate advertisements, and to obtain the matching degree between the candidate advertisements and the target advertisement; A filtering module is used to filter candidate ads to be recommended to the user based on the index of the candidate ads and the matching degree between the target ad and the candidate ads; The input module is used to input the matching degree between the candidate advertisement to be recommended and the target advertisement, the user's preference for the target advertisement, and the candidate advertisement to be recommended into the first supervised model, and the first supervised model predicts the user's interest in the candidate advertisement; The determination module is used to determine the advertisements to be recommended to the user based on the level of interest.

16. A computer storage medium having a computer program stored thereon, which, when executed by a processor, implements the recommended method as described in any one of claims 1-12.

Images