Recommendation processing method, related device and medium

Abstract

The invention relates to the field of big data, in particular to a recommendation processing method, a related device and a medium. The recommendation processing method comprises the steps of obtaining a plurality of candidate recommendation objects of a target object based on object relation data and object interaction data in a plurality of scenes in an application; obtaining the relationship intimacy between each candidate recommendation object and the target object; predicting a conversion rate after each candidate recommendation object is recommended to the target object; obtaining preference conformity of each candidate recommendation object to the target object; and sorting the plurality of candidate recommendation objects based on the relationship intimacy, the conversion rate and the preference conformity for display to the target object. According to the embodiment of the invention, the sorting accuracy of the application in multiple scenes is improved, and the conversion rate of recommendation processing is improved. The embodiment of the invention can be applied to various scenes such as big data and data processing.

Description

Technical Field

[0001] The present disclosure relates to the field of big data, and in particular to a recommendation processing method, related devices and media. Background Art

[0002] In big data processing, object recommendation refers to recommending objects that the target object may be willing to establish associations (such as adding friends) to the target object, thereby increasing the object association rate of the application. This is of great significance for improving the relationship chain coverage and activity of the application. At present, object recommendation generally goes through the stages of recall and sorting. In the recall stage, candidate objects that the target object may be interested in are obtained. In the sorting stage, the candidate objects recalled in the recall stage are sorted and displayed to the target object. At present, the sorting stage is mainly based on a single standard for sorting.

[0003] Currently, many applications integrate a large number of scenarios. For example, in addition to instant conversations, instant messaging applications may also integrate scenarios such as object groups, channels, spaces, and short videos. The sorting stage is based on a single standard. In the case of multiple scenarios, due to the complex relationships between objects in multiple scenarios and the complex relationships with scenario items (such as content, author), the sorting accuracy is poor and cannot effectively promote the conversion rate of recommendation processing (the probability of the target object establishing an association with the recommended object after being recommended to the target object). Summary of the invention

[0004] The disclosed embodiments provide a recommendation processing method, a related device and a medium, which can improve the sorting accuracy in multiple application scenarios and improve the conversion rate of recommendation processing.

[0005] According to one aspect of the present disclosure, a recommendation processing method is provided, comprising:

[0006] Based on object relationship data and object interaction data in multiple scenarios in the application, multiple candidate recommended objects of the target object are obtained;

[0007] Obtaining the degree of intimacy between each candidate recommendation object and the target object;

[0008] Predicting a conversion rate after recommending each of the candidate recommended objects to the target object, wherein the conversion rate is a probability that the target object pays attention to the candidate recommended object after the candidate recommended object is recommended to the target object;

[0009] Obtaining the preference conformity of each candidate recommendation object to the target object;

[0010] Based on the relationship intimacy, the conversion rate, and the preference conformity, the plurality of candidate recommendation objects are sorted for display to the target object.

[0011] According to one aspect of the present disclosure, there is provided a recommendation processing device, comprising:

[0012] A first acquisition unit, configured to acquire multiple candidate recommended objects of a target object based on object relationship data and object interaction data in multiple scenes in an application;

[0013] A second acquisition unit is used to acquire the relationship intimacy between each of the candidate recommendation objects and the target object;

[0014] A prediction unit, configured to predict a conversion rate after recommending each of the candidate recommended objects to the target object, wherein the conversion rate is a probability that the target object pays attention to the candidate recommended object after the candidate recommended object is recommended to the target object;

[0015] A third acquisition unit is used to acquire the preference conformity of each candidate recommendation object to the target object;

[0016] A sorting unit is used to sort the multiple candidate recommendation objects based on the relationship intimacy, the conversion rate, and the preference compliance, so as to display them to the target object.

[0017] Optionally, the second acquiring unit is used to:

[0018] Obtaining the attribute similarity between each of the candidate recommended objects and the target object;

[0019] Obtaining the interaction frequency between each of the candidate recommendation objects and the target object in the application;

[0020] Obtaining the number of commonly associated objects between each of the candidate recommended objects and the target object in the application;

[0021] Obtaining a main interaction type between each of the candidate recommendation objects and the target object in the application;

[0022] Based on the attribute similarity, the interaction frequency, the number of commonly associated objects, and the main interaction type, the relationship closeness between each of the candidate recommended objects and the target object is determined.

[0023] Optionally, the second acquiring unit is specifically configured to:

[0024] Acquire first attribute values ​​of multiple first attributes of each of the candidate recommendation objects and second attribute values ​​of multiple first attributes of the target object;

[0025] generating a first vector based on the first attribute values ​​of the plurality of first attributes of the candidate recommendation objects, and generating a second vector based on the second attribute values ​​of the plurality of first attributes of the target object;

[0026] The attribute similarity is obtained based on the distance between the first vector and the second vector.

[0027] Optionally, the second acquiring unit is specifically configured to:

[0028] Identify the interaction between the candidate recommendation object and the target object in the application within a predetermined period before the current time;

[0029] Counting the number of interactions within the predetermined period;

[0030] The interaction frequency is determined based on the counted number of interactions and the predetermined period length.

[0031] Optionally, the second acquiring unit is specifically configured to:

[0032] In a multi-scenario element interconnection graph generated based on the object relationship data and the object interaction data, identifying other object element nodes adjacent to the object element node of the candidate recommended object as associated objects of the candidate recommended object in the application, and identifying other object element nodes adjacent to the object element node of the target object as associated objects of the target object in the application;

[0033] The number of objects in the intersection of the associated objects of the candidate recommended object in the application and the associated objects of the target object in the application is determined as the number of commonly associated objects.

[0034] Optionally, the second acquiring unit is specifically configured to:

[0035] Identify the interaction between the candidate recommendation object and the target object in the application within a predetermined period before the current time;

[0036] Obtaining the identified interaction type of the interaction;

[0037] Determine the number of interactions of each interaction type captured;

[0038] Based on the number of interactions, the main interaction type is determined among the identified interaction types.

[0039] Optionally, the second acquiring unit is specifically configured to:

[0040] Based on the attribute similarity, determining a first score of the candidate recommendation object;

[0041] Based on the interaction frequency, determining a second score of the candidate recommendation object;

[0042] Determining a third score of the candidate recommendation object based on the number of commonly associated objects;

[0043] Determining a fourth score of the candidate recommendation object based on the main interaction type;

[0044] Based on the first score, the second score, the third score, and the fourth score of each of the candidate recommendation objects, the closeness of the relationship between the candidate recommendation object and the target object is determined.

[0045] Optionally, the prediction unit is used to:

[0046] Acquire a first object interest tag of the target object;

[0047] Obtain a second object interest tag for each of the candidate recommendation objects;

[0048] The first object interest tag and the second object interest tag are input into a conversion rate prediction model to obtain the predicted conversion rate.

[0049] Optionally, the third acquiring unit is used to:

[0050] Obtaining historical recommendation records for the target object;

[0051] Acquire multiple post-conversion recommended objects in which the target object has been converted in the historical recommendation record;

[0052] Based on the attributes of the plurality of the converted recommended objects, clustering the plurality of the converted recommended objects to form a plurality of recommended object classes;

[0053] generating a conversion preference label for the target object based on a recommended object class that meets a predetermined standard among the plurality of recommended object classes;

[0054] The preference compliance is obtained based on the matching of the conversion preference label of the target object and the attributes of the candidate recommendation object.

[0055] Optionally, the sorting unit is used to:

[0056] Based on the relationship intimacy, the conversion rate, and the preference conformity, obtaining a first ranking of the plurality of candidate recommendation objects;

[0057] Obtaining object attributes of a plurality of candidate recommendation objects;

[0058] If, for each of the object attributes, among the first first number of the candidate recommendation objects in the first sorting, the number of the candidate recommendation objects carrying the object attribute is greater than a second number, then the number of the candidate recommendation objects carrying the object attribute is reduced to no greater than the second number, thereby obtaining a second sorting, wherein the second number is less than the first number;

[0059] Based on the second ranking, a plurality of the candidate recommendation objects are displayed.

[0060] Optionally, the sorting unit is specifically used to:

[0061] Taking the relationship intimacy, the conversion rate, and the preference conformity as multiple targets of the recommendation process, and using multi-target fusion to determine a first weight of the relationship intimacy, a second weight of the conversion rate, and a third weight of the preference conformity;

[0062] Based on the first weight, the second weight, and the third weight, performing a weighted sum of the relationship intimacy, the conversion rate, and the preference compliance;

[0063] Based on the weighted sum, the first sorting is performed on the plurality of candidate recommendation objects.

[0064] Optionally, the first acquiring unit is used to:

[0065] Based on the object relationship data and the object interaction data in the multiple scenes in the application, a multi-scene element interconnection graph is generated, wherein the multi-scene element interconnection graph includes multiple element nodes, the multiple element nodes include object element nodes and auxiliary element nodes, the object element nodes and the relationship between each of the object element nodes are generated by the object relationship data, and the auxiliary element nodes and the relationship between the auxiliary element nodes and the object element nodes are generated by the object interaction data;

[0066] Based on the interconnected graph of multiple scene elements and the target object, multiple candidate recommended objects for the target object are obtained.

[0067] Optionally, the auxiliary element node includes a content author node and an object interest tag node; and the first acquisition unit is specifically used to:

[0068] Acquire a first candidate recommendation object based on the object element node, the content author node, and the target object;

[0069] Based on the object interest tag node and the target object, obtaining a second candidate recommended object;

[0070] The interconnected graph of multiple scene elements and the target object input relationship and interest fusion recommendation model are used to obtain a third candidate recommendation object;

[0071] The first candidate recommendation object, the second candidate recommendation object, and the third candidate recommendation object are integrated into a plurality of the candidate recommendation objects.

[0072] Optionally, the sorting unit is used to:

[0073] sorting the plurality of candidate recommendation objects;

[0074] While displaying the candidate recommendation objects according to the sorting, sources of the candidate recommendation objects are displayed, and the sources include: the process of obtaining a first candidate recommendation object based on the object element node, the content author node, and the target object; the process of obtaining a second candidate recommendation object based on the object interest tag node and the target object; and the process of fusing the multi-scenario element interconnection graph and the target object input relationship with the interest recommendation model to obtain a third candidate recommendation object.

[0075] Optionally, the first acquiring unit is specifically configured to:

[0076] In the multi-scenario element interconnection graph, the object element node having a common associated object element node with the object element node of the target object is obtained, and the object corresponding to the obtained object element node is used as the first sub-candidate recommendation object;

[0077] In each object element node of the multi-scenario element interconnection graph, the object element node having the same specific attribute as the target object is obtained, and the object corresponding to the obtained object element node is used as the second sub-candidate recommendation object;

[0078] Based on the attributes of each object element node of the multi-scenario element interconnection graph, group each object element node, obtain the object element nodes that are grouped in the same group as the object element node of the target object, and use the object corresponding to the obtained object element node as the third sub-candidate recommendation object;

[0079] From the objects corresponding to the object element nodes of the multi-scenario element interconnection graph, an object belonging to the same application session group as the target object is obtained as a fourth sub-candidate recommendation object;

[0080] In the multi-scenario element interconnection graph, the content author node that has an interactive relationship with the object element node of the target object is obtained, and the object corresponding to the obtained content author node is used as the fifth sub-candidate recommendation object;

[0081] Based on the list of associated objects of the target object outside the application, obtaining a sixth sub-candidate recommended object;

[0082] In the multi-scenario element interconnection graph, the content author node that is mutually concerned with the object element node of the target object is obtained, and the object corresponding to the obtained content author node is used as the seventh sub-candidate recommendation object;

[0083] The first sub-candidate recommendation object, the second sub-candidate recommendation object, the third sub-candidate recommendation object, the fourth sub-candidate recommendation object, the fifth sub-candidate recommendation object, the sixth sub-candidate recommendation object, and the seventh sub-candidate recommendation object are integrated into the first candidate recommendation object.

[0084] Optionally, the auxiliary element node includes a content node; and the first acquisition unit is specifically configured to:

[0085] Based on the label similarity of the object interest label node of the target object and the object interest label nodes of each other object, obtaining an eighth sub-candidate recommended object;

[0086] Based on the interaction data of the target object node with each of the content nodes in the multi-scenario element interconnection graph, a first point of interest vector of the target object node is obtained; based on the interaction data of each of the other object nodes with each of the content nodes in the multi-scenario element interconnection graph, a second point of interest vector of each of the other object nodes is obtained; and based on a comparison between the first point of interest vector and the second point of interest vector, a ninth sub-candidate recommendation object is obtained;

[0087] The eighth sub-candidate recommendation object and the ninth sub-candidate recommendation object are integrated into the second candidate recommendation object.

[0088] According to one aspect of the present disclosure, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the recommendation processing method as described above when executing the computer program.

[0089] According to one aspect of the present disclosure, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the recommendation processing method as described above is implemented.

[0090] According to one aspect of the present disclosure, a computer program product is provided. The computer program product includes a computer program. The computer program is read and executed by a processor of a computer device, so that the computer device executes the recommendation processing method as described above.

[0091] The disclosed embodiment obtains multiple candidate recommended objects for a target object based on object relationship data and object interaction data in multiple scenes. When sorting multiple candidate recommended objects, not only the relationship intimacy between the candidate recommended objects and the target object is considered, but also the predicted conversion rate and the target object's preference compliance are considered. Relationship intimacy represents the distance between objects in multiple scenes based on the object relationship data in multiple scenes, but multiple scenes are not necessarily scenes that are equally important to the target object. Although it is determined based on the object relationship data in multiple scenes that the relationship between objects may be very close, it may not meet the target object's preference. And according to the various performances of the target object before, it may be predicted that the target object does not have a very high conversion rate for an object that seems to have a close relationship with it. Therefore, the disclosed embodiment sorts multiple candidate recommended objects based on multiple indicators such as relationship intimacy, conversion rate, and preference compliance, which improves the sorting accuracy in multiple scenes of the application and improves the conversion rate of the recommendation process.

[0092] Other features and advantages of the present disclosure will be described in the following description, and partly become apparent from the description, or understood by practicing the present disclosure. The purpose and other advantages of the present disclosure can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] The accompanying drawings are used to provide further understanding of the technical solution of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the technical solution of the present disclosure and do not constitute a limitation on the technical solution of the present disclosure.

[0094] Figure 1 is a system architecture diagram of a recommended processing method of an embodiment of the present disclosure;

[0095] Figure 2A -E is a schematic diagram of an interface in which the recommendation processing method of an embodiment of the present disclosure is applied to a friend recommendation scenario;

[0096] Figure 3 is a flowchart of a recommendation processing method according to an embodiment of the present disclosure;

[0097] Figure 4 yes Figure 3 A specific flow chart of step 310;

[0098] Figure 5 yes Figure 4 A schematic diagram of a multi-scenario element interconnection graph of step 410;

[0099] Figure 6 yes Figure 4 A specific flow chart of step 410;

[0100] Fig. 7A yes Figure 6 A schematic diagram of a specific implementation process of generating object element node data in step 620;

[0101] Figure 7B yes Figure 6 A schematic diagram of a specific implementation process of generating auxiliary element node data in step 620;

[0102] Figure 8 yes Figure 6 A specific flow chart of step 620 in FIG.

[0103] Fig. 9 yes Figure 6 Another specific implementation process diagram of step 620 generating auxiliary element node data;

[0104] Fig.10 yes Figure 6 A schematic diagram of a specific implementation process of generating a multi-scenario element interconnection graph in step 630;

[0105] Fig.11 yes Figure 6 Another specific implementation process diagram of step 630 generating a multi-scenario element interconnection graph;

[0106] Fig.12 yes Figure 4 A specific flow chart of step 420;

[0107] Fig.13 yes Fig.12 A schematic diagram of a specific implementation process of obtaining multiple candidate recommendation objects;

[0108] Fig.14 yes Fig.12 A specific flow chart of step 1210 in FIG.

[0109] Fig.15 yes Fig.12 A specific flow chart of step 1220 in FIG.

[0110] Fig.16 yes Figure 3 A specific flow chart of step 320 in FIG.

[0111] Fig.17 yes Fig.16 A schematic diagram of a specific implementation process of determining attribute similarity in step 1610;

[0112] Fig.18 yes Fig.16 A schematic diagram of a specific implementation process of determining the interaction frequency in step 1620;

[0113] Fig.19 yes Fig.16 A schematic diagram of a specific implementation process of determining the commonly associated object data in step 1630;

[0114] Fig. 20 yes Fig.16 A schematic diagram of a specific implementation process of determining the main interaction type in step 1640;

[0115] Fig.21 yes Fig.16 A specific flow chart of step 1650;

[0116] Fig. 22 yes Figure 3 A specific flow chart of step 330;

[0117] Fig.23 yes Figure 3 A specific flow chart of step 340;

[0118] Fig.24 yes Figure 3 A schematic diagram of a specific implementation process of obtaining the preference compliance in step 340;

[0119] Fig.25 yes Figure 3 A first specific flow chart of step 350;

[0120] Fig.26 yes Fig.25 A specific flow chart of step 2510;

[0121] Fig. 27 yes Figure 3 A second specific flow chart of step 350;

[0122] Fig.28 yes Fig. 27 A schematic diagram of a display interface showing candidate recommendation objects and sources;

[0123] Fig.29 yes Figure 3 A third specific flow chart of step 350;

[0124] Fig.30 is a schematic diagram of the overall process of a recommended processing method according to an embodiment of the present disclosure;

[0125] Fig.31 It is a schematic diagram of the implementation process of the recommendation processing method according to the embodiment of the present disclosure in the friend recommendation scenario;

[0126] Fig.32 is a module diagram of a recommendation processing device according to an embodiment of the present disclosure;

[0127] Fig.33 According to the embodiment of the present disclosure Figure 3 The terminal structure diagram of the recommended processing method shown;

[0128] Fig.34 According to the embodiment of the present disclosure Figure 3 The server structure diagram of the recommended processing method is shown. DETAILED DESCRIPTION

[0129] In order to make the purpose, technical solution and advantages of the present disclosure more clear, the present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present disclosure and are not used to limit the present disclosure.

[0130] It should be noted that in various specific implementations of the present disclosure, when it comes to the need to perform relevant processing based on data related to the characteristics of the target object such as the target object attribute information or attribute information set, the permission or consent of the target object will be obtained first, and the collection, use and processing of these data will comply with relevant laws, regulations and standards. In addition, when the embodiment of the present disclosure needs to obtain the attribute information of the target object, the separate permission or separate consent of the target object will be obtained through a pop-up window or jump to a confirmation page. After clearly obtaining the separate permission or separate consent of the target object, the necessary target object-related data for the normal operation of the embodiment of the present disclosure will be obtained.

[0131] Before further describing the embodiments of the present disclosure in detail, the nouns and terms involved in the embodiments of the present disclosure are described. The nouns and terms involved in the embodiments of the present disclosure are subject to the following interpretations:

[0132] Multiple scenarios in an application: refers to multiple scenarios of different services under the same application. For example, in an instant messaging application, there are multiple scenarios such as object groups, channels, spaces, and short videos.

[0133] Object recommendation refers to recommending objects that the target object may be willing to establish associations with (such as adding friends) to improve the object association rate of the application. Currently, object recommendation generally goes through the stages of recall and sorting. In the recall stage, candidate objects that the target object may be interested in are obtained. In the sorting stage, the candidate objects recalled in the recall stage are sorted and displayed to the target object. Currently, the sorting stage mainly sorts based on a single criterion.

[0134] In the case of multiple scenes, due to the complex relationships between objects in multiple scenes and the complex relationships with scene items (such as content, author), the sorting accuracy is poor and cannot effectively promote the conversion rate of recommendation processing (the probability of establishing an association between the target object and the recommended object after being recommended to the target object).

[0135] System architecture and scenario description of the application of the embodiments of the present disclosure

[0136] Figure 1 1 is a system architecture diagram of a recommendation processing method according to an embodiment of the present disclosure, which includes a recommendation processing server 110, the Internet 120, and a target terminal 130.

[0137] The recommendation processing server 110 refers to a computer system that can provide recommendation processing services to the target terminal 130. Compared with the target terminal 130, the recommendation processing server 110 has higher requirements in terms of stability, security, performance, etc. The recommendation processing server 110 can be a high-performance computer in a network platform, a cluster of multiple high-performance computers, a part of a high-performance computer (such as a virtual machine), a combination of parts of multiple high-performance computers (such as virtual machines), etc. The recommendation processing server 110 can also communicate with the Internet 120 in a wired or wireless manner to exchange data.

[0138] The object terminal 130 is a device used to obtain object relationship data and object interaction data in multiple scenes in the application to perform recommendation processing. It includes various forms such as touch screen terminals, desktop computers, laptops, PDAs (personal digital assistants), mobile phones, car terminals, home theater terminals, and dedicated terminals. In addition, it can be a single device or a collection of multiple devices. For example, multiple devices are connected through a local area network and share a display device to work together, forming a terminal. The terminal can also communicate with the Internet 120 in a wired or wireless manner to exchange data.

[0139] The disclosed embodiments can be applied to any scenario where a recommendation processing service is used, especially Figure 2A -Friend recommendation scenarios shown in E, etc.

[0140] like Figure 2A As shown, the target terminal 130 is a mobile terminal. The target terminal 130 displays multiple applications on the interface, including an instant messaging application. When the target sees this interface on the target terminal 130, it can click the "instant messaging application" button, and then the target terminal 130 will display the interface of the instant messaging application.

[0141] like Figure 2B As shown in FIG. 1 , after the object clicks the “instant messaging application” button, the object terminal 130 displays the object’s avatar and conversation record in the instant messaging application. In the interface displayed by the object terminal 130, the instant messaging application is currently in the “message” scene ( Figure 2BIn addition to the "message" scene, the instant messaging application also includes multiple scenes such as "small world", "contacts", and "dynamics". When the object sees this interface on the object terminal 130, he can click the "contacts" button, and then the object terminal 130 will display the interface about contacts.

[0142] like Figure 2C As shown, after the subject clicks the "Contact" button on the subject terminal 130, the subject terminal 130 will start the subject recommendation service and recommend some people that the subject may want to know. The subject terminal 130 recommends two recommended friends for the subject, recommended friend 1 is "Fei Xiang De Qiu Qiu" and recommended friend 2 is "Chong Er". After the subject sees this interface on the subject terminal 130, he can click the "Add" buttons corresponding to the two recommended friends to add friends.

[0143] like Figure 2D As shown, when the subject clicks the "Add" button corresponding to "Flying Qiuqiu", the subject terminal 130 will send an add request to the recommended friend 1 "Flying Qiuqiu" and display a prompt pop-up window on the interface. The content of the prompt pop-up window is "An add request has been sent to Feixiang Qiuqiu, and the other party's consent needs to be waited for."

[0144] like Figure 2E As shown, after the recommended friend 1 "Flying Qiuqiu" agrees to the adding request, "Flying Qiuqiu" will become the friend of the object. The object terminal 130 interface displays "Flying Qiuqiu" in the friend list of the object, and deletes the recommended friend 1 "Flying Qiuqiu", but retains the recommended friend 2 "Chonger".

[0145] It should be understood that the above content only illustrates some application scenarios of the present disclosure, and the business scenarios to which the present disclosure can be applied may include, but are not limited to, the specific embodiments listed above.

[0146] General description of the disclosed embodiments

[0147] It should be emphasized that the embodiments of the present disclosure can be applied to a variety of application scenarios, such as Figure 2A -E's friend recommendation scenario. The recommendation processing of the related art has problems such as low sorting accuracy and low recommendation processing conversion rate. According to some embodiments of the present disclosure, a recommendation processing method, a related device and a medium are provided, which can improve the sorting accuracy in multiple application scenarios and improve the conversion rate of recommendation processing.

[0148] The recommendation processing method is a method for displaying multiple candidate recommendation objects to a target object so that the target object can perform object conversion on the candidate recommendation objects. It includes displaying a candidate recommendation object when no candidate recommendation object was originally displayed, and changing some candidate recommendation objects to other candidate recommendation objects when some candidate recommendation objects were originally displayed. The latter case is the updating of the candidate recommendation objects. The candidate recommendation object refers to the object recommended to the target object. The target object refers to the object that calls the recommendation processing service.

[0149] The recommendation processing method of the embodiment of the present disclosure may be executed on the target terminal 130 , may be executed on the recommendation processing server 110 , or may be partially executed on the target terminal 130 and the other partially executed on the recommendation processing server 110 .

[0150] like Figure 3 As shown, the recommendation processing method according to an embodiment of the present disclosure may include:

[0151] Step 310: Based on the object relationship data and object interaction data in multiple scenes in the application, obtain multiple candidate recommended objects for the target object;

[0152] Step 320: Obtain the relationship intimacy between each candidate recommendation object and the target object;

[0153] Step 330: predicting the conversion rate after recommending each candidate recommended object to the target object, where the conversion rate is the probability that the target object pays attention to the candidate recommended object after the candidate recommended object is recommended to the target object;

[0154] Step 340: Obtain the preference conformity of each candidate recommendation object to the target object;

[0155] Step 350: sort multiple candidate recommendation objects based on relationship intimacy, conversion rate, and preference compliance for display to the target object.

[0156] Steps 310 to 350 are described in detail below.

[0157] In step 310, multiple candidate recommended objects for the target object are obtained based on the object relationship data and object interaction data in multiple scenes in the application. An application refers to software or a program that improves a certain function, such as an instant messaging application. The multiple scenes in the application here are different from the friend recommendation scene described above. The multiple scenes in the application mainly refer to the multiple business modules or business scenes under a certain application. For example, in an instant messaging application, multiple scenes include scenes such as object groups, channels, spaces, and short videos.

[0158] The object relationship data in multiple scenarios in an application specifically refers to data used to characterize the relationship between objects in different scenarios of the same application. For example, the object relationship data includes, in an instant messaging application, object 1 and object 2 are related objects, object 2 and object 3 are related objects, object 2 and object 4 are related objects, and object 4 and object 5 are related objects.

[0159] The object interaction data in multiple scenes in an application specifically refers to data used to characterize the interactive behavior of objects in different scenes of the same application. Interaction behaviors include, for example, likes, comments, messages, etc. For example, the object interaction data includes, in the space scene of the instant messaging application, object 2 likes the space of object 3, and object 3 leaves a message for the space of object 4. For another example, the object interaction data includes, in the short video scene of the instant messaging application, object 1 comments on the short video posted by object 5, and object 3 likes the short video posted by object 2.

[0160] Based on the object relationship data and the object interaction data, multiple candidate recommendation objects for the target object can be obtained. For example, taking object 1 as the target object, based on the object relationship data above, it can be known that object 2 and the target object are mutually related objects, and object 3 and object 4 are both mutually related objects with object 2. Therefore, based on the object relationship data, object 3 and object 4 can be used as candidate recommendation objects. Then based on the object interaction data above, it can be known that the target object commented on the short video posted by object 5. Therefore, based on the object interaction data, object 5 can be used as a candidate recommendation object. Finally, based on the object relationship data and the object interaction data, it is jointly determined that object 3 is the candidate recommendation object 1, object 4 is the candidate recommendation object 2, and object 5 is the candidate recommendation object 3.

[0161] After obtaining multiple candidate recommendation objects for the target object, it is necessary to obtain the relationship intimacy, conversion rate, and preference compliance, and then sort the multiple candidate recommendation objects.

[0162] Regarding the relationship intimacy, in step 320, the relationship intimacy between each candidate recommendation object and the target object is obtained. The relationship intimacy refers to the degree of intimacy between the candidate recommendation object and the target object. The higher the relationship intimacy, the higher the ranking of the candidate recommendation object.

[0163] For example, the relationship intimacy between candidate recommendation object 1 and the target object is 0.8, the relationship intimacy between candidate recommendation object 2 and the target object is 0.6, and the relationship intimacy between candidate recommendation object 3 and the target object is 0.2.

[0164] Regarding the conversion rate, in step 330, the conversion rate after each candidate recommendation object is recommended to the target object is predicted. The conversion rate is the probability that the target object pays attention to the candidate recommendation object after the candidate recommendation object is recommended to the target object. The higher the conversion rate, the higher the ranking of the candidate recommendation object.

[0165] In one example, for the candidate recommendation objects 1-3 in the above example, the conversion rates after the candidate recommendation objects 1-3 are recommended to the target object can be predicted. For example, the conversion rate of the target object to the candidate recommendation object 1 is predicted to be 0.7, the conversion rate of the target object to the candidate recommendation object 2 is predicted to be 0.5, and the conversion rate of the target object to the candidate recommendation object 3 is predicted to be 0.1.

[0166] Regarding the preference conformity, in step 340, the preference conformity of each candidate recommendation object to the target object is obtained. The preference conformity refers to the degree to which the candidate recommendation object conforms to the personalized preferences of the target object. If the preference conformity of a candidate recommendation object is higher, it means that the target object is more likely to pay attention to the candidate recommendation object, and the candidate recommendation object is ranked higher.

[0167] In one example, for the candidate recommendation objects 1-3 in the above example, the preference conformity of the candidate recommendation objects 1-3 to the target object can be obtained. For example, the preference conformity of the candidate recommendation object 1 to the target object is 0.3, the preference conformity of the candidate recommendation object 2 to the target object is 0.8, and the preference conformity of the candidate recommendation object 1 to the target object is 0.6.

[0168] After acquiring the relationship intimacy, conversion rate, and preference conformity, in step 350, multiple candidate recommendation objects are sorted based on the relationship intimacy, conversion rate, and preference conformity for display to the target object.

[0169] In one example, for the candidate recommendation objects 1-3 mentioned above, as well as the acquired relationship intimacy, conversion rate, and preference compliance, the candidate recommendation objects 1-3 can be sorted. Specifically, candidate recommendation object 1 {relationship intimacy is 0.8, conversion rate is 0.7, preference compliance is 0.3}, candidate recommendation object 2 {relationship intimacy is 0.6, conversion rate is 0.5, preference compliance is 0.8}, candidate recommendation object 3 {relationship intimacy is 0.2, conversion rate is 0.1, preference compliance is 0.6}. Each indicator can be summed and then sorted based on the summed result. The summed result of candidate recommendation object 1 is 1.8, the summed result of candidate recommendation object 2 is 1.9, and the summed result of candidate recommendation object 3 is 0.9. After sorting, the display result of the candidate recommendation objects displayed to the target object is {candidate recommendation object 2, candidate recommendation object 1, candidate recommendation object 3}.

[0170] The benefit of the embodiment of the above steps 310-350 is that, since multiple candidate recommendation objects are sorted using relationship intimacy, conversion rate and preference conformity, the sorting accuracy is greatly improved compared to sorting using only one or two indicators. In addition, relationship intimacy represents the distance between objects in multiple scenes based on the object relationship data in multiple scenes, but multiple scenes are not necessarily scenes that are equally important to the target object. Although the objects may be closely related based on the object relationship data in multiple scenes, they may not meet the preferences of the target object. And according to the various performances of the target object before, it may be predicted that the target object does not have a very high conversion rate for an object that seems to be relatively close to it. Therefore, this embodiment sorts multiple candidate recommendation objects based on multiple indicators such as relationship intimacy, conversion rate, and preference conformity, improves the sorting accuracy in multiple scenes of the application, improves the conversion rate of the recommendation process, and improves the universality of the recommendation process.

[0171] The above is a general description of steps 310-350. The specific implementation process of steps 310-350 will be described in detail below.

[0172] Detailed description of step 310

[0173] In step 310, multiple candidate recommended objects of the target object are obtained based on object relationship data and object interaction data in multiple scenes in the application.

[0174] Reference Figure 4 In one embodiment, step 310 includes:

[0175] Step 410: Generate a multi-scenario element interconnection graph based on object relationship data and object interaction data in multiple scenes in the application;

[0176] Step 420: Based on the interconnected graph of multiple scene elements and the target object, obtain multiple candidate recommended objects for the target object.

[0177] The following is an overview of steps 410-420.

[0178] In step 410, a multi-scenario element interconnection graph is generated based on the object relationship data and object interaction data in multiple scenes in the application. The multi-scenario element interconnection graph includes multiple element nodes, and the multiple element nodes include object element nodes and auxiliary element nodes. The object element nodes and the relationship between each object element node are generated by the object relationship data, and the auxiliary element nodes and the relationship between the auxiliary element nodes and the object element nodes are generated by the object interaction data.

[0179] Reference Figure 5 , Figure 5A multi-scenario element interconnection graph is shown. On the one hand, the multi-scenario element interconnection graph includes 8 object element nodes and the relationship between the 8 object element nodes. For example, in the multi-scenario element interconnection graph, object element node 1 is connected to object element node 2, object element node 2 is connected to object element node 3, and so on. On the other hand, the multi-scenario element interconnection graph includes 24 auxiliary element nodes and the relationship between the 24 auxiliary element nodes and 8 object element nodes. These 24 auxiliary element nodes include 3 content author nodes, 3 content author attribute nodes, 5 recommended activity nodes, 7 content nodes, 3 content attribute nodes, and 3 object interest tag nodes. The relationship between the 24 auxiliary element nodes and the 8 object element nodes includes that the content author node 1 is connected to the object element node 8, the content author attribute node 1 is connected to the object element node 8 through the content author node 1, the recommended activity node 1 is connected to the object element node 1, the content node 1 is connected to the object element node 1, the content attribute node 1 is connected to the object element node 1 through the content node 1, the object interest tag node 1 is connected to the object element node 1, and so on.

[0180] Notice, Figure 5 The multi-scenario element interconnection graph shown also includes an object profile node. The object profile node is generated by object attribute information and is used to indicate the attribute information of the object. For example, the attribute information corresponding to the object profile node includes interests, work, educational background, geographic location, etc.

[0181] After generating the multi-scenario element interconnection graph, in step 420, based on the multi-scenario element interconnection graph and the target object, multiple candidate recommended objects of the target object are obtained.

[0182] Continue to refer to Figure 5 , assuming that the object corresponding to object element node 4 is the target object, based on the multi-scenario element interconnection graph, the candidate recommended objects obtained may be {the object corresponding to object element node 1, the object corresponding to object element node 2, the object corresponding to object element node 7}. It should be noted that the objects corresponding to the three object element nodes, object element node 3, object element node 5 and object element node 6, are already associated with the target object and will not be recommended as candidate objects for the target object.

[0183] The advantage of the embodiment of the above steps 410-420 is that a multi-scenario element interconnection graph is generated based on object relationship data and object interaction data, so that the network structure of object relationship data and the network structure of object interaction data are aggregated in the same multi-scenario element interconnection graph, thereby realizing the aggregation of complex and diverse data. The candidate recommended objects obtained based on the multi-scenario element interconnection graph can further improve the recommendation accuracy and recommendation conversion rate.

[0184] Steps 410 - 420 are described in detail below.

[0185] In step 410, a multi-scenario element interconnection graph is generated based on the object relationship data and object interaction data in multiple scenes in the application.

[0186] Reference Figure 6 In one embodiment, step 410 includes:

[0187] Step 610: establishing a first subspace corresponding to the object element node and a plurality of second subspaces corresponding to the plurality of types of auxiliary element nodes;

[0188] Step 620: Generate object element node data based on the object relationship data in the multiple scenes and place it in the first subspace, and generate multiple types of auxiliary element node data based on the object interaction data in the multiple scenes and place them in multiple second subspaces;

[0189] Step 630: Generate a multi-scene element interconnection graph based on the object element node data in the first subspace and the object interaction data in multiple second subspaces.

[0190] Steps 610-630 are described in detail below.

[0191] In step 610, the first subspace refers to a container for storing object element node data, and the second subspace refers to a container for storing auxiliary element node data. It can be understood that since the embodiment of the present disclosure is aimed at object relationship data and object interaction data in multiple scenarios in an application, data isolation can be achieved by establishing the first subspace and the second subspace, which can improve data processing efficiency and also help improve the efficiency and accuracy of generating a multi-scenario element interconnection graph.

[0192] In step 620, the explanation of the object relationship data and the object interaction data has been introduced in the above step 310 and will not be repeated here.

[0193] The object element node data is generated from the object relationship data and refers to data used to represent the object element nodes and the relationship between the object element nodes. Fig. 7AIn one example, the object relationship data includes {object 1 and object 2 are mutually related objects, object 2 and object 3 are mutually related objects}, and the object element node data generated based on the object relationship data includes {object 1 corresponds to object element node 1, object 2 corresponds to object element node 2, object 3 corresponds to object element node 3, object element node 1 is connected to object element node 2, and object element node 2 is connected to object element node 3}. After the object element node data is generated, the obtained object element node data is placed in the first subspace.

[0194] For auxiliary element node data, the auxiliary element node data is generated by object interaction data, and refers to data used to represent auxiliary element nodes and the relationship between auxiliary element nodes. In one embodiment, the object interaction data includes first interaction data between the object and the content, second interaction data between the object and the content author, and third interaction data between the object and the recommended activity item. The auxiliary element node data includes: object interest tag node data, content node data, content author node data, and recommended activity node data.

[0195] In one embodiment, in step 620, multiple types of auxiliary element node data are generated based on the object interaction data in multiple scenes and placed in multiple second subspaces, including:

[0196] Based on the first interaction data, content node data is generated and placed in a second subspace corresponding to the content node data;

[0197] Based on the second interaction data, content author node data is generated and placed in a second subspace corresponding to the content author node data;

[0198] Based on the third interaction data, generating recommended activity node data, and putting the data into a second subspace corresponding to the recommended activity node data;

[0199] Based on the first interaction data between the object and each content, object interest label node data is generated and placed in the second subspace corresponding to the object interest label node data.

[0200] Specifically, in this embodiment, referring to Figure 7B , the first interaction data includes {object 1 liked content 1, object 2 commented on content 2}, then the object interest label node data generated based on the first interaction data includes {content 1 corresponds to object interest label node 1, content 2 corresponds to object interest label node 2, object interest label node 1 is connected to object element node 1, and object interest label node 2 is connected to object element node 2}. The obtained object interest label node data is put into the second subspace corresponding to the object interest label node data.

[0201] Continue to refer to Figure 7BThe content node data generated based on the first interaction data includes {content 1 corresponds to content node 1, content 2 corresponds to content node 2, content 1 is connected to object 1, and content 2 is connected to object 2}. The obtained content node data is placed in the second subspace corresponding to the content node data.

[0202] Continue to refer to Figure 7B , the second interaction data includes {object 1 follows content author 1, object 3 leaves a message to content author 2}, then the content author node data generated based on the second interaction data includes {content author 1 corresponds to content author node 1, content author 2 corresponds to content author node 2, content author node 1 is connected to object element node 1, content author node 2 is connected to object element node 3}. The obtained content author node data is put into the second subspace corresponding to the content author node data.

[0203] Continue to refer to Figure 7B , the third interaction data includes {object 1 forwarded recommended activity item 1, object 1 collected recommended activity item 2, object 2 forwarded recommended activity item 2}, and the recommended activity node data generated based on the third interaction data includes {recommended activity item 1 corresponds to recommended activity node 1, recommended activity item 2 corresponds to recommended activity node 2, recommended activity node 1 is connected to object element node 1, recommended activity node 1 is connected to object element node 2, and recommended activity node 2 is connected to object element node 2}. The obtained recommended activity node data is put into the second subspace corresponding to the recommended activity node data.

[0204] In another embodiment, referring to Figure 8 In step 620, based on the object interaction data in the multiple scenes, multiple types of auxiliary element node data are generated and placed in multiple second subspaces, including:

[0205] Step 810: Obtain content-related data in multiple scenes;

[0206] Step 820: Based on the object interaction data and content association data in the multiple scenes, multiple types of auxiliary element node data are generated and placed in multiple second subspaces.

[0207] Steps 810-820 are described in detail below.

[0208] In step 810, content-related data refers to data representing content-related information. Content-related data includes: content data, content attribute data, and content author attribute data. Auxiliary element node data also includes: content attribute node data, content author attribute node data, and content node relationship data.

[0209] In one embodiment, step 820 includes:

[0210] Based on the content attribute data, content attribute node data is generated and placed in a second subspace corresponding to the content attribute node data;

[0211] Based on the content author attribute data, generate content author attribute node data, and put it into the second subspace corresponding to the content author attribute node data;

[0212] Based on the content data, content node relationship data is generated and put into a second subspace corresponding to the content node relationship data.

[0213] In this embodiment, refer to Fig. 9 , the content attribute data includes {the attribute of content 1 is a short video, the attribute of content 2 is a picture}, and the content attribute node data generated based on the content attribute data includes {the attribute of content 1 corresponds to content attribute node 1, the attribute of content 2 corresponds to content attribute node 2, content attribute node 1 is connected to content node 1, and content attribute node 2 is connected to content node 2}. The obtained content attribute node data is placed in the second subspace corresponding to the content attribute node data.

[0214] Continue to refer to Fig. 9 , the content author attribute data includes {the attribute of content author 1 is that the geographical location is G city, and the attribute of content author 2 is that the geographical location is L city}, and the content author attribute node data generated based on the content author attribute data includes {the attribute of content author 1 corresponds to content author attribute node 1, the attribute of content author 2 corresponds to content author attribute node 2, content author attribute node 1 is connected to content author node 1, and content author attribute node 2 is connected to content author node 2}. The obtained content author attribute node data is put into the second subspace corresponding to the content author attribute node data.

[0215] Continue to refer to Fig. 9 , the content data includes {content 1 is a short video about food tutorials, content 2 is a picture about tourist attractions}, and the content node relationship data generated based on the content data includes {content node 1 is connected to content node 2}. The obtained content node relationship data is placed in the second subspace corresponding to the content node relationship data.

[0216] After step 620, in step 630, a multi-scene element interconnection graph is generated based on the object element node data in the first subspace and the object interaction data in multiple second subspaces.

[0217] In one embodiment, the auxiliary element nodes include: object interest tag nodes, content nodes, content author nodes, and recommended activity nodes, and step 630 includes:

[0218] Based on the object element node data in the first subspace, generate object element nodes and the relationship between the object element nodes in the multi-scene element interconnection graph;

[0219] Based on the content node data in the second subspace, generating content nodes and the relationship between object element nodes and content nodes in the multi-scenario element interconnection graph;

[0220] Based on the content author node data in the second subspace, generate the content author node and the relationship between the object element node and the content author node in the multi-scenario element interconnection graph;

[0221] Based on the recommended activity node data in the second subspace, generating the recommended activity nodes and the relationship between the object element nodes and the recommended activity nodes in the multi-scenario element interconnection graph;

[0222] Based on the object interest label node data in the second subspace, object interest label nodes and the relationship between object element nodes and object interest label nodes are generated in the multi-scene element interconnection graph.

[0223] Specifically, in this embodiment, referring to Fig.10 , based on the object element node data in the first subspace, object element nodes 1-3 in the multi-scenario element interconnection graph are generated, and object element node 1 is connected to object element node 2, and object element node 2 is connected to object element node 3. Based on the object interest tag node data in the second subspace corresponding to the object interest tag node data, object interest tag nodes 1-2 in the multi-scenario element interconnection graph are generated, and object element node 1 is connected to object interest tag node 2, and object element node 2 is connected to object interest tag node 1. Based on the content node data in the second subspace corresponding to the content node data, content nodes 1-2 in the multi-scenario element interconnection graph are generated, and object element node 1 is connected to content node 1, and object element node 2 is connected to content node 2. Based on the content author node data in the second subspace corresponding to the content author node data, content author nodes 1-2 in the multi-scenario element interconnection graph are generated, and object element node 1 is connected to content author node 1, and object element node 3 is connected to content author node 2. Based on the recommended activity node data in the second subspace corresponding to the recommended activity node data, the recommended activity node 1-2 in the multi-scenario element interconnection graph is generated, and the object element node 1 is connected to the recommended activity node 1, the object element node 2 is connected to the recommended activity node 1, and the object element node 2 is connected to the recommended activity node 2. After the above process, the following is finally generated: Fig.10 The interconnected map of multiple scene elements is shown.

[0224] In another embodiment, the auxiliary element node further includes: a content attribute node and a content author attribute node, and step 630 includes:

[0225] Based on the content attribute node data in the second subspace, generate content attribute nodes and the relationship between content nodes and content attribute nodes in the multi-scenario element interconnection graph;

[0226] Based on the content author attribute node data in the second subspace, generate content author attribute nodes and the relationship between content author nodes and content author attribute nodes in the multi-scenario element interconnection graph;

[0227] Based on the relationship data between content nodes in the second subspace, the relationship between content nodes is generated in the multi-scenario element interconnection graph.

[0228] Specifically, in this embodiment, referring to Fig.11 Based on the content attribute node data in the second subspace corresponding to the content attribute node data, content attribute nodes 1-2 in the multi-scenario element interconnection graph are generated, and content node 1 is connected to content attribute node 1, and content node 2 is connected to content attribute node 2. Based on the content author attribute node data in the second subspace corresponding to the content author attribute node data, content author attribute nodes 1-2 in the multi-scenario element interconnection graph are generated, and content author node 1 is connected to content author attribute node 1, and content author node 2 is connected to content author attribute node 2. Based on the content node relationship data in the second subspace corresponding to the content node data, content node 1 is connected to content node 2 in the multi-scenario element interconnection graph. After the above process, Fig.10 Based on this, the following Fig.11 The interconnected map of multiple scene elements is shown.

[0229] The advantage of the embodiment of the above steps 610-630 is that data isolation can be achieved by establishing a first subspace and multiple second subspaces. In addition to improving data processing efficiency, it can also improve the efficiency and accuracy of generating an interconnected graph of multi-scenario elements.

[0230] The above is a detailed description of step 410 , and step 420 is described in detail below.

[0231] In step 420, based on the interconnected graph of multiple scene elements and the target object, multiple candidate recommended objects of the target object are obtained.

[0232] Reference Fig.12 In one embodiment, the auxiliary element nodes include a content author node and an object interest tag node, and step 420 includes:

[0233] Step 1210: acquiring a first candidate recommendation object based on the object element node, the content author node, and the target object;

[0234] Step 1220: obtaining a second candidate recommended object based on the object interest tag node and the target object;

[0235] Step 1230: Input the multi-scenario element interconnection graph and the target object into the relationship and interest fusion recommendation model to obtain a third candidate recommendation object;

[0236] Step 1240: Integrate the first candidate recommendation object, the second candidate recommendation object, and the third candidate recommendation object into a plurality of candidate recommendation objects.

[0237] Steps 1210 - 1240 are described in detail below.

[0238] In step 1210, the first candidate recommended object is obtained based on the object element node, the content author node, and the target object. The object element node corresponds to a certain object, and the content author node also corresponds to a certain object. The connection relationship between the object element nodes can reflect the probability of two objects that are not related objects becoming mutually related objects. The connection relationship between the content author node and the object element node can reflect whether the two objects that are not related objects follow the same content author. If they follow the same content author, it means that the probability of the two objects becoming mutually related objects is relatively high. Therefore, the first candidate recommended object recommended for the target object can be obtained based on the object element node and the content author node.

[0239] For example, refer to Fig.13 , Fig.13 A multi-scenario element interconnection graph is shown. Assume that object 2 corresponding to object element node 2 is the target object. Based on object element nodes 1-5 in the multi-scenario element interconnection graph, it can be known that object element node 1 is indirectly connected to object element node 2 through object element node 3. It can be seen that object 1 corresponding to object element node 1 is not yet an associated object of the target object, but object 1 and the target object are both associated objects of object 3, so object 1 corresponding to object element node 1 can be used as the first candidate recommended object. Based on content author nodes 1-2, it can be known that object element node 2 and other object element nodes have no commonly connected content author nodes, that is, the target object and other objects have no common content authors of concern. Based on the above reasons, the final first candidate recommended object includes {object 1}.

[0240] In step 1220, based on the object interest tag node and the target object, a second candidate recommended object is obtained. The object interest tag node may indicate the interest tag of the object. If the object interest tag node connected to a certain object element node is consistent with the object interest tag node connected to the target object, it indicates that the interest tag of the object is consistent with the interest tag of the target object, thereby indicating that the probability of the object and the target object becoming mutually related objects is relatively high.

[0241] For example, continue to refer to Fig.13 , continue to assume that object 2 corresponding to object element node 2 is the target object. Object interest tag node 2 is not only connected to object element node 2, but also connected to object element node 4. It can be seen that object 2 and object 4 corresponding to object element node 4 have a common interest tag. In this way, based on object interest tag node 2, object 4 can be used as the second candidate recommended object for the target object. Based on the above reasons, the second candidate recommended object finally includes {object 4}.

[0242] In step 1230, the multi-scenario element interconnection graph and the target object are input into the relationship and interest fusion recommendation model to obtain a third candidate recommendation object. The relationship and interest fusion recommendation model refers to a model that converts the multi-scenario element interconnection graph and the target object into a third candidate recommendation object, such as a convolutional neural network (CNN), a decision tree model, etc.

[0243] Different from step 1210, which determines the first candidate recommendation object based on the object element node and the content author node, and different from step 1220, which determines the second candidate recommendation object based on the object interest tag node, step 1230 determines the third candidate recommendation object based on the entire multi-scenario element interconnection graph. It can be understood that the multi-scenario element interconnection graph not only includes the relationship between the object element nodes, but also includes the relationship between the object element nodes and the object interest tag nodes. Therefore, the third candidate recommendation object determined based on the entire multi-scenario element interconnection graph can fully reflect the fusion of the third candidate recommendation object and the target object in terms of relationship and interest.

[0244] For example, refer to Fig.13 , continue to assume that object 2 corresponding to object element node 2 is the target object. Object element node 5 is not connected to object element node 2, and there is no commonly connected object interest label node. However, object element node 5 and object element node 1 have a commonly connected object interest label node 1, and object element node 1 is indirectly connected to object element node 2. In this case, the relationship and interest fusion recommendation model can mine the potential connection between object element node 5 and object element node 2 based on the input multi-scenario interconnected graph and the target object, thereby taking object 5 as the third candidate recommended object. Based on the above reasons, the final third candidate recommended object includes {object 5}.

[0245] After the above steps 1210-1230, in step 1240, the first candidate recommendation object, the second candidate recommendation object, and the third candidate recommendation object are integrated into a plurality of candidate recommendation objects. For example, the plurality of candidate recommendation objects include {object 1, object 4, object 5}.

[0246] The benefit of the implementation of the above steps 1210-1240 is that recommendations are made based on relationships, based on interests, and based on the fusion of relationships and interests. Multiple candidate recommendation objects are determined by using these three recommendation channels, which increases the conversion rate of target objects to candidate recommendation objects and improves the recommendation processing conversion rate.

[0247] The above is a detailed description of steps 1210-1240. The following is a detailed description of the specific implementation of steps 1210 and 1220.

[0248] In one embodiment, referring to Fig.14 , step 1210 includes:

[0249] Step 1410: in the multi-scenario element interconnection graph, obtain an object element node that has a common associated object element node with the object element node of the target object, and use the object corresponding to the obtained object element node as the first sub-candidate recommendation object;

[0250] Step 1420: Obtain an object element node having the same specific attribute as the target object from each object element node in the multi-scenario element interconnection graph, and use the object corresponding to the obtained object element node as the second sub-candidate recommendation object;

[0251] Step 1430: group the object element nodes based on the attributes of the multi-scenario element interconnection graph, obtain the object element nodes that are grouped in the same group as the object element node of the target object, and use the object corresponding to the obtained object element node as the third sub-candidate recommendation object;

[0252] Step 1440: From the objects corresponding to the object element nodes in the multi-scenario element interconnection graph, obtain an object that belongs to the same application session group as the target object as a fourth sub-candidate recommendation object;

[0253] Step 1450: in the multi-scenario element interconnection graph, obtain a content author node that has an interactive relationship with the object element node of the target object, and use the object corresponding to the obtained content author node as the fifth child candidate recommendation object;

[0254] Step 1460: Based on the list of associated objects of the target object outside the application, obtain a sixth sub-candidate recommended object;

[0255] Step 1470: in the multi-scenario element interconnection graph, obtain a content author node that is mutually concerned with the object element node, and use the object corresponding to the obtained content author node as the seventh sub-candidate recommendation object;

[0256] Step 1480, integrate the first sub-candidate recommendation object, the second sub-candidate recommendation object, the third sub-candidate recommendation object, the fourth sub-candidate recommendation object, the fifth sub-candidate recommendation object, the sixth sub-candidate recommendation object, and the seventh sub-candidate recommendation object into the first candidate recommendation object.

[0257] Steps 1410-1480 are described in detail below.

[0258] In step 1410, the commonly associated object element node means that two object element nodes are associated with the same object element node. Fig.13 , object element node 2 and object element node 1 are both associated with object element node 3, then the commonly associated object element node is object element node 3. Assuming that object 2 corresponding to object element node 2 is the target object, then Fig.13 In the multi-scenario element interconnection graph shown, object element node 1 is obtained, and object 1 corresponding to the obtained object element node 1 is used as the first sub-candidate recommended object.

[0259] In step 1420, the specific attribute refers to an attribute selected in advance from multiple attributes of the target object. For example, if the multiple attributes of the target object include educational background, work, geographical location, etc., the specific attribute may be at least one of educational background, work, and geographical location. If an object element node is identical to the target object in a specific attribute, it means that the object corresponding to the object element node should be recommended as the second sub-candidate object. For example, assume that the geographical location of object element node 2 is City G, assume that the geographical location of object element node 1 is City G, and assume that object 2 corresponding to object element node 2 is the target object. If the specific attribute is the geographical location, the object element node whose specific attribute is identical to the target object is obtained as object element node 1, and object 1 corresponding to the obtained object element node 1 is used as the second sub-candidate object.

[0260] In step 1430, based on the attributes of each object element node of the multi-scenario element interconnection graph, each object element node is grouped, and the object element node that is grouped in the same group as the object element node of the target object is obtained, and the object corresponding to the obtained object element node is used as the third sub-candidate recommendation object. For example, it is assumed that the attributes of object element node 1 include {school attended is University A, work is a student}, it is assumed that the attributes of object element node 2 include {school attended is University A, work is a student}, and it is assumed that object 2 corresponding to object element node 2 is the target object. After grouping object element node 1 and object element node 2, object element node 1 and object element node 2 may be grouped in the same alumni group. In this way, object element node 1 is obtained, and object 1 corresponding to the obtained object element node 1 is used as the third sub-candidate recommendation object.

[0261] In step 1440, among the objects corresponding to the object element nodes of the multi-scenario element interconnection graph, an object belonging to the same application session group as the target object is obtained as the fourth sub-candidate recommendation object. For example, it is assumed that the object of object element node 1 belongs to the instant messaging application C session group, it is assumed that the object of object element node 2 belongs to the instant messaging application C session group, and it is assumed that object 2 corresponding to object element node 2 is the target object. In this way, the object obtained that belongs to the instant messaging application C session group with the target object is object 1, and object 1 is taken as the fourth sub-candidate recommendation object.

[0262] In step 1450, in the multi-scenario element interconnection graph, a content author node that has an interactive relationship with the object element node of the target object is obtained, and the object corresponding to the obtained content author node is used as the fifth child candidate recommendation object. For example, assuming that object 2 corresponding to object element node 2 is the target object, and the target object has an interactive relationship with content author 1, and the object corresponding to content author 1 is object 1, object 1 is used as the fifth child recommendation object.

[0263] In step 1460, based on the list of associated objects of the target object outside the application, the sixth sub-candidate recommended object is obtained. The list of associated objects outside the application may refer to a list of associated objects of the target object in the real world (address book / alumni / relatives / colleagues) outside the application. For example, it is assumed that object 2 corresponding to object element node 2 is the target object, and it is assumed that the list of associated objects is the address book of object 2. If object 1 is included in the address book, object 1 is taken as the sixth sub-recommended object.

[0264] In step 1470, in the multi-scenario element interconnection graph, the content author node that is mutually concerned with the object element node of the target object is obtained, and the object corresponding to the obtained content author node is used as the seventh child candidate recommendation object. For example, it is assumed that object 2 corresponding to object element node 2 is the target object, and it is assumed that the content author node that is mutually concerned with element node 2 is content author node 1. The object corresponding to content author node 1 is obtained as object 1, and object 1 is used as the seventh child recommendation object.

[0265] After the above steps 1410-1470, in step 1480, the first sub-candidate recommendation object, the second sub-candidate recommendation object, the third sub-candidate recommendation object, the fourth sub-candidate recommendation object, the fifth sub-candidate recommendation object, the sixth sub-candidate recommendation object, and the seventh sub-candidate recommendation object are integrated into the first candidate recommendation object. In summary, the first candidate recommendation object includes {object 1}.

[0266] The advantage of the implementation of the above steps 1410-1480 is that it utilizes the object element nodes and content author nodes in the multi-scenario element interconnection graph to fully mine the first candidate recommendation objects that can be the target objects with seven selection dimensions, thereby improving the conversion rate of the recommendation processing.

[0267] Reference Fig.15 In one embodiment, the auxiliary element node includes a content node, and step 1220 includes:

[0268] Step 1510: acquiring an eighth sub-candidate recommended object based on the object interest label node of the target object and the label similarity between the object interest label nodes of each other object;

[0269] Step 1520: based on the interaction data of the object element node of the target object with each content node in the multi-scenario element interconnection graph, obtain a first interest point vector of the object element node of the target object; based on the interaction data of the object element nodes of each other object in the multi-scenario element interconnection graph with each content node, obtain a second interest point vector of the object element nodes of each other object; and based on the comparison between the first interest point vector and the second interest point vector, obtain a ninth sub-candidate recommendation object;

[0270] Step 1530: Integrate the eighth sub-candidate recommendation object and the ninth sub-candidate recommendation object into a second candidate recommendation object.

[0271] Steps 1510-1530 are described in detail below.

[0272] In step 1510, if the target object has a higher label similarity with a certain object, it means that the object should be selected as the eighth sub-candidate recommended object. For example, an object whose label similarity is greater than a predetermined similarity threshold may be selected as the eighth sub-recommended object. Alternatively, multiple objects may be sorted based on label similarity, and the object with the highest ranking may be selected as the eighth sub-recommended object.

[0273] In step 1520, the interest points of the target object can be counted based on the interaction data of the object element nodes of the target object with each content node in the multi-scene element interconnection graph, so that the first interest point vector can be obtained. The interest points of each other object can be counted based on the interaction data of the object element nodes of each other object in the multi-scene element interconnection graph with each content node, so that the second interest point vector of each other object can be obtained. The smaller the distance between the first interest point vector of the target object and the second interest point vector of a certain object is, the more the object should be used as the ninth sub-candidate recommendation object. For example, an object whose distance between the first interest point vector and the second interest point vector is less than a predetermined distance threshold can be used as the ninth sub-recommended object. It is also possible to sort multiple objects based on the distance between the first interest point vector and the second interest point vector, and use the top-ranked object as the ninth sub-recommended object.

[0274] After step 1510 and step 1520, in step 1530, the eighth sub-candidate recommendation object and the ninth sub-candidate recommendation object are integrated into a second candidate recommendation object.

[0275] The advantage of the embodiment of the above steps 1510-1530 is that the object label similarity is used to reflect the explicit matching between the target object and other objects in the interest dimension, and the first interest point vector and the second interest point vector are used to reflect the implicit matching between the target object and other objects in the interest dimension, thereby improving the sorting accuracy and the conversion rate of the recommendation processing.

[0276] Detailed description of step 320

[0277] In step 320, the degree of intimacy between each candidate recommendation object and the target object is obtained.

[0278] Reference Fig.16 In one embodiment, step 320 includes:

[0279] Step 1610: Obtain the attribute similarity between each candidate recommendation object and the target object;

[0280] Step 1620: Obtain the interaction frequency between each candidate recommendation object and the target object in the application;

[0281] Step 1630: Obtain the number of commonly associated objects between each candidate recommendation object and the target object in the application;

[0282] Step 1640: Obtain the main interaction type between each candidate recommendation object and the target object in the application;

[0283] Step 1650: Determine the closeness of the relationship between each candidate recommendation object and the target object based on attribute similarity, interaction frequency, number of commonly associated objects, and main interaction type.

[0284] Steps 1610-1650 are described in detail below.

[0285] In step 1610, the attribute similarity between each candidate recommendation object and the target object is obtained. Attribute similarity refers to the degree of similarity between the attributes of the candidate recommendation object and the attributes of the target object. If the attribute similarity of a candidate recommendation object is higher, the ranking of the candidate recommendation object will be higher.

[0286] In one embodiment, step 1610 includes:

[0287] Obtaining first attribute values ​​of multiple first attributes of each candidate recommendation object and second attribute values ​​of multiple first attributes of the target object;

[0288] Generate a first vector based on first attribute values ​​of a plurality of first attributes of the candidate recommendation object, and generate a second vector based on second attribute values ​​of a plurality of first attributes of the target object;

[0289] Based on the distance between the first vector and the second vector, the attribute similarity is obtained.

[0290] Reference Fig.17 In this embodiment, it is assumed that the candidate recommendation objects include {candidate recommendation object 1, candidate recommendation object 2}. The first attribute of candidate recommendation object 1 includes {interest, work, educational background, geographical location}, and the first attribute value of the first attribute of candidate recommendation object 1 includes {playing badminton, student, university bachelor, G city}, and the generated first vector of candidate recommendation object 1 is 0110101011110101. The first attribute of candidate recommendation object 2 includes {interest, work, educational background, geographical location}, and the first attribute value of the first attribute of candidate recommendation object 1 includes {swimming, teacher, university doctor, L city}, and the generated first vector of candidate recommendation object 2 is 0100100010010110. The first attribute of the target object includes {interest, work, geographical location}, and the first attribute value of the target object includes {playing badminton, student, L city}, and the second vector of the target object is 0110101000000110.

[0291] Continue to refer to Fig.17, in the rectangular coordinate system, the first vector of candidate recommendation object 1, the first vector of candidate recommendation object 2, and the second vector of the target object are mapped into a point, and the distance between each first vector and the second vector can be clearly seen. Specifically, the distance between candidate recommendation object 1 and the target object is small, and the distance between candidate recommendation object 2 and the target object is large. Finally, based on the distance between the first vector and the second vector, the attribute similarity is obtained, including {the attribute similarity between candidate recommendation object 1 and the target object is 0.5, and the attribute similarity between candidate recommendation object 2 and the target object is 0.25}.

[0292] The advantage of this embodiment is that the attribute similarity is measured by the distance between the first vector and the second vector, which can flexibly target the difference in the number of attributes between different objects and reduce the calculation complexity of the attribute similarity.

[0293] In step 1620, the interaction frequency between each candidate recommendation object and the target object in the application is obtained. If the interaction frequency between a candidate recommendation object and the target object in the application is higher, the ranking of the candidate recommendation object is higher.

[0294] In one embodiment, step 1620 includes:

[0295] Identify interactions between the candidate recommendation object and the target object in the application within a predetermined period before the current time;

[0296] Counting the number of interactions within a predetermined period;

[0297] Based on the number of counted interactions and the predetermined period length, an interaction frequency is determined.

[0298] In this embodiment, refer to Fig.18 , assuming that the interactions between the candidate recommendation object and the target object in the application are identified as {13:30, the interaction between the candidate recommendation object 1 and the target object is a like; 13:35, the interaction between the candidate recommendation object 1 and the target object is a comment; 13:40, the interaction between the candidate recommendation object 1 and the target object is a message}. The scheduled period is from 13:30 to 13:40, then the number of interactions within the scheduled period is 3. Then the interaction frequency determined based on the number of interactions and the length of the scheduled period is 3 / 10, in times / minute.

[0299] The advantage of this embodiment is that the interaction frequency can be quickly determined based on the number of interactions and the predetermined period, thereby improving the efficiency of determining the interaction frequency.

[0300] In step 1630, the number of objects commonly associated with each candidate recommendation object and the target object in the application is obtained. If a candidate recommendation object has more objects commonly associated with the target object in the application, the ranking of the candidate recommendation object will be higher.

[0301] In one embodiment, step 1630 includes:

[0302] In a multi-scenario element interconnection graph generated based on object relationship data and object interaction data, other object element nodes adjacent to the object element node of the candidate recommendation object are identified as associated objects of the candidate recommendation object in the application, and other object element nodes adjacent to the object element node of the target object are identified as associated objects of the target object in the application;

[0303] The number of objects in the intersection of the associated objects of the candidate recommendation object in the application and the associated objects of the target object in the application is determined as the number of commonly associated objects.

[0304] In this embodiment, refer to Fig.19 , Fig.19 The multi-scenario element interconnection graph includes 5 object element nodes, but other auxiliary element nodes are omitted. Assume that object 1 corresponding to object element node 1 is a candidate recommended object, and assume that object 2 corresponding to object element node 2 is the target object. Based on the multi-scenario element interconnection graph, it can be identified that other object element nodes adjacent to object element node 1 of the candidate recommended object include {object element node 3, object element node 5}. Based on the multi-scenario element interconnection graph, it can also be identified that other object element nodes adjacent to object element node 2 of the target object include {object element node 3, object element node 4, object element node 5}. The intersection is then determined to be {object element node 3, object element node 5}, and the final number of commonly associated objects is 2.

[0305] The advantage of this embodiment is that the accuracy and efficiency of determining the number of commonly associated objects can be improved by using the interconnected graph of multiple scene elements to determine the number of commonly associated objects.

[0306] In step 1640, the main interaction type between each candidate recommendation object and the target object in the application is obtained. The interaction type includes the like type, comment type, message type, conversation type, etc. The main interaction type refers to the interaction type with the most interactions in the interaction between the candidate recommendation object and the target object in the application, thereby obtaining the main interaction type. Different main interaction types have different importance, for example, the message type and conversation type are more important than the like type and comment type.

[0307] In one embodiment, step 1640 includes:

[0308] Identify interactions between the candidate recommendation object and the target object in the application within a predetermined period before the current time;

[0309] Get the interaction type of the identified interaction;

[0310] Determine the number of interactions of each interaction type captured;

[0311] Based on the number of interactions, a primary interaction type is determined among the identified interaction types.

[0312] In this embodiment, refer to Fig. 20 , the interactions between the candidate recommended object and the target object in the application include {13:30, the interaction between the candidate recommended object 1 and the target object is like; 13:35, the interaction between the candidate recommended object 1 and the target object is comment; 13:40, the interaction between the candidate recommended object 1 and the target object is message; 13:45, the interaction between the candidate recommended object 1 and the target object is dialogue; 13:50, the interaction between the candidate recommended object 1 and the target object is message; 13:55, the interaction between the candidate recommended object 1 and the target object is like; 13:56, the interaction between the candidate recommended object 1 and the target object is message}. The identified interaction types include {13:30, like type; 13:35, comment type; 13:40, message type; 13:45, dialogue type; 13:50, message type; 13:55, like type; 13:56, message type}. The number of interactions in the interaction type includes {the number of interactions of the like type is 2, the number of interactions of the comment type is 1, the number of interactions of the message type is 3, and the number of interactions of the conversation type is 1}. The interaction type with the largest number of interactions can be used as the main interaction type, that is, the message type with an interaction number of 3 can be used as the main interaction type.

[0313] The advantage of this embodiment is that the main interaction type is determined from the interaction types based on the number of interactions, thereby improving the efficiency of determining the main interaction type.

[0314] After steps 1610 - 1640 , in step 1650 , the closeness of the relationship between each candidate recommendation object and the target object is determined based on attribute similarity, interaction frequency, number of commonly associated objects, and main interaction type.

[0315] The advantage of the embodiment of the above steps 1610-1650 is that the relationship intimacy is determined based on attribute similarity, interaction frequency, number of commonly associated objects, and main interaction type, which improves the accuracy of determining the relationship intimacy and thus improves the recommendation processing conversion rate.

[0316] In one embodiment, referring to Fig.21 , step 1650 includes:

[0317] Step 2110: Determine a first score of the candidate recommendation object based on the attribute similarity;

[0318] Step 2120: Determine a second score of the candidate recommendation object based on the interaction frequency;

[0319] Step 2130: Determine a third score of the candidate recommendation object based on the number of commonly associated objects;

[0320] Step 2140: Determine a fourth score of the candidate recommendation object based on the main interaction type;

[0321] Step 2150: Determine the degree of closeness between the candidate recommendation object and the target object based on the first score, the second score, the third score, and the fourth score of each candidate recommendation object.

[0322] Steps 2110-2150 are described in detail below.

[0323] In step 2110, based on the attribute similarity, a first score of the candidate recommendation object is determined, which can be done by searching a table comparing the attribute similarity range with the first score, or by using a formula method.

[0324] (1) The attribute similarity range and the first score comparison table lists the corresponding relationship between the attribute similarity range and the first score. First, the attribute similarity range is obtained according to the attribute similarity, and then the attribute similarity range and the first score comparison table are searched accordingly to obtain the first score. The following is an example of an attribute similarity range and the first score comparison table:

[0325] Attribute similarity range First Score 0.9 or above 100 0.8-0.9 90 0.6-0.8 80 0.3-0.6 70 0.2-0.3 60 …… ……

[0326] Table 1

[0327] For example, assuming that the attribute similarity is 0.7, the attribute similarity range is 0.6-0.8, and by looking up Table 1, the corresponding first score is 80.

[0328] The above method of searching the attribute similarity range and the first score comparison table has the advantages of being simple and easy to implement with low processing overhead.

[0329] (2) When the formula method is used, the first score can be set to be proportional to the attribute similarity, for example:

[0330] F1=K1·S1 Formula 1

[0331] Among them, F1 represents the first score, S1 represents the attribute similarity, and K1 is a preset constant that can be set according to actual needs. For example, K1=800 / 7, after substituting S1=0.7, the first score F1=80.

[0332] The advantages of the above method of determining the first score by means of a formula are high accuracy, and the formula can be adjusted as required, which is highly flexible.

[0333] In step 2120, based on the interaction frequency, the second score of the candidate recommendation object is determined, which can be done by looking up a comparison table between the interaction frequency range and the second score, or by using a formula method.

[0334] (1) The interaction frequency range and second score comparison table lists the corresponding relationship between the interaction frequency range and the second score. First, the interaction frequency range is obtained according to the interaction frequency, and then the interaction frequency range and second score comparison table is searched to obtain the second score. The following is an example of an interaction frequency range and second score comparison table:

[0335] Interaction frequency range (unit: times / minute) Second score 10 or more 100 8-9 90 6-8 80 3-6 70 2-3 60 …… ……

[0336] Table 2

[0337] For example, assuming the interaction frequency is 5 times / minute, the interaction frequency range is 3-6, and the corresponding second score is 70 obtained by looking up Table 2. The above method of looking up the interaction frequency range and the second score comparison table has the advantages of being simple and easy to implement with low processing overhead.

[0338] (2) When using the formula method, the second score can be set to be proportional to the interaction frequency, for example:

[0339] F2=K2·S2 Formula 2

[0340] Wherein, F2 represents the second score, S2 represents the interaction frequency, and K2 is a preset constant that can be set according to actual needs. For example, K2=14, after substituting S2=5, the second score F2=70.

[0341] The advantages of the above method of determining the second score by means of a formula are high accuracy, and the formula can be adjusted as required, which is highly flexible.

[0342] In step 2130, based on the number of commonly associated objects, a third score of the candidate recommendation object is determined, which may be done by searching a comparison table between the number of commonly associated objects and the third score, or by using a formula.

[0343] (1) The comparison table of the number range of commonly associated objects and the third score lists the corresponding relationship between the number range of commonly associated objects and the third score. First, the number range of commonly associated objects is obtained according to the number of commonly associated objects, and then the comparison table of the number range of commonly associated objects and the third score is searched to obtain the third score. The following is an example of the comparison table of the number range of commonly associated objects and the third score:

[0344] The range of the number of commonly associated objects The third score 10 or more 100 8-9 90 6-8 80 3-6 70 2-3 60 …… ……

[0345] Table 3

[0346] For example, assuming the number of commonly associated objects is 4, the range of commonly associated objects is 3-6, and the corresponding third score is 70 when searching Table 3. The above method of searching the comparison table of commonly associated object range and third score has the advantages of being simple and easy to implement with low processing overhead.

[0347] (2) When the formula method is used, the third score can be set to be proportional to the number of commonly associated objects, for example:

[0348] F3=K3·S3 Formula 3

[0349] Wherein, F3 represents the third score, S3 represents the number of commonly associated objects, and K3 is a preset constant that can be set according to actual needs. For example, K3=35 / 2, after substituting S3=4, the third score F3=70.

[0350] The advantages of the above method of determining the third score by means of a formula are high accuracy, and the formula can be adjusted as required, which is highly flexible.

[0351] In step 2140, based on the main interaction type, the fourth score of the candidate recommendation object is determined, which can be done by looking up a comparison table of the main interaction type and the fourth score, or by other methods. The comparison table of the main interaction type and the fourth score lists the correspondence between the main interaction type and the fourth score. The fourth score can be obtained by looking up the comparison table of the main interaction type and the fourth score according to the main interaction type. The following is an example of a comparison table of the main interaction type and the fourth score:

[0352] Primary interaction type Fourth score Types of conversations 100 Message Type 90 Comment Type 80 Like Type 70 …… ……

[0353] Table 4

[0354] For example, assuming that the main interaction type is the message type, then by looking up Table 4, the corresponding fourth score is 90. The above method of looking up the comparison table between the main interaction type and the fourth score has the advantages of being simple and easy to implement with low processing overhead.

[0355] In step 2150, based on the first score, the second score, the third score, and the fourth score of each candidate recommendation object, the closeness of the relationship between the candidate recommendation object and the target object is determined.

[0356] In one embodiment, step 2150 includes:

[0357] Determining a total score based on the first score, the second score, the third score, and the fourth score of each candidate recommendation object;

[0358] Based on the total score, the closeness of the relationship between the candidate recommendation object and the target object is determined.

[0359] Specifically, the total score may be obtained by calculating the average or weighted average of the first score, the second score, the third score, and the fourth score.

[0360] When calculating the average of the first score, the second score, the third score, and the fourth score as the total score, for example, the first score is 80, the second score is 70, the third score is 70, and the fourth score is 90, then the total score is (80+70+70+90) / 4=77.5. The advantage of using the average to calculate the total score is that it can equally reflect the influence of attribute similarity, interaction frequency, number of commonly associated objects, and main interaction type on determining relationship intimacy.

[0361] When calculating the weighted average of the first score, the second score, the third score, and the fourth score as the total score, for example, the weights of the first score, the second score, the third score, and the fourth score are set to 0.3, 0.2, 0.4, and 0.1 respectively, the first score is 80, the second score is 70, the third score is 70, and the fourth score is 90, then the total score is 80×0.3+70×0.2+70×0.4+90×0.1=75. The advantage of using the weighted average to calculate the total score is that different weights can be set for attribute similarity, interaction frequency, number of commonly associated objects, and main interaction type, which increases the flexibility of determining relationship intimacy.

[0362] Based on the total score, the relationship intimacy between the candidate recommendation object and the target object is determined, which can be done by looking up a comparison table of the total score range and the relationship intimacy, or by using a formula method, etc. The specific process of determining the relationship intimacy is similar to the specific implementation process of determining the first score in step 2110, determining the second score in step 2120, and determining the third score in step 2130, and will not be repeated here.

[0363] The benefit of the embodiment of steps 2110-2150 is that the flexibility and accuracy of determining relationship intimacy can be improved by respectively calculating the first score of attribute similarity, the second score of interaction frequency, the third score of commonly associated objects, and the fourth score of main interaction type.

[0364] Detailed description of step 330

[0365] In step 330, the conversion rate after each candidate recommended object is recommended to the target object is predicted. The conversion rate is the probability that the target object pays attention to the candidate recommended object after the candidate recommended object is recommended to the target object.

[0366] Reference Fig. 22 In one embodiment, step 330 includes:

[0367] Step 2210: Obtain the first object interest tag of the target object;

[0368] Step 2220: Obtain the second object interest tag of each candidate recommendation object;

[0369] Step 2230: Input the first object interest tag and the second object interest tag into the conversion rate prediction model to obtain the predicted conversion rate.

[0370] Steps 2210-2230 are described in detail below.

[0371] In step 2210, the first object interest tag of the target object is obtained. The first object interest tag is used to indicate the interest point of the target object. For example, if the target object likes to play badminton, the first object interest tag is {play badminton}. The first object interest tag can be obtained from the object profile of the target object, which includes the interest point of the target object. The first object interest tag can also be obtained from the object interest tag node adjacent to the object element node of the target object in the multi-scene element interconnection graph.

[0372] In step 2220, the second object interest tag of each candidate recommended object is obtained. The first object interest tag is used to indicate the interest point of the target object. For example, if the target object likes swimming, the second object interest tag is {swimming}. The second object interest tag can be obtained from the object profile of the candidate recommended object, which includes the interest point of the candidate recommended object. The second object interest tag can also be obtained from the object interest tag node adjacent to the object element node of the candidate recommended object in the multi-scene element interconnection graph.

[0373] In step 2230, the first object interest tag and the second object interest tag are input into the conversion rate prediction model to obtain the predicted conversion rate. The conversion rate prediction model is a model that converts the first object interest tag and the second object interest tag into a conversion rate, such as a convolutional neural network (CNN), a decision tree model, etc. For example, the first object interest tag is {playing badminton}, and the second object interest tag is {swimming}, then the conversion rate prediction model obtains a predicted conversion rate of 0.25 based on {playing badminton} and {swimming}. For another example, the first object interest tag is {playing badminton, swimming}, and the second object interest tag is {swimming}, then the conversion rate prediction model obtains a predicted conversion rate of 0.5 based on {playing badminton, swimming} and {swimming}.

[0374] In one embodiment, the conversion rate prediction model is trained in the following manner:

[0375] Construct a sample set of recommended events;

[0376] Input each recommended event sample into the conversion rate prediction model to obtain the predicted probability of adding the sample recommended object as an associated object for the sample target object;

[0377] Based on the conversion label and predicted probability of each recommended event sample, a loss function is constructed, and the conversion rate prediction model is trained based on the loss function.

[0378] In this embodiment, the recommended event sample set includes a plurality of recommended event samples. The more the number of recommended event samples, the better the training effect. The recommended event samples include sample target objects, sample recommended objects recommended for the sample target objects, and conversion labels. The sample target objects refer to objects taken out of the object set and used as samples. The sample recommended objects refer to objects taken out of the recommended objects recommended for the sample target objects and used as samples. The sample target objects and sample recommended objects here are similar to the target objects and candidate recommended objects of step 310, except that the sample target objects and sample recommended objects are used for model training, and the target objects and candidate recommended objects are used for actual use.

[0379] A conversion label is added to the recommended event samples in the recommended event sample set in advance, and the conversion label indicates whether the sample target object adds the sample recommended object as an associated object of the sample target object after the sample recommended object is recommended for the sample target object. Labeling can be done by manual annotation.

[0380] Next, each recommended event sample is input into the conversion rate prediction model to obtain the predicted probability of adding the sample recommended object as an associated object to the sample target object, and then a loss function is constructed based on the conversion label and the predicted probability of each recommended event sample, and the conversion rate prediction model is trained based on the loss function. For example, the cross entropy loss function can be calculated based on the conversion label and the predicted probability to construct the first error function. For another example, the vector distance can be calculated based on the conversion label and the predicted probability to construct the first error function.

[0381] After constructing the first error function, the parameters of the conversion rate prediction model can be adjusted based on the first error function. Specifically, a first threshold can be preset. When the first error function is less than the first threshold, the training process ends. When the first error function is greater than or equal to the first threshold, the parameters of the conversion rate prediction model are adjusted until the first error function is less than the first threshold.

[0382] The benefit of the implementation of steps 2210-2230 is that conversion rate prediction is performed on the interest tags of the first object and the second object based on the conversion rate prediction model, which can reflect the degree of conformity between the target object and the candidate recommended object in the interest tags, thereby improving the processing efficiency of the conversion rate prediction and also improving the accuracy of the conversion rate prediction.

[0383] Detailed description of step 340

[0384] In step 340, the preference conformity of each candidate recommendation object to the target object is obtained.

[0385] Reference Fig.23 In one embodiment, step 340 includes:

[0386] Step 2310: Obtain historical recommendation records for the target object;

[0387] Step 2320: Acquire multiple converted recommended objects of the target object in the historical recommendation records;

[0388] Step 2330: clustering the multiple converted recommended objects based on the attributes of the multiple converted recommended objects to form multiple recommended object classes;

[0389] Step 2340: Generate a conversion preference label for the target object based on the recommended object class that meets the predetermined criteria among the multiple recommended object classes;

[0390] Step 2350: Obtain preference compliance based on the conversion preference label of the target object and the matching of the attributes of the candidate recommendation objects.

[0391] Steps 2310-2350 are described in detail below.

[0392] In step 2310, the historical recommendation record refers to the record of recommending candidate recommendation objects for the target object before the current recommendation processing service. The candidate recommendation record can be obtained from the object profile of the target object or from the recommendation service log. Fig.24 , the historical recommendation records include {on XX / XX / XX, historical object 1, historical object 2, historical object 3, historical object 4, and historical object 5 were recommended to the target object. Then, the target object converted historical object 2, historical object 3, and historical object 5}.

[0393] In step 2320, the historical recommendation record recommends multiple historical objects for the target object. These historical objects may or may not be converted by the target object. The conversion here means that after the historical object is recommended to the target object, the target object and the historical object become associated objects. Fig.24 , it is obtained from the historical recommendation records that the target object has been converted into historical object 2, historical object 3, and historical object 5. Next, historical object 2 is used as the converted recommended object 1, historical object 3 is used as the converted recommended object 2, and historical object 5 is used as the converted recommended object 3, and the converted recommended objects include {converted recommended object 1, converted recommended object 2, converted recommended object 3}.

[0394] In step 2330, clustering is performed based on the attributes of the multiple post-conversion recommended objects, with the purpose of forming multiple recommended object classes, so as to obtain representative conversion preference labels based on the recommended object classes. Fig.24 , the attributes of the converted recommended object 1 include {interests, work, educational background, geographic location}, and the attribute values ​​of the converted recommended object 1 include {playing badminton, student, university bachelor, L city}. The attributes of the converted recommended object 2 include {interests, work, educational background, geographic location}, and the first attribute value of the converted recommended object 2 includes {playing badminton, student, university doctor, G city}. The attributes of the converted recommended object 3 are {interests, work, geographic location}, and the first attribute value of the converted recommended object 3 is {playing tennis, student, M city}. The converted recommended objects 1-3 are clustered, and the recommended object classes include {recommended object class 1, recommended object class 2}. Recommended object class 1 includes {converted recommended object 1, converted recommended object 2}. Recommended object class 2 includes {converted recommended object 3}.

[0395] In step 2340, based on the recommended object classes that meet the predetermined criteria in the multiple recommended object classes, a conversion preference label for the target object is generated. The predetermined criteria refers to a pre-set screening condition for the recommended object class. For example, the predetermined criteria is that the number of recommended objects after conversion in the recommended object class is greater than or equal to a quantity threshold. For another example, the predetermined criteria is that the number of a certain attribute in the recommended object class is greater than or equal to a quantity threshold.

[0396] In one embodiment, step 2340 includes:

[0397] Determining a recommended object class that meets a predetermined criterion among a plurality of recommended object classes;

[0398] Obtain recommended object attributes of each recommended object in the recommended object class that meets the predetermined criteria;

[0399] For each acquired recommended object attribute, the ratio of recommended objects having the recommended object attribute in the recommended object class is determined, and if the recommended object ratio meets a predetermined condition, the recommended object attribute is used as a conversion preference label of the target object.

[0400] In this embodiment, please refer to the above for the explanation of the predetermined criteria. The predetermined condition refers to a pre-set screening condition for the recommended object ratio. For example, the predetermined condition is that the recommended object ratio is greater than or equal to the ratio threshold. For another example, the predetermined condition is that the recommended object ratio ranks among the top several among multiple recommended object ratios.

[0401] In one example, it is assumed that the predetermined standard is the first case above, and the quantity threshold is 2. Fig.24, since the number of converted recommended objects in recommended object class 1 is 2, which is equal to the quantity threshold, the recommended object class that meets the predetermined criteria is determined to be recommended object class 1. Recommended object class 1 includes recommended object 1 and recommended object 2. The recommended object attributes obtained for recommended object 1 include {(interest, playing badminton), (work, student), (educational background, university bachelor), (geographical location, L city)}. The recommended object attributes obtained for recommended object 2 include {(interest, playing badminton), (work, student), (educational background, university doctor), (geographical location, G city)}. For the recommended object attribute of "interest, playing badminton", it can be found that recommended object 1 and recommended object 2 both have the recommended object attribute, so it is determined that the ratio of recommended objects with the recommended object attribute of "interest, playing badminton" is 100%. For the recommended object attribute of "work, student", it can be found that recommended object 1 and recommended object 2 both have the recommended object attribute, so it is determined that the ratio of recommended objects with the recommended object attribute of "work, student" is 100%. For the recommended object attribute of "educational background, university bachelor's degree", it can be found that recommended object 1 has this recommended object attribute, so it is determined that the ratio of recommended objects with the recommended object attribute of "educational background, university bachelor's degree" is 50%. For the recommended object attribute of "educational background, university doctorate", it can be found that recommended object 2 has this recommended object attribute, so it is determined that the ratio of recommended objects with the recommended object attribute of "educational background, university doctorate" is 50%. The process of determining the recommended object ratios of other recommended object attributes is similar and will not be repeated here. Then assume that the predetermined condition is that the recommended object ratio is greater than or equal to the ratio threshold, and the ratio threshold is 70%. Since the ratio of recommended objects with the recommended object attribute of "work, student" is 100%, and the ratio of recommended objects with the recommended object attribute of "work, student" is 100%, both of which are greater than the ratio threshold, therefore, the converted preference labels generated for the target object include {(interest, playing badminton), (work, student)}.

[0402] In step 2350, based on the matching of the conversion preference label of the target object and the attributes of the candidate recommended object, the preference conformity is obtained. The higher the matching degree between the conversion preference label and the attributes of the candidate recommended object, the higher the preference conformity. Fig.24, assuming that the first attribute value of candidate recommendation object 1 includes {playing badminton, student, university bachelor, G city}; and assuming that the first attribute value of candidate recommendation object 2 includes {swimming, teacher, university doctor, L city}. Since the target object and candidate recommendation object 1 have the same attribute values ​​in interests and work, the target object's preference conformity for candidate recommendation object 1 can be 0.7. Since the target object and candidate recommendation object 2 do not have the same attribute values, the target object's preference conformity for candidate recommendation object 2 can be 0.1. The preference conformity finally obtained includes {the target object's preference conformity for candidate recommendation object 1 is 0.7, and the target object's preference conformity for candidate recommendation object 2 is 0.1}.

[0403] The benefit of the implementation of steps 2310-2250 is that by utilizing historical recommendation records and clustering algorithms, the conversion preference labels of the target objects can be statistically calculated, thereby calculating the preference compliance of each candidate recommendation object using the conversion preference labels, thereby improving the accuracy of obtaining the preference compliance and improving the sorting accuracy.

[0404] Detailed description of step 350

[0405] In step 350, based on relationship intimacy, conversion rate, and preference conformity, multiple candidate recommendation objects are sorted for display to the target object.

[0406] In one embodiment, step 350 includes:

[0407] Based on the relationship intimacy, conversion rate, and preference conformity, obtaining a first ranking of multiple candidate recommendation objects;

[0408] Based on the first ranking, a plurality of candidate recommendation objects are displayed.

[0409] In this embodiment, multiple candidate recommendation objects may be ranked based on the average or weighted average of relationship intimacy, conversion rate, and preference conformity to obtain a first ranking. Then, multiple candidate recommendation objects are displayed based on the first ranking. The advantage of this is that the ranking accuracy is improved by using the three ranking indicators of relationship intimacy, conversion rate, and preference conformity to rank each candidate recommendation object, thereby improving the conversion rate of the recommendation process.

[0410] In another embodiment, considering the influence of the diversity of candidate recommendation objects on the recommendation process, it is necessary to perform a second sorting based on the first sorting to improve the diversity of candidate recommendation objects.

[0411] When this embodiment is implemented, refer to Fig.25 , step 350 comprises:

[0412] Step 2510: Obtain a first ranking of multiple candidate recommendation objects based on relationship intimacy, conversion rate, and preference conformity;

[0413] Step 2520: Obtain object attributes of multiple candidate recommendation objects;

[0414] Step 2530: If, for each object attribute, among the first first number of candidate recommendation objects in the first sorting, the number of candidate recommendation objects carrying the object attribute is greater than a second number, then the number of candidate recommendation objects carrying the object attribute is reduced to no greater than the second number, thereby obtaining a second sorting, wherein the second number is less than the first number;

[0415] Step 2540: Based on the second ranking, display multiple candidate recommendation objects.

[0416] Steps 2510-2540 are described in detail below.

[0417] In step 2510, multiple candidate recommendation objects are ranked based on relationship intimacy, conversion rate, and preference conformity. Referring to the above, an average may be calculated based on relationship intimacy, conversion rate, and preference conformity, and then multiple candidate recommendation objects are ranked based on the average.

[0418] In one embodiment, referring to Fig.26 , step 2510 includes:

[0419] Step 2610: taking the relationship intimacy, conversion rate, and preference conformity as multiple targets for recommendation processing, and using multi-target fusion to determine a first weight for the relationship intimacy, a second weight for the conversion rate, and a third weight for the preference conformity;

[0420] Step 2620: Based on the first weight, the second weight, and the third weight, a weighted sum is performed on the relationship intimacy, the conversion rate, and the preference compliance;

[0421] Step 2630: Based on the weighted sum, perform a first sorting of the multiple candidate recommendation objects.

[0422] Steps 2610-2630 are described in detail below.

[0423] In step 2610, multi-objective fusion refers to a method of fusing multiple objectives using multi-objective fusion technology to obtain multiple weights. Multi-objective fusion technology is a hyperparameter optimization problem. The current typical hyperparameter optimization methods are mainly divided into Bayesian optimization, evolutionary learning-based methods and reinforcement learning-based methods. The simplest solution is to configure hyperparameters by artificial experience. In this embodiment, multi-objective fusion technology can be implemented using a multi-objective fusion (MTF) model. For example, relationship intimacy, conversion rate, and preference compliance are used as multiple objectives for recommendation processing, and multiple objectives are input into the multi-objective fusion model to obtain a first weight of relationship intimacy, a second weight of conversion rate, and a third weight of preference compliance.

[0424] After step 2610, a weighted sum is calculated for the relationship intimacy, conversion rate, and preference conformity based on the first weight, the second weight, and the third weight in step 2620. Then, based on the weighted sum, a first ranking is performed on the plurality of candidate recommendation objects in step 2630. It can be understood that the higher the weighted sum of the candidate recommendation objects, the higher the first ranking.

[0425] The benefit of the embodiment of steps 2610-2630 is that by using multi-objective fusion to determine the first weight, the second weight, and the third weight, and then calculating the weighted sum, different weights can be set for relationship closeness, conversion rate, and preference compliance, thereby improving the flexibility of the first sorting.

[0426] Then in steps 2520-2540, object attributes of multiple candidate recommendation objects are obtained; if for each object attribute, among the first first number of candidate recommendation objects in the first sort, the number of candidate recommendation objects carrying the object attributes is greater than a second number, then the number of candidate recommendation objects carrying the object attributes is reduced to no more than the second number, thereby obtaining a second sort; based on the second sort, multiple candidate recommendation objects are displayed.

[0427] The advantage of the embodiment of steps 2510-2540 is that, based on the first sorting, if the number of candidate recommendation objects carrying a certain object attribute is greater than the second number, the number of candidate recommendation objects is reduced. In this way, among the first number of candidate recommendation objects in the first sorting, no more than the second number of candidate recommendation objects have the same object attribute, which improves the diversity of the candidate recommendation objects in terms of object attributes, thereby improving the sorting accuracy and the conversion rate of the recommendation process.

[0428] In another embodiment, referring to Fig. 27 , step 350 comprises:

[0429] Step 2710: sort multiple candidate recommendation objects;

[0430] Step 2720: while displaying the candidate recommendation objects in order, display the sources of the candidate recommendation objects, including: a process of obtaining a first candidate recommendation object based on object element nodes, content author nodes, and target objects; a process of obtaining a second candidate recommendation object based on object interest tag nodes and target objects; and a process of obtaining a third candidate recommendation object by integrating a multi-scenario element interconnection graph and a target object input relationship and interest into a recommendation model.

[0431] For steps 2710-2720, in one example, refer to Fig.28 , after sorting multiple candidate recommendation objects, the sorting is {candidate recommendation object 1 "Flying Qiuqiu", candidate recommendation object 2 "Chong Er", candidate recommendation object 3 "Sky"}. Then, while displaying the candidate recommendation objects in the sorting order, the source of the candidate recommendation objects needs to be displayed. Assuming that candidate recommendation object 1 comes from "the process of obtaining the first candidate recommendation object based on the object element node, the content author node, and the target object", while displaying candidate recommendation object 1 "Flying Qiuqiu", "Source: Recommended based on relationship chain" is displayed. Assuming that candidate recommendation object 2 comes from "the process of obtaining the second candidate recommendation object based on the object interest tag node and the target object", while displaying candidate recommendation object 2 "Chong Er", "Source: Recommended based on interest points" is displayed. Assuming that candidate recommendation object 3 comes from "the process of integrating the multi-scene element interconnection graph and the target object input relationship and interest recommendation model to obtain the third candidate recommendation object", while displaying candidate recommendation object 3 "Sky", "Source: Recommended based on relationship chain and interest points" is displayed.

[0432] The advantage of the implementation of steps 2710-2720 is that, when recommending candidate recommendation objects, the segmented channels of the recommendation sources can be well segmented, thereby giving very clear reasons for the recommendation, which is very helpful for the target object to understand, thereby effectively improving the actual conversion rate of the recommendation results to friend relationships, and helping to make the recommendation processing platform active.

[0433] like Fig.29 As shown, the recommendation processing method according to an embodiment of the present disclosure may include:

[0434] Step 310: Based on the object relationship data and object interaction data in multiple scenes in the application, obtain multiple candidate recommended objects for the target object;

[0435] Step 320: Obtain the relationship intimacy between each candidate recommendation object and the target object;

[0436] Step 330: predicting the conversion rate after recommending each candidate recommended object to the target object, where the conversion rate is the probability that the target object pays attention to the candidate recommended object after the candidate recommended object is recommended to the target object;

[0437] Step 340: Obtain the preference conformity of each candidate recommendation object to the target object;

[0438] Step 2910: predicting the retention rate after recommending each candidate recommended object to the target object;

[0439] Step 350 includes: Step 2920, sorting multiple candidate recommendation objects based on relationship intimacy, conversion rate, retention rate and preference compliance.

[0440] Since steps 310-340 have been described in detail above, they will not be repeated here. Steps 2910 and 2920 are described in detail below.

[0441] In step 2910, the retention rate is the probability that after the candidate recommendation object is recommended to the target object, the target object follows the candidate recommendation object and does not cancel the follow-up within a predetermined time period after the current time. The higher the retention rate, the higher the candidate recommendation object is ranked in the sorting. The process of predicting the retention rate is similar to the process of predicting the conversion rate, which will not be repeated here. For example, assuming there are candidate recommendation object 1 and candidate recommendation object 2, the predicted retention rate of candidate recommendation object 1 is 0.7, and the predicted retention rate of candidate recommendation object 2 is 0.3.

[0442] In step 2920, multiple candidate recommendation objects are sorted based on relationship intimacy, conversion rate, retention rate, and preference compliance. Compared with the process of sorting multiple candidate recommendation objects based on relationship intimacy, conversion rate, and preference compliance in step 350, step 2920 has an additional retention rate, but the specific sorting process of step 2920 is similar to the process of sorting multiple candidate recommendation objects based on relationship intimacy, conversion rate, and preference compliance, and will not be repeated here.

[0443] The advantages of the implementation of steps 310-340 and steps 2910-2920 are that (1) they can make full use of the object relationship data and object interaction data in multiple scenarios of the application to organize them into a huge multi-scenario element interconnected graph, avoiding the limitations of single-scenario data and the insufficient number of objects, and can more accurately recommend candidate recommended objects of interest to the target object, thereby improving the effect of the recommendation processing and the object satisfaction; (2) through the design of recall and sorting layers, different layers can be optimized separately without affecting each other, and the final recommended results can often have multiple associated relationships with the target object, thereby improving diversity; (3) because the target objects have different feedback preferences, comprehensive consideration of multiple sorting indicators for fusion and rearrangement improves sorting accuracy and the conversion rate of recommendation processing, thereby maximizing long-term business gains.

[0444] Implementation details of the recommended processing method of the disclosed embodiment

[0445] Refer to the following Fig.30 , the implementation details of the recommended processing method of the embodiment of the present disclosure are explained in detail and by way of example.

[0446] The embodiments of the present disclosure specifically include the following implementation details:

[0447] (1) Obtain object relationship data, object interaction data, and content association data in multiple scenarios in the application.

[0448] (2) Based on the object relationship data in multiple scenes, object element node data is generated and placed in the first subspace, and based on the object interaction data and content association data in multiple scenes, multiple types of auxiliary element node data are generated and placed in multiple second subspaces; wherein the auxiliary element node data includes: object profile node data, object interest tag node data, content node data, content author node data, recommended activity node data, content attribute node data, content author attribute node data, and content node relationship data.

[0449] (3) Based on the object element node data in the first subspace and the auxiliary element node data in multiple second subspaces, a multi-scene element interconnection graph is generated.

[0450] (4) At the recall layer, multiple candidate recommendation objects are obtained based on the interconnected graph of multiple scene elements. Specifically, relationship chain recall, point of interest recall, and relationship chain and point of interest recall can be performed. Among them, relationship chain recall includes multiple dimensions, such as commonly associated objects, the same specific attributes, the same group, the same application session group, content authors with interactive relationships, lists of associated objects outside the application, and content authors who follow each other. Point of interest recall includes multiple dimensions, such as label similarity of object interest label nodes, and comparison based on the first point of interest vector and the second point of interest vector. Relationship chain and point of interest recall includes multiple dimensions, such as deep learning networks (such as the DSSM dual-tower model), and relationship and interest recommendation models.

[0451] (5) At the ranking layer, multiple candidate recommendation objects are ranked using relationship closeness, conversion rate, preference conformity, and diversity.

[0452] (6) At the recommendation processing application layer, while displaying the candidate recommendation objects in order, the sources of the candidate recommendation objects are displayed. For example, the source of the candidate recommendation objects is displayed as a relationship chain recall.

[0453] Advantages of this embodiment include, but are not limited to, improving sorting accuracy and recommendation processing conversion rate.

[0454] Implementation process of the recommendation processing method of the disclosed embodiment in the friend recommendation scenario

[0455] Refer to the following Fig.31 , the implementation process of the recommended processing method of the embodiment of the present disclosure is described in detail and exemplarily.

[0456] The embodiment of the present disclosure specifically includes the following implementation process:

[0457] Step 3111: The object terminal 130 reports data of multiple application scenarios to the reporting and analysis interface. The data of multiple application scenarios include: object relationship data, object interaction data, and content association data.

[0458] Step 3112: The statistical analysis database reads and stores the behavior statistical information from the reporting and analysis interface. Specifically, the object relationship data, the object interaction data, and the content association data are preliminarily analyzed to obtain statistical information, such as object element node data and auxiliary element node data, and then the object element node data is placed in the first subspace of the statistical analysis database, and the auxiliary element node data is placed in the second subspace of the statistical analysis database.

[0459] Step 3113: The statistical analysis database mines the object profile information and writes the object profile information results. For example, the object profile information includes object attributes, and the object profile information results include age, gender, education, city, company, etc.

[0460] Step 3114: Generate a heterogeneous GNN graph network model based on the statistical analysis database and object profile information. The heterogeneous GNN graph network model is the multi-scenario element interconnection graph mentioned above, for example, generating object element nodes in the multi-scenario element interconnection graph based on the object element node data in the first subspace, and generating auxiliary element nodes in the multi-scenario element interconnection graph based on the auxiliary element node data in the second subspace.

[0461] Step 3115: In the friend recommendation scenario, a friend recommendation model is formed based on the heterogeneous GNN graph network model.

[0462] Step 3116: transform the friend recommendation model into a service model to obtain a friend recommendation service.

[0463] Step 3117: The target terminal 130 calls the friend recommendation service to recommend friends to the target that calls the friend recommendation service.

[0464] Step 3121: The target terminal 130 sends a message to the message and content service access server or obtains content.

[0465] Step 3122: The message and content service access server writes messages to the message system, or accesses different sub-services to the full-scenario content service.

[0466] Step 3123: The message system writes and stores the message to the message and content database.

[0467] Step 3124: The message system sends the message to the message and content service access server.

[0468] Step 3125: The message system sends the sub-service recommendation result to the message and content service access server.

[0469] Step 3126: The message and content service access server sends a synchronization message to the target terminal or sends a recommendation result.

[0470] The advantages of this embodiment include but are not limited to: by utilizing the interaction between the object terminal and the content service access server, message system and other modules, the object relationship data, object interaction data, and content association data of the object terminal in different application scenarios can be collected to generate a friend recommendation model, which can improve the conversion rate of friend recommendations in the friend recommendation scenario.

[0471] Description of the apparatus and device of the present disclosure

[0472] It is to be understood that, although the steps in the above-mentioned flowcharts are sequentially displayed according to the characterization of arrows, these steps are not necessarily executed in sequence according to the order of arrow characterization. Unless there is a clear description in the present embodiment, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the above-mentioned flowcharts can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.

[0473] It should be noted that in various specific implementations of the present disclosure, when it comes to the need to perform relevant processing based on data related to the characteristics of the target object such as the target object attribute information or attribute information set, the permission or consent of the target object will be obtained first, and the collection, use and processing of these data will comply with relevant laws, regulations and standards. In addition, when the embodiment of the present disclosure needs to obtain the attribute information of the target object, the separate permission or separate consent of the target object will be obtained through a pop-up window or jump to a confirmation page. After clearly obtaining the separate permission or separate consent of the target object, the necessary target object-related data for the normal operation of the embodiment of the present disclosure will be obtained.

[0474] Reference Fig.32 According to one aspect of the present disclosure, a recommendation processing device 3200 is provided, comprising:

[0475] A first acquisition unit 3210 is used to acquire multiple candidate recommended objects of a target object based on object relationship data and object interaction data in multiple scenes in an application;

[0476] The second acquisition unit 3220 is used to acquire the relationship intimacy between each candidate recommendation object and the target object;

[0477] A prediction unit 3230 is used to predict a conversion rate after each candidate recommendation object is recommended to the target object. The conversion rate is a probability that the target object pays attention to the candidate recommendation object after the candidate recommendation object is recommended to the target object.

[0478] The third acquisition unit 3240 is used to obtain the preference conformity of each candidate recommendation object to the target object;

[0479] The sorting unit 3250 is used to sort multiple candidate recommendation objects based on relationship intimacy, conversion rate, and preference compliance for display to the target object.

[0480] Optionally, the second acquiring unit 3220 is used to:

[0481] Obtain the attribute similarity between each candidate recommendation object and the target object;

[0482] Obtain the interaction frequency between each candidate recommendation object and the target object in the application;

[0483] Get the number of common related objects between each candidate recommendation object and the target object in the application;

[0484] Obtain the main interaction type between each candidate recommendation object and the target object in the application;

[0485] Based on attribute similarity, interaction frequency, number of commonly associated objects, and main interaction type, the relationship closeness between each candidate recommendation object and the target object is determined.

[0486] Optionally, the second acquiring unit 3220 is specifically configured to:

[0487] Obtaining first attribute values ​​of multiple first attributes of each candidate recommendation object and second attribute values ​​of multiple first attributes of the target object;

[0488] Generate a first vector based on first attribute values ​​of a plurality of first attributes of the candidate recommendation object, and generate a second vector based on second attribute values ​​of a plurality of first attributes of the target object;

[0489] Based on the distance between the first vector and the second vector, the attribute similarity is obtained.

[0490] Optionally, the second acquiring unit 3220 is specifically configured to:

[0491] Identify interactions between the candidate recommendation object and the target object in the application within a predetermined period before the current time;

[0492] Counting the number of interactions within a predetermined period;

[0493] Based on the number of counted interactions and the predetermined period length, an interaction frequency is determined.

[0494] Optionally, the second acquiring unit 3220 is specifically configured to:

[0495] In a multi-scenario element interconnection graph generated based on object relationship data and object interaction data, other object element nodes adjacent to the object element node of the candidate recommendation object are identified as associated objects of the candidate recommendation object in the application, and other object element nodes adjacent to the object element node of the target object are identified as associated objects of the target object in the application;

[0496] The number of objects in the intersection of the associated objects of the candidate recommendation object in the application and the associated objects of the target object in the application is determined as the number of commonly associated objects.

[0497] Optionally, the second acquiring unit 3220 is specifically configured to:

[0498] Identify interactions between the candidate recommendation object and the target object in the application within a predetermined period before the current time;

[0499] Get the interaction type of the identified interaction;

[0500] Determine the number of interactions of each interaction type captured;

[0501] Based on the number of interactions, a primary interaction type is determined among the identified interaction types.

[0502] Optionally, the second acquiring unit 3220 is specifically configured to:

[0503] Based on the attribute similarity, determining a first score of the candidate recommendation object;

[0504] Based on the interaction frequency, determining a second score of the candidate recommendation object;

[0505] Determining a third score of the candidate recommendation object based on the number of commonly associated objects;

[0506] Based on the primary interaction type, determining a fourth score of the candidate recommendation object;

[0507] Based on the first score, the second score, the third score, and the fourth score of each candidate recommendation object, the relationship closeness between the candidate recommendation object and the target object is determined.

[0508] Optionally, the prediction unit 3230 is used to:

[0509] Obtain a first object interest tag of a target object;

[0510] Obtain a second object interest tag for each candidate recommendation object;

[0511] The first object interest tag and the second object interest tag are input into the conversion rate prediction model to obtain a predicted conversion rate.

[0512] Optionally, the third obtaining unit 3240 is used to:

[0513] Obtain historical recommendation records for the target object;

[0514] Obtain multiple post-conversion recommended objects in which the target object has been converted in the historical recommendation records;

[0515] Based on the attributes of the multiple post-conversion recommendation objects, the multiple post-conversion recommendation objects are clustered to form multiple recommendation object classes;

[0516] generating a conversion preference label for a target object based on a recommended object class that meets a predetermined standard among a plurality of recommended object classes;

[0517] Based on the conversion preference label of the target object and the matching of the attributes of the candidate recommendation object, the preference compliance is obtained.

[0518] Optionally, the sorting unit 3250 is used to:

[0519] Based on the relationship intimacy, conversion rate, and preference conformity, obtaining a first ranking of multiple candidate recommendation objects;

[0520] Get object attributes of multiple candidate recommendation objects;

[0521] If, for each object attribute, among the first first number of candidate recommendation objects in the first sorting, the number of candidate recommendation objects carrying the object attribute is greater than a second number, then the number of candidate recommendation objects carrying the object attribute is reduced to no greater than the second number, thereby obtaining a second sorting, wherein the second number is less than the first number;

[0522] Based on the second ranking, a plurality of candidate recommendation objects are displayed.

[0523] Optionally, the sorting unit 3250 is specifically configured to:

[0524] Taking relationship intimacy, conversion rate, and preference conformity as multiple targets for recommendation processing, and using multi-target fusion to determine a first weight of relationship intimacy, a second weight of conversion rate, and a third weight of preference conformity;

[0525] Based on the first weight, the second weight, and the third weight, a weighted sum is performed on the relationship intimacy, the conversion rate, and the preference compliance;

[0526] Based on the weighted sum, a first ranking is performed on the plurality of candidate recommendation objects.

[0527] Optionally, the first acquiring unit 3210 is used to:

[0528] Based on the object relationship data and object interaction data in multiple scenes in the application, a multi-scene element interconnection graph is generated, wherein the multi-scene element interconnection graph includes multiple element nodes, the multiple element nodes include object element nodes and auxiliary element nodes, the object element nodes and the relationship between each object element node are generated by the object relationship data, and the auxiliary element nodes and the relationship between the auxiliary element nodes and the object element nodes are generated by the object interaction data;

[0529] Based on the interconnected graph of multiple scene elements and the target object, multiple candidate recommended objects of the target object are obtained.

[0530] Optionally, the auxiliary element node includes a content author node and an object interest tag node; the first acquisition unit 3210 is specifically used to:

[0531] Based on the object element node, the content author node, and the target object, obtain a first candidate recommendation object;

[0532] Based on the object interest tag node and the target object, obtain a second candidate recommended object;

[0533] The interconnected graph of multiple scene elements and the target object are input into the relationship and interest fusion recommendation model to obtain the third candidate recommendation object;

[0534] The first candidate recommendation object, the second candidate recommendation object, and the third candidate recommendation object are integrated into a plurality of candidate recommendation objects.

[0535] Optionally, the sorting unit 3250 is used to:

[0536] Sort multiple candidate recommendation objects;

[0537] While displaying the candidate recommendation objects in order, the sources of the candidate recommendation objects are also displayed, including: the process of obtaining the first candidate recommendation object based on the object element node, the content author node, and the target object; the process of obtaining the second candidate recommendation object based on the object interest tag node and the target object; the process of obtaining the third candidate recommendation object by integrating the multi-scenario element interconnection graph and the target object input relationship and interest into the recommendation model.

[0538] Optionally, the first acquiring unit 3210 is specifically configured to:

[0539] In the multi-scenario element interconnection graph, an object element node having a common associated object element node with the object element node of the target object is obtained, and the object corresponding to the obtained object element node is used as the first sub-candidate recommendation object;

[0540] In each object element node of the multi-scenario element interconnection graph, an object element node having the same specific attribute as the target object is obtained, and the object corresponding to the obtained object element node is used as the second sub-candidate recommendation object;

[0541] Based on the attributes of each object element node in the multi-scenario element interconnection graph, each object element node is grouped, object element nodes that are grouped in the same group as the object element node of the target object are obtained, and the object corresponding to the obtained object element node is used as the third sub-candidate recommendation object;

[0542] From the objects corresponding to the object element nodes of the multi-scenario element interconnection graph, an object belonging to the same application session group as the target object is obtained as a fourth sub-candidate recommendation object;

[0543] In the multi-scenario element interconnection graph, a content author node having an interactive relationship with an object element node of the target object is obtained, and the object corresponding to the obtained content author node is used as the fifth child candidate recommendation object;

[0544] Based on the list of associated objects of the target object outside the application, obtain a sixth sub-candidate recommendation object;

[0545] In the multi-scenario element interconnection graph, a content author node that is mutually concerned with the object element node of the target object is obtained, and the object corresponding to the obtained content author node is used as the seventh sub-candidate recommendation object;

[0546] The first sub-candidate recommendation object, the second sub-candidate recommendation object, the third sub-candidate recommendation object, the fourth sub-candidate recommendation object, the fifth sub-candidate recommendation object, the sixth sub-candidate recommendation object, and the seventh sub-candidate recommendation object are integrated into the first candidate recommendation object.

[0547] Optionally, the auxiliary element node includes a content node; the first acquisition unit 3210 is specifically used to:

[0548] Based on the object interest label node of the target object and the label similarity of the object interest label nodes of each other object, an eighth sub-candidate recommendation object is obtained;

[0549] Based on the interaction data of the target object node with each content node in the multi-scenario element interconnection graph, a first point of interest vector of the target object node is obtained; based on the interaction data of each other object node with each content node in the multi-scenario element interconnection graph, a second point of interest vector of each other object node is obtained; and based on the comparison between the first point of interest vector and the second point of interest vector, a ninth sub-candidate recommendation object is obtained;

[0550] The eighth sub-candidate recommendation object and the ninth sub-candidate recommendation object are integrated into the second candidate recommendation object.

[0551] Reference Fig.33 , Fig.33 The structural block diagram of the terminal part for implementing the recommended processing method of the embodiment of the present disclosure includes: a radio frequency (RF) circuit 3310, a memory 3315, an input unit 3330, a display unit 3340, a sensor 3350, an audio circuit 3360, a wireless fidelity (WiFi) module 3370, a processor 3380, and a power supply 3390. Those skilled in the art can understand that Fig.33 The terminal structure shown in the figure does not constitute a limitation on the mobile phone or computer, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0552] The RF circuit 3310 can be used for receiving and sending signals during information transmission or calls. In particular, after receiving the downlink information from the base station, it is sent to the processor 3380 for processing; in addition, the designed uplink data is sent to the base station.

[0553] The memory 3315 may be used to store software programs and modules. The processor 3380 executes various functional applications and data processing of the terminal by running the software programs and modules stored in the memory 3315 .

[0554] The input unit 3330 may be used to receive input digital or character information and generate key signal input related to the terminal's settings and function control. Specifically, the input unit 3330 may include a touch panel 3331 and other input devices 3332 .

[0555] The display unit 3340 may be used to display input information or provided information and various menus of the terminal. The display unit 3340 may include a display panel 3341.

[0556] The audio circuit 3360, the speaker 3361, and the microphone 3362 may provide an audio interface.

[0557] In this embodiment, the processor 3380 included in the terminal can execute the recommended processing method of the previous embodiment.

[0558] The terminals of the embodiments of the present disclosure include but are not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle terminals, aircraft, etc. The embodiments of the present disclosure can be applied to various scenarios, including but not limited to big data scenarios, data processing scenarios, etc.

[0559] Reference Fig.34 , Fig.34The structural block diagram of part of the server for implementing the recommended processing method of the embodiment of the present disclosure, the server may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 3422 (for example, one or more processors) and memory 3432, one or more storage media 3430 (for example, one or more mass storage devices) storing application programs 3442 or data 3444. Among them, the memory 3432 and the storage medium 3430 can be short-term storage or permanent storage. The program stored in the storage medium 3430 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 3422 can be configured to communicate with the storage medium 3430 to execute a series of instruction operations in the storage medium 3430 on the server.

[0560] The server may also include one or more power supplies 3426, one or more wired or wireless network interfaces 3450, one or more input and output interfaces 3458, and / or one or more operating systems 3441, such as Windows ServerTM, Mac OS XTM, UnixTM, Li nuxTM, FreeBSDTM, etc.

[0561] The central processor 3422 in the server can be used to execute the recommendation processing method of the embodiment of the present disclosure.

[0562] The embodiments of the present disclosure also provide a computer-readable storage medium, which is used to store program codes, and the program codes are used to execute the recommendation processing methods of the aforementioned embodiments.

[0563] The embodiment of the present disclosure also provides a computer program product, which includes a computer program. A processor of a computer device reads and executes the computer program, so that the computer device executes and implements the above-mentioned recommendation processing method.

[0564] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can, for example, be implemented in an order other than those illustrated or described herein. In addition, the terms "comprises" and "comprising" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0565] It should be understood that in the present disclosure, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0566] It should be understood that in the description of the embodiments of the present disclosure, the meaning of multiple (or multiple items) is more than two, greater than, less than, exceed, etc. are understood to not include the number, and above, below, within, etc. are understood to include the number.

[0567] In the several embodiments provided in the present disclosure, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0568] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0569] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0570] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present disclosure is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store program codes.

[0571] It should also be understood that the various implementations provided in the embodiments of the present disclosure can be combined arbitrarily to achieve different technical effects.

[0572] The above is a specific description of the implementation methods of the present disclosure, but the present disclosure is not limited to the above implementation methods. Technical personnel familiar with the art can also make various equivalent modifications or substitutions without violating the spirit of the present disclosure. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present disclosure.

Claims

1. A recommendation processing method, characterized in that: include: Based on object relationship data and object interaction data in multiple scenarios in the application, multiple candidate recommended objects of the target object are obtained; Obtaining the degree of intimacy between each of the candidate recommendation objects and the target object; Predicting a conversion rate after recommending each of the candidate recommended objects to the target object, wherein the conversion rate is a probability that the target object pays attention to the candidate recommended object after the candidate recommended object is recommended to the target object; Obtaining the preference conformity of each candidate recommendation object to the target object; Based on the relationship intimacy, the conversion rate, and the preference conformity, the plurality of candidate recommendation objects are sorted for display to the target object.

2. The recommendation processing method according to claim 1, characterized in that: The obtaining of the degree of intimacy between each candidate recommendation object and the target object includes: Obtaining the attribute similarity between each of the candidate recommended objects and the target object; Obtaining the interaction frequency between each of the candidate recommendation objects and the target object in the application; Obtaining the number of commonly associated objects between each of the candidate recommended objects and the target object in the application; Obtaining a main interaction type between each of the candidate recommendation objects and the target object in the application; Based on the attribute similarity, the interaction frequency, the number of commonly associated objects, and the main interaction type, the relationship closeness between each of the candidate recommended objects and the target object is determined.

3. The recommendation processing method according to claim 2, characterized in that: The obtaining of the attribute similarity between each of the candidate recommended objects and the target object includes: Acquire first attribute values ​​of multiple first attributes of each of the candidate recommendation objects and second attribute values ​​of multiple first attributes of the target object; generating a first vector based on the first attribute values ​​of the plurality of first attributes of the candidate recommendation objects, and generating a second vector based on the second attribute values ​​of the plurality of first attributes of the target object; The attribute similarity is obtained based on the distance between the first vector and the second vector.

4. The recommendation processing method according to claim 2, characterized in that: The obtaining the interaction frequency between each of the candidate recommendation objects and the target object in the application includes: Identify the interaction between the candidate recommendation object and the target object in the application within a predetermined period before the current time; Counting the number of interactions within the predetermined period; The interaction frequency is determined based on the counted number of interactions and the predetermined period length.

5. The recommendation processing method according to claim 2, characterized in that: The obtaining the number of common associated objects between each candidate recommendation object and the target object in the application includes: In a multi-scenario element interconnection graph generated based on the object relationship data and the object interaction data, identifying other object element nodes adjacent to the object element node of the candidate recommended object as associated objects of the candidate recommended object in the application, and identifying other object element nodes adjacent to the object element node of the target object as associated objects of the target object in the application; The number of objects in the intersection of the associated objects of the candidate recommended object in the application and the associated objects of the target object in the application is determined as the number of commonly associated objects.

6. The recommendation processing method according to claim 2, characterized in that: The obtaining of the main interaction type between each of the candidate recommendation objects and the target object in the application includes: Identify the interaction between the candidate recommendation object and the target object in the application within a predetermined period before the current time; Obtaining the identified interaction type of the interaction; Determine the number of interactions of each interaction type captured; Based on the number of interactions, the main interaction type is determined among the identified interaction types.

7. The recommendation processing method according to claim 2, characterized in that: The determining the relationship intimacy between each of the candidate recommended objects and the target object based on the attribute similarity, the interaction frequency, the number of the commonly associated objects, and the main interaction type includes: Based on the attribute similarity, determining a first score of the candidate recommendation object; Based on the interaction frequency, determining a second score of the candidate recommendation object; Determining a third score of the candidate recommendation object based on the number of commonly associated objects; Determining a fourth score of the candidate recommendation object based on the main interaction type; Based on the first score, the second score, the third score, and the fourth score of each of the candidate recommendation objects, the closeness of the relationship between the candidate recommendation object and the target object is determined.

8. The recommendation processing method according to claim 1, characterized in that: The predicting of the conversion rate after recommending each candidate recommended object to the target object includes: Acquire a first object interest tag of the target object; Obtain a second object interest tag for each of the candidate recommendation objects; The first object interest tag and the second object interest tag are input into a conversion rate prediction model to obtain the predicted conversion rate.

9. The recommendation processing method according to claim 1, characterized in that: The obtaining of the preference conformity of each candidate recommendation object to the target object includes: Obtaining historical recommendation records for the target object; Acquire multiple post-conversion recommended objects in which the target object has been converted in the historical recommendation record; Based on the attributes of the plurality of the converted recommended objects, clustering the plurality of the converted recommended objects to form a plurality of recommended object classes; generating a conversion preference label for the target object based on a recommended object class that meets a predetermined standard among the plurality of recommended object classes; The preference compliance is obtained based on the matching of the conversion preference label of the target object and the attributes of the candidate recommendation object.

10. The recommendation processing method according to claim 1, characterized in that: The step of sorting the plurality of candidate recommendation objects based on the relationship intimacy, the conversion rate, and the preference conformity for display to the target object includes: Based on the relationship intimacy, the conversion rate, and the preference conformity, obtaining a first ranking of the plurality of candidate recommendation objects; Obtaining object attributes of a plurality of candidate recommendation objects; If, for each of the object attributes, among the first first number of the candidate recommendation objects in the first sorting, the number of the candidate recommendation objects carrying the object attribute is greater than a second number, then the number of the candidate recommendation objects carrying the object attribute is reduced to no greater than the second number, thereby obtaining a second sorting, wherein the second number is less than the first number; Based on the second ranking, a plurality of the candidate recommendation objects are displayed.

11. The recommendation processing method according to claim 10, characterized in that: The obtaining a first ranking of the plurality of candidate recommendation objects based on the relationship intimacy, the conversion rate, and the preference conformity includes: Taking the relationship intimacy, the conversion rate, and the preference conformity as multiple targets of the recommendation process, and using multi-target fusion to determine a first weight of the relationship intimacy, a second weight of the conversion rate, and a third weight of the preference conformity; Based on the first weight, the second weight, and the third weight, performing a weighted sum of the relationship intimacy, the conversion rate, and the preference compliance; Based on the weighted sum, the first sorting is performed on the plurality of candidate recommendation objects.

12. The recommendation processing method according to claim 1, characterized in that: The step of obtaining multiple candidate recommended objects for a target object based on object relationship data and object interaction data in multiple scenarios in the application includes: Based on the object relationship data and the object interaction data in the multiple scenes in the application, a multi-scene element interconnection graph is generated, wherein the multi-scene element interconnection graph includes multiple element nodes, the multiple element nodes include object element nodes and auxiliary element nodes, the object element nodes and the relationship between each of the object element nodes are generated by the object relationship data, and the auxiliary element nodes and the relationship between the auxiliary element nodes and the object element nodes are generated by the object interaction data; Based on the interconnected graph of multiple scene elements and the target object, multiple candidate recommended objects for the target object are obtained.

13. The recommendation processing method according to claim 12, characterized in that: The auxiliary element nodes include content author nodes and object interest tag nodes; The acquiring a plurality of candidate recommended objects for the target object based on the multi-scenario element interconnection graph and the target object comprises: Acquire a first candidate recommendation object based on the object element node, the content author node, and the target object; Based on the object interest tag node and the target object, obtaining a second candidate recommended object; The interconnected graph of multiple scene elements and the target object input relationship and interest fusion recommendation model are used to obtain a third candidate recommendation object; The first candidate recommendation object, the second candidate recommendation object, and the third candidate recommendation object are integrated into a plurality of the candidate recommendation objects.

14. The recommendation processing method according to claim 13, characterized in that: The step of sorting the plurality of candidate recommendation objects for display to the target object includes: sorting the plurality of candidate recommendation objects; While displaying the candidate recommendation objects according to the sorting, sources of the candidate recommendation objects are displayed, and the sources include: the process of obtaining a first candidate recommendation object based on the object element node, the content author node, and the target object; the process of obtaining a second candidate recommendation object based on the object interest tag node and the target object; and the process of fusing the multi-scenario element interconnection graph and the target object input relationship with the interest fusion recommendation model to obtain a third candidate recommendation object.

15. The recommendation processing method according to claim 13, characterized in that: The acquiring a first candidate recommendation object based on the object element node, the content author node, and the target object includes: In the multi-scenario element interconnection graph, the object element node having a common associated object element node with the object element node of the target object is obtained, and the object corresponding to the obtained object element node is used as the first sub-candidate recommendation object; In each object element node of the multi-scenario element interconnection graph, the object element node having the same specific attribute as the target object is obtained, and the object corresponding to the obtained object element node is used as the second sub-candidate recommendation object; Based on the attributes of each object element node of the multi-scenario element interconnection graph, group each object element node, obtain the object element nodes that are grouped in the same group as the object element node of the target object, and use the object corresponding to the obtained object element node as the third sub-candidate recommendation object; From the objects corresponding to the object element nodes of the multi-scenario element interconnection graph, an object belonging to the same application session group as the target object is obtained as a fourth sub-candidate recommendation object; In the multi-scenario element interconnection graph, the content author node that has an interactive relationship with the object element node of the target object is obtained, and the object corresponding to the obtained content author node is used as the fifth sub-candidate recommendation object; Based on the list of associated objects of the target object outside the application, obtaining a sixth sub-candidate recommended object; In the multi-scenario element interconnection graph, the content author node that is mutually concerned with the object element node of the target object is obtained, and the object corresponding to the obtained content author node is used as the seventh sub-candidate recommendation object; The first sub-candidate recommendation object, the second sub-candidate recommendation object, the third sub-candidate recommendation object, the fourth sub-candidate recommendation object, the fifth sub-candidate recommendation object, the sixth sub-candidate recommendation object, and the seventh sub-candidate recommendation object are integrated into the first candidate recommendation object.

16. The recommendation processing method according to claim 13, characterized in that: The auxiliary element nodes include content nodes; The acquiring a second candidate recommended object based on the object interest tag node and the target object includes: Based on the label similarity of the object interest label node of the target object and the object interest label nodes of each other object, obtaining an eighth sub-candidate recommended object; Based on the interaction data of the target object node with each of the content nodes in the multi-scenario element interconnection graph, a first point of interest vector of the target object node is obtained; based on the interaction data of each of the other object nodes with each of the content nodes in the multi-scenario element interconnection graph, a second point of interest vector of each of the other object nodes is obtained; and based on a comparison between the first point of interest vector and the second point of interest vector, a ninth sub-candidate recommendation object is obtained; The eighth sub-candidate recommendation object and the ninth sub-candidate recommendation object are integrated into the second candidate recommendation object.

17. A recommendation processing device, characterized in that: include: A first acquisition unit, configured to acquire multiple candidate recommended objects of a target object based on object relationship data and object interaction data in multiple scenes in an application; A second acquisition unit is used to acquire the relationship intimacy between each of the candidate recommendation objects and the target object; A prediction unit, configured to predict a conversion rate after recommending each of the candidate recommended objects to the target object, wherein the conversion rate is a probability that the target object pays attention to the candidate recommended object after the candidate recommended object is recommended to the target object; A third acquisition unit is used to acquire the preference conformity of each candidate recommendation object to the target object; A sorting unit is used to sort the multiple candidate recommendation objects based on the relationship intimacy, the conversion rate, and the preference compliance, so as to display them to the target object.

18. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the recommendation processing method according to any one of claims 1 to 16 is implemented.

19. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the recommendation processing method according to any one of claims 1 to 16 is implemented.

20. A computer program product, comprising a computer program, wherein the computer program is read and executed by a processor of a computer device, so that the computer device executes the recommendation processing method according to any one of claims 1 to 16.

Images