Interest degree prediction and model training method and device, electronic equipment and storage medium

Abstract

The present disclosure provides an interest degree prediction and model training method and device, electronic equipment and storage medium, and relates to the technical fields of artificial intelligence such as big data, intelligent search and machine learning. The specific implementation scheme is: based on a pre-constructed heterogeneous graph, obtaining feature information of a target user and associated feature information of the target user; the heterogeneous graph is constructed based on two types of nodes, namely user information nodes and query statement category nodes; based on the feature information of the target user and the associated feature information of the target user, an interest degree prediction model of a specified field is pre-trained, and the interest degree of the target user for information of the specified field is predicted. The technology of the present disclosure can effectively improve the accuracy of interest degree prediction.

Description

Methods, devices, electronic equipment and storage media for interest prediction and model training Technical Field

[0001] This disclosure relates to the field of computer technology, specifically to the fields of big data, intelligent search, machine learning and other artificial intelligence technologies, and particularly to a method, apparatus, electronic device and storage medium for predicting interest and training models. Background Technology

[0002] In existing technologies, attribute information of multiple users can be collected from multiple channels. Each user is treated as a node, and the social relationships between users are used as edges, enabling different users with social relationships to be connected and constructing a social relationship graph.

[0003] For any given user, based on the constructed social relationship graph, it is possible to identify related users with whom the user has social relationships. Furthermore, based on the characteristic information of these related users, it is possible to predict the user's interest in a particular business area. For example, if user A's related user B frequently participates in activities related to a specific business, it can be predicted that user A will also have a high level of interest in that specific business. Summary of the Invention

[0004] This disclosure provides a method, apparatus, electronic device, and storage medium for interest prediction and model training.

[0005] According to one aspect of this disclosure, an interest prediction method is provided, comprising:

[0006] Based on a pre-constructed heterogeneous graph, the feature information of the target user and the associated feature information of the target user are obtained; the heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes.

[0007] Based on the target user's feature information and the target user's associated feature information, a pre-trained interest prediction model for a specified domain is used to predict the target user's interest in information in the specified domain.

[0008] According to another aspect of this disclosure, a method for training an interest prediction model is provided, comprising:

[0009] Obtain the sample user identifier and the sample user's tag from the log information of the specified domain;

[0010] Based on the sample user identifier and the pre-constructed heterogeneous graph, the feature information of the sample user and the associated feature information of the sample user are obtained; the heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes.

[0011] The interest prediction model for the specified domain is trained based on the feature information of the sample users, the associated feature information of the sample users, and the tags of the sample users.

[0012] According to another aspect of this disclosure, an apparatus for predicting interest is provided, comprising:

[0013] The feature acquisition module is used to acquire the feature information of the target user and the associated feature information of the target user based on a pre-constructed heterogeneous graph; the heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes.

[0014] The prediction module is used to predict the target user's interest in information in the specified domain by using a pre-trained interest prediction model for the specified domain, based on the target user's feature information and the target user's associated feature information.

[0015] According to another aspect of this disclosure, a training apparatus for an interest prediction model is provided, comprising:

[0016] The sample acquisition module is used to obtain the sample user identifier and the tag of the sample user from the log information of the specified domain;

[0017] The feature acquisition module is used to acquire the feature information of the sample user and the associated feature information of the sample user based on the sample user identifier and a pre-constructed heterogeneous graph; the heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes.

[0018] The training module is used to train the interest prediction model for the specified domain based on the feature information of the sample users, the associated feature information of the sample users, and the labels of the sample users.

[0019] According to another aspect of this disclosure, an electronic device is provided, comprising:

[0020] At least one processor; and

[0021] A memory communicatively connected to the at least one processor; wherein,

[0022] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the methods described above and any possible implementations.

[0023] According to yet another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing the computer to perform the methods described above and any possible implementation thereof.

[0024] According to another aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the aspects and any possible implementations described above.

[0025] According to the technology disclosed herein, the accuracy of interest prediction can be effectively improved.

[0026] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0027] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0028] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure;

[0029] Figure 2 is a schematic diagram according to a second embodiment of the present disclosure;

[0030] Figure 3 is a partial schematic diagram of a social relationship graph provided in this disclosure.

[0031] Figure 4 is a partial schematic diagram of a heterogeneous graph provided in this disclosure;

[0032] Figure 5 is a schematic diagram according to a third embodiment of the present disclosure;

[0033] Figure 6 is a schematic diagram according to the fourth embodiment of the present disclosure;

[0034] Figure 7 is a schematic diagram according to the fifth embodiment of the present disclosure;

[0035] Figure 8 is a schematic diagram according to the sixth embodiment of the present disclosure;

[0036] Figure 9 is a schematic diagram according to the seventh embodiment of the present disclosure;

[0037] Figure 10 is a block diagram of an electronic device used to implement the methods of embodiments of the present disclosure. Detailed Implementation

[0038] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0039] Obviously, the described embodiments are only some, not all, of the embodiments disclosed herein. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0040] It should be noted that the terminal devices involved in the embodiments of this disclosure may include, but are not limited to, smart devices such as mobile phones, personal digital assistants (PDAs), wireless handheld devices, and tablet computers; the display devices may include, but are not limited to, personal computers, televisions, and other devices with display functions.

[0041] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0042] Figure 1 is a schematic diagram according to the first embodiment of the present disclosure; as shown in Figure 1, this embodiment provides an interest prediction method, which predicts the interest of any target user in information of a specified domain, and may specifically include the following steps:

[0043] S101. Based on the pre-built heterogeneous graph, obtain the characteristic information of the target user and the associated characteristic information of the target user; the heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes.

[0044] S102. Based on the target user's feature information and related feature information, a pre-trained interest prediction model for the specified domain is used to predict the target user's interest in information in the specified domain.

[0045] The heterogeneous graph in this embodiment is pre-built and includes two types of nodes: one is user information nodes, and the other is query statement (Query) category nodes.

[0046] The user information node can include a lot of basic user information, such as user identifier, gender, age, occupation, education, hobbies, etc. The user identifier can be an identification code configured by the platform for the user, or other identifiers that can uniquely identify the user. All information involved in the heterogeneous graph in this embodiment is obtained with the user's authorization. Furthermore, sensitive user information has been anonymized.

[0047] In this embodiment, the query category node is mainly used to connect different user information nodes. This type of node does not need to carry too much information. For example, it can only carry the name of the query category, or it can also include feature representations such as vectors. In this embodiment, the query category refers to the category to which the user's query belongs during the search, such as very specific categories like credit, insurance, children's education, baby care, Western music, Japanese and Korean movies, etc.

[0048] Given that in existing technologies, users constructing social relationship graphs typically come from different channels, there are very few social connections between different users across different channels. This results in poor connectivity between users in the social relationship graph, often leaving some users without any associated users. Consequently, the accuracy of predicting user interests based on the existing social relationship graph is poor.

[0049] Based on this, this disclosure proposes a heterogeneous graph-based method for predicting user interests. This heterogeneous graph includes not only user information nodes but also query category nodes that are heterogeneous with the user information nodes. In this way, even users from different channels and without social connections can be linked through the query categories they use. For example, user A is from channel 1, and user B is from channel 2; the two users have no social connection, but both used queries related to the "children's education" category. Therefore, user A and user B can be linked through the "children's education" node. Thus, compared to existing social relationship graphs, the heterogeneous graph used in this disclosure has stronger connectivity. Therefore, based on this heterogeneous graph, more comprehensive and accurate association feature information of target users can be obtained; consequently, the interest of target users in information within a specified domain can be predicted more efficiently and accurately.

[0050] The interest prediction method in this embodiment can obtain the target user's feature information and related feature information based on a pre-constructed heterogeneous node graph that includes two types of nodes: user information nodes and query statement category nodes. This makes the obtained related feature information of the target user more comprehensive and accurate. Furthermore, based on the target user's feature information and related feature information, a pre-trained interest prediction model for a specified domain is used to predict the target user's interest in information in the specified domain, which can effectively improve the accuracy of the predicted interest.

[0051] Optionally, in one embodiment of this disclosure, the method may further include the following step before step S101 of the embodiment shown in FIG1:

[0052] Based on the collected user information and the query statements used by each user, a heterogeneous graph is constructed.

[0053] Specifically, multiple user information and all queries used by each user can be collected from the logs. Optionally, to ensure the timeliness of the information, the above information can be collected only from logs containing records from a certain preset time period prior to the current moment. The preset time period can be set according to actual needs, such as a week, a month, a quarter, or other time periods, and is not limited here.

[0054] Figure 2 is a schematic diagram according to the second embodiment of the present disclosure; as shown in Figure 2, this embodiment provides a method for constructing a heterogeneous graph, specifically used to construct a heterogeneous graph based on multiple collected user information and the query statements used by the users corresponding to each user information, which may specifically include the following steps:

[0055] S201. Collect information on multiple users and the query statements used by each user from the log information;

[0056] In this embodiment, each piece of user information collected may include user ID, gender, age, occupation, education level, hobbies, etc. Simultaneously, all queries used by each user can be utilized. This embodiment is used to construct a heterogeneous graph applicable to various fields, thus allowing the collection of multiple user information and their query statements from log information across different fields.

[0057] S202. For the query statements used by each user, a pre-trained feature representation model is used to obtain the feature representation of the query statement.

[0058] The feature representation of the query statement in this embodiment can be expressed in vector form. The feature representation model in this embodiment can be the FastText model.

[0059] Suppose that the query statement can include K categories across all domains, and the specific number of K varies depending on the size of the classification granularity.

[0060] Before using the fasttext model, it can be trained using K categories of labeled query statements. This ensures that the similarity of feature representations (vectors) of query statements within the same category reaches a preset similarity threshold. The preset similarity threshold can be set according to actual needs. After this training, the fasttext model can generate feature representations for query statements of unknown categories.

[0061] S203. Identify the category of the query statement based on the feature expressions of other query statements of known categories and the feature expressions of the query statement itself.

[0062] Specifically, the similarity between the feature representation of the query statement and the feature representation of other query statements of known categories can be calculated, and the category of the query statement can be identified as the category of the other query statements with the highest similarity.

[0063] S204. Based on multiple user information and the query statement category corresponding to each user information, each user information and each query statement category are respectively used as nodes; two different user information nodes with the same query statement category are connected through the same query statement category node to obtain a heterogeneous graph based on user information and query statement category.

[0064] S205. In a heterogeneous graph, based on the social relationships between different user information in multiple user information, connect two user information nodes that have social relationships.

[0065] The social relationships in this embodiment can include likes, shares, and follows.

[0066] For example, Figure 3 is a partial schematic diagram of a social relationship graph provided in this disclosure. As shown in Figure 3, users corresponding to ID7, ID1, and ID9 come from the first channel; users corresponding to ID2, ID3, and ID10 come from the second channel; and users corresponding to ID17, ID5, and ID8 come from the third channel. In this social relationship, the user with ID2 and the user with ID10 have a follow relationship; the user with ID10 and the user with ID3 have a comment relationship; the user with ID1 and the user with ID7 have a follow relationship; the user with ID7 and the user with ID8 have a like relationship; the user with ID17 and the user with ID5 have a follow relationship; and the user with ID17 and the user with ID8 have a share relationship.

[0067] According to the existing technical solutions, there are no social relationships between users of the three channels, so the connectivity between users in the social relationship graph is very weak.

[0068] Figure 4 is a partial schematic diagram of a heterogeneous graph provided in this disclosure. Specifically, Figure 4 is a heterogeneous graph constructed based on the above steps S201-S204 of this embodiment.

[0069] As shown in Figure 4, user identifiers UID1 and UID2 are associated through Query category 1. User identifier UID1 is associated with user identifiers UID3 and UID4 respectively through Query category 2. Compared with Figure 3, the heterogeneous graph of this disclosure adds nodes of the Query category, enhancing the connectivity between users.

[0070] In this embodiment, by employing the above method, a highly accurate and efficient heterogeneous graph can be constructed. Moreover, compared to existing social relationship graphs, the heterogeneous graph constructed in this embodiment has excellent connectivity, effectively connecting user information that has no social relationship.

[0071] Furthermore, compared to existing social relationship graphs, query characteristics were previously used as attributes on user information nodes. However, by extracting query categories as a new type of node in heterogeneous graphs, a large number of features on user information nodes are released, effectively expanding the capacity limit of the graph structure.

[0072] Figure 5 is a schematic diagram according to the third embodiment of this disclosure; as shown in Figure 5, this embodiment further describes the technical solution of this disclosure in more detail based on the embodiment shown in Figure 1 above. As shown in Figure 5, an interest prediction method of this embodiment may specifically include the following steps:

[0073] S501. Obtain the feature information of the target user from the nodes corresponding to the target user in the heterogeneous graph;

[0074] S502. Obtain the association feature information of the target user from at least two-order neighbor nodes of the heterogeneous graph that are associated with the node corresponding to the target user;

[0075] The heterogeneous graph in this embodiment can adopt the structure shown in Figure 4. The heterogeneous graph includes two types of nodes: user information nodes and query statement category nodes.

[0076] In this embodiment, the target user can be any user information corresponding to any user information in the heterogeneous graph. Specifically, to predict a target user's interest in information from a specific domain, the identifier of the target user can be obtained first. Then, the characteristic information of the target user can be obtained from the attribute information stored in the node corresponding to the target user in the heterogeneous graph. For example, this may include the target user's gender, age, occupation, education level, hobbies, frequently visited websites, frequently used software categories, etc.

[0077] In a heterogeneous graph, the neighboring nodes directly connected to the target user are the first-order nodes corresponding to the target user; the neighboring nodes directly connected to the first-order nodes of the target user are the second-order nodes; and so on, third-order, fourth-order, and other multi-order nodes of the target user can be obtained. The node corresponding to the target user can be considered as the zeroth-order node of the target user. Specifically, the characteristic information of the target user's associated users can be obtained from at least two-order neighboring nodes, serving as the target user's associated characteristic information. However, in practical applications, the information of higher-order neighboring nodes that are far from the target user's information is not very meaningful. Therefore, it is generally preferable to obtain the target user's associated characteristic information from the two-order neighboring nodes. Of course, in practical applications, if the information from the two-order neighboring nodes is limited, the target user's associated characteristic information can also be obtained from the third-order or fourth-order neighboring nodes.

[0078] For example, to obtain the association feature information of the target user from the two-level neighbor nodes corresponding to the target user, the specific implementation may include the following steps:

[0079] (1) Obtain multiple first-order nodes that are the nearest neighbors of the node corresponding to the target user from the heterogeneous graph;

[0080] (2) From the heterogeneous graph, obtain the second-order node that is the nearest neighbor of each first-order node, and obtain a total of multiple second-order nodes;

[0081] (3) Obtain the feature information of the target user’s associated users from multiple first-order nodes and multiple second-order nodes respectively, and use it as the associated feature information of the target user.

[0082] Specifically, multiple first-order nodes may include some query statement category nodes and some user information nodes. The query statement category nodes only serve a connecting function and do not contain any user information. Therefore, in the implementation, query statement category nodes can be removed from the multiple first-order and multiple second-order nodes. Then, from the remaining nodes, the characteristic information of the target user's associated users can be obtained as the target user's associated characteristic information. If there are multiple remaining nodes, the characteristic information of the associated users in each remaining node needs to be obtained to collectively constitute the target user's associated characteristic information.

[0083] By using the above method, multiple first-order nodes and multiple second-order nodes corresponding to the target user can be accurately obtained, thereby accurately obtaining the associated feature information of the target user.

[0084] In one embodiment of this disclosure, if there are a large number of query statement categories corresponding to the target user, and / or a large number of users with social relationships with the target user, obtaining all the first-order nodes that are the nearest neighbors of the node corresponding to the target user will result in a large number of first-order nodes. In this case, according to a preset filtering strategy, some of the first-order nodes that are the nearest neighbors of the node corresponding to the target user can be obtained from the heterogeneous graph to obtain multiple first-order nodes.

[0085] Similarly, following the preset filtering strategy, the second-order nodes that are nearest neighbors to each first-order node are obtained from the heterogeneous graph, and some second-order nodes are filtered to obtain multiple second-order nodes corresponding to the target user.

[0086] For example, the preset filtering strategy in this embodiment may include: obtaining multiple first-order nodes from the heterogeneous graph, and obtaining multiple second-order nodes from the heterogeneous graph. The number of the multiple first-order nodes is not greater than a first preset threshold, and the number of the multiple second-order nodes is not greater than a second preset threshold; or the sum of the number of the multiple first-order nodes and the number of the multiple second-order nodes is not greater than a third preset threshold. The first, second, and third preset thresholds are all positive integers, and their specific values ​​can be set according to actual needs.

[0087] For example, in one embodiment of this disclosure, multiple first-order nodes can be randomly obtained from the heterogeneous graph, and multiple second-order nodes can be randomly obtained from the heterogeneous graph.

[0088] Alternatively, multiple first-order nodes can be obtained from the heterogeneous graph according to certain rules, and the multiple second-order nodes can also be obtained from the heterogeneous graph according to the same rules. For example, the rules here may include obtaining nodes based on the probability of each node being selected in the historical record, and each time a node is obtained, nodes with a low probability of being selected in the historical record are selected first, thereby making the probability of all nodes being selected more even.

[0089] Alternatively, multiple first-order nodes can be randomly obtained from the heterogeneous graph according to a first preset ratio, and multiple second-order nodes can be randomly obtained from the heterogeneous graph according to a second preset ratio; wherein the first preset ratio and the second preset ratio can be the same or different, and the specific values ​​are set according to actual needs.

[0090] Of course, in practical applications, other preset filtering strategies can also be used to obtain some first-order nodes and some second-order nodes of the target user, resulting in multiple first-order nodes and multiple second-order nodes. Examples will not be elaborated here.

[0091] By adopting the above method, the acquisition efficiency of multiple first-order nodes and multiple second-order nodes can be effectively improved, thereby improving the acquisition efficiency of the target user's associated feature information.

[0092] S503. Based on the target user's feature information and the target user's associated feature information, a pre-trained interest prediction model for the specified domain is used to predict the target user's interest in information in the specified domain.

[0093] Specifically, the target user's feature information and related feature information are input into the interest prediction model. This model, based on the input information, performs embedding representation and then computational processing to predict and output the target user's interest level in information within a specified domain. The interest level is a probability value between 0 and 1. A higher probability value indicates greater interest in the specified domain, while a lower probability value indicates less interest.

[0094] S504. If the target user's interest in information in the specified field is greater than or equal to a preset interest threshold, push information in the specified field to the target user.

[0095] In this embodiment, the target user's interest in information in a specified field can be applied to intelligent search and recommendation scenarios, which can effectively improve the accuracy and efficiency of information push.

[0096] The interest prediction method in this embodiment, by adopting the above scheme and based on heterogeneous graphs, can obtain very accurate association feature information of target users, and thus can efficiently and accurately predict the target user's interest in information in a specified field.

[0097] Figure 6 is a schematic diagram according to the fourth embodiment of this disclosure; as shown in Figure 6, this embodiment provides a training method for an interest prediction model, which may specifically include the following steps:

[0098] S601. Obtain the sample user identifier and the sample user's tag from the log information of the specified domain;

[0099] S602. Based on the sample user identifier and the pre-constructed heterogeneous graph, obtain the feature information of the sample user and the associated feature information of the sample user;

[0100] S603. Based on the feature information of sample users, the associated feature information of sample users, and the labels of sample users, train an interest prediction model for a specified domain.

[0101] Specifically, in any scenario, log information can record the display of all information, user click behavior, and so on. For example, log information can record the push of information in a specific field to users. It can also record whether the user saw the pushed information and chose to click and view it. If the log information records that the user chose to click and view, then the user is considered interested in the information in that specific field, and this user is a positive sample user; conversely, if the log information does not record that the user chose to click and view, then the user is considered not interested in the information in that specific field, and this user is a negative sample user.

[0102] Based on this, the identifiers and tags of sample users used to train the interest prediction model for a specified domain can be obtained from the log information of that domain. These tags indicate whether the sample user is interested in information within the specified domain. For example, 1 can represent interest, and 0 can represent no interest.

[0103] Further, based on the sample user identifier, the feature information and associated feature information of the sample user are obtained from the pre-constructed heterogeneous graph. This heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes. For a detailed explanation of the implementation of step S602, please refer to the specific implementation of step S101 in the embodiment shown in Figure 1 above; it will not be repeated here.

[0104] During training, the feature information and associated feature information of sample users are input into a domain-specific interest prediction model. This model predicts and outputs the sample user's interest level in information within the specified domain based on the input information. This interest level is a probability value between 0 and 1. Then, based on the predicted interest level and the sample user's label, the parameters of the interest prediction model are adjusted so that the prediction result closely matches the result identified by the sample user's label. For example, if a sample user's label is 1, identifying them as a positive user, the predicted probability of their interest in the specified domain is considered to be 1. If a sample user's label is 0, identifying them as a negative user, the predicted probability of their interest in the specified domain is considered to be 0.

[0105] In this embodiment, the interest prediction model can be trained using information from multiple sample users through multiple rounds of training, as described above. Training continues until a preset threshold is reached, or until the model's predicted interest in a specific domain consistently matches the sample user's label across multiple training rounds. At this point, training ends, the parameters of the interest prediction model are determined, and the model itself is finalized.

[0106] The training method of the interest prediction model in this embodiment can accurately obtain the associated feature information of sample users by adopting the above-mentioned approach. Then, based on the feature information, associated feature information, and labels of sample users, the interest prediction model for a specified domain can be trained accurately and efficiently, which can effectively improve the training efficiency and accuracy of the interest prediction model.

[0107] In one embodiment of this disclosure, step S601 in the above embodiments may include the following steps:

[0108] (a1) Obtain the identifiers of positive sample users with operational behaviors from the log information of the specified domain; and configure the tags of positive sample users; and

[0109] (b1) Obtain the negative sample user identifiers without any operational behavior from the log information of the specified domain; and configure the labels of the negative sample users.

[0110] It should be noted that the actions described here are all predicated on a display action. If the user takes action after the display, they are considered a positive sample user; otherwise, they are considered a negative sample user. The label for a positive sample user can be 1, indicating a probability of 1 for the positive sample user's interest in information within the specified domain. The label for a negative sample user can be 0, indicating a probability of 0 for the negative sample user's interest in information within the specified domain.

[0111] In one embodiment of this disclosure, step S602 in the above embodiments may include the following steps:

[0112] (a2) Obtain the feature information of the sample users from the nodes corresponding to the sample users in the heterogeneous graph;

[0113] (b2) Obtain the association feature information of the sample user from the neighboring nodes of at least two orders of magnitude of the heterogeneous graph that are associated with the node corresponding to the sample user.

[0114] Specifically, refer to the specific implementation of steps S501 and S502 in the embodiment shown in Figure 5 above.

[0115] For example, to obtain the association feature information of a sample user from its two-level neighbor nodes, the specific implementation may include the following steps:

[0116] (a3) From the heterogeneous graph, obtain multiple first-order nodes that are the nearest neighbors of the node corresponding to the sample user;

[0117] (b3) From the heterogeneous graph, obtain the second-order node that is the nearest neighbor of each first-order node, and obtain a total of multiple second-order nodes;

[0118] (b3) Obtain the feature information of the associated users of the sample user from multiple first-order nodes and multiple second-order nodes respectively, and use it as the associated feature information of the sample user.

[0119] By using the above method, multiple first-order nodes and multiple second-order nodes corresponding to the sample user can be accurately obtained, thereby accurately obtaining the associated feature information of the sample user.

[0120] In one embodiment of this disclosure, if there are a large number of query statement categories corresponding to the sample user, and / or a large number of users with social relationships with the sample user, obtaining all the nearest first-order nodes of the node corresponding to the sample user would result in a large number of first-order nodes. In this case, according to a preset filtering strategy, some of the nearest first-order nodes of the node corresponding to the sample user can be obtained from the heterogeneous graph to obtain multiple first-order nodes.

[0121] Similarly, following the preset filtering strategy, the second-order nodes that are nearest neighbors to each first-order node are obtained from the heterogeneous graph, and some second-order nodes are filtered to obtain multiple second-order nodes corresponding to the sample users.

[0122] In this embodiment, the preset filtering strategy can be referred to the relevant description of the embodiment shown in Figure 5 above, and will not be repeated here.

[0123] By adopting the above scheme, based on heterogeneous graphs, the associated feature information of target users can be obtained very accurately and efficiently, thereby enabling efficient and accurate training of interest prediction models.

[0124] Figure 7 is a schematic diagram according to the fifth embodiment of the present disclosure; as shown in Figure 7, this embodiment provides an interest prediction device 700, including:

[0125] The feature acquisition module 701 is used to acquire the feature information of the target user and the associated feature information of the target user based on a pre-constructed heterogeneous graph; the heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes.

[0126] The prediction module 702 is used to predict the target user's interest in information in the specified domain based on the target user's feature information and the target user's associated feature information, using a pre-trained interest prediction model for the specified domain.

[0127] The interest prediction device 700 in this embodiment achieves the same implementation principle and technical effect as the above-mentioned related method embodiments by using the above-mentioned modules. For details, please refer to the description of the above-mentioned related method embodiments, which will not be repeated here.

[0128] Figure 8 is a schematic diagram according to the sixth embodiment of the present disclosure; as shown in Figure 8, this embodiment provides an interest prediction device 800, including: a feature acquisition module 801 and a prediction module 802, which adopt the same name and function as the embodiment shown in Figure 7 above.

[0129] As shown in Figure 8, in this embodiment, the feature acquisition module 801 includes:

[0130] The first acquisition unit 8011 is used to acquire feature information of the target user from the node corresponding to the target user in the heterogeneous graph;

[0131] The second acquisition unit 8012 is used to acquire the association feature information of the target user from at least two levels of neighboring nodes of the heterogeneous graph that are associated with the node corresponding to the target user.

[0132] Furthermore, in one embodiment of this disclosure, the second acquisition unit 8012 is used for:

[0133] From the heterogeneous graph, obtain multiple first-order nodes that are the nearest neighbors of the node corresponding to the target user;

[0134] From the heterogeneous graph, the second-order nodes that are nearest neighbors to each first-order node are obtained, resulting in a total of multiple second-order nodes;

[0135] The feature information of the associated users of the target user is obtained from the plurality of first-order nodes and the plurality of second-order nodes respectively, and is used as the associated feature information of the target user.

[0136] Furthermore, in one embodiment of this disclosure, the second acquisition unit 8012 is used for:

[0137] According to the preset filtering strategy, the plurality of first-order nodes that are the nearest neighbors of the node corresponding to the target user are obtained from the heterogeneous graph;

[0138] From the heterogeneous graph, the second-order nodes that are nearest neighbors to each of the first-order nodes are obtained, resulting in a total of multiple second-order nodes, including:

[0139] According to the preset filtering strategy, the second-order nodes that are nearest neighbors to each first-order node are obtained from the heterogeneous graph, and a total of the multiple second-order nodes are obtained.

[0140] Furthermore, in one embodiment of this disclosure, the preset filtering strategy includes:

[0141] Obtain the plurality of first-order nodes from the heterogeneous graph, and obtain the plurality of second-order nodes from the heterogeneous graph; wherein, the number of the plurality of first-order nodes is not greater than a first preset threshold, and the number of the plurality of second-order nodes is not greater than a second preset threshold; or the sum of the number of the plurality of first-order nodes and the number of the plurality of second-order nodes is not greater than a third preset threshold; or

[0142] The plurality of first-order nodes are randomly obtained from the heterogeneous graph according to a first preset ratio, and the plurality of second-order nodes are randomly obtained from the heterogeneous graph according to a second preset ratio.

[0143] Furthermore, in one embodiment of this disclosure, the second acquisition unit 8012 is used for:

[0144] Remove nodes belonging to the query statement category from the plurality of first-order nodes and the plurality of second-order nodes;

[0145] From the remaining nodes after removal, obtain the feature information of the associated users of the target user, and use it as the associated feature information of the target user.

[0146] Furthermore, as shown in FIG8, in one embodiment of this disclosure, the interest prediction device 800 further includes:

[0147] The construction module 803 is used to construct the heterogeneous graph based on the collected user information and the query statements used by the users corresponding to each user information.

[0148] Furthermore, in one embodiment of this disclosure, the construction module 803 is configured to:

[0149] For each query statement used by the user, a pre-trained feature representation model is used to obtain the feature representation of the query statement;

[0150] The category of the query statement is identified based on the feature expressions of other query statements of known categories and the feature expressions of the query statement itself.

[0151] Based on the multiple user information and the query statement category corresponding to each user information, each user information and each query statement category are respectively used as nodes; two different user information nodes with the same query statement category are connected through the same query statement category node to obtain a heterogeneous graph based on user information and query statement category;

[0152] In the heterogeneous graph, based on the social relationships between different user information in the multiple user information, two user information nodes with social relationships are connected.

[0153] The interest prediction device 800 in this embodiment achieves the same implementation principle and technical effect as the above-mentioned related method embodiments by using the above-mentioned modules. For details, please refer to the description of the above-mentioned related method embodiments, which will not be repeated here.

[0154] Figure 9 is a schematic diagram according to the seventh embodiment of the present disclosure; as shown in Figure 9, this embodiment provides a training device 900 for an interest prediction model, including:

[0155] The sample acquisition module 901 is used to acquire the sample user identifier and the tag of the sample user from the log information of the specified domain;

[0156] The feature acquisition module 902 is used to acquire the feature information of the sample user and the associated feature information of the sample user based on the sample user identifier and a pre-constructed heterogeneous graph; the heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes.

[0157] Training module 903 is used to train the interest prediction model for the specified domain based on the feature information of the sample users, the associated feature information of the sample users, and the labels of the sample users.

[0158] The training device 900 for the interest prediction model in this embodiment achieves the same implementation principle and technical effect as the above-mentioned related method embodiments by using the above-mentioned modules. For details, please refer to the description of the above-mentioned related method embodiments, which will not be repeated here.

[0159] Furthermore, in one embodiment of this disclosure, the sample acquisition module 901 is used for:

[0160] From the log information of the specified domain, obtain the identifiers of positive sample users with operational behaviors; and configure the tags of the positive sample users; and

[0161] Obtain the negative sample user identifiers that have no operational behavior from the log information of the specified domain; and configure the labels of the negative sample users.

[0162] Furthermore, in one embodiment of this disclosure, the feature acquisition module 902 is used for:

[0163] The feature information of the sample user is obtained from the nodes corresponding to the sample user in the heterogeneous graph;

[0164] The association feature information of the sample user is obtained from at least two levels of neighboring nodes of the heterogeneous graph that are associated with the node corresponding to the sample user.

[0165] The training device 900 for the interest prediction model in the above embodiments achieves the same implementation principle and technical effect as the above-mentioned related method embodiments by using the above-mentioned modules. For details, please refer to the description of the above-mentioned related method embodiments, which will not be repeated here.

[0166] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0167] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0168] Figure 10 illustrates a schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0169] As shown in Figure 10, device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1002 or a computer program loaded from storage unit 1008 into random access memory (RAM) 1003. The RAM 1003 may also store various programs and data required for the operation of device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. Input / output (I / O) interface 1005 is also connected to bus 1004.

[0170] Multiple components in device 1000 are connected to I / O interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1007, such as various types of monitors, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0171] The computing unit 1001 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as the methods of this disclosure. For example, in some embodiments, the methods of this disclosure can be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program can be loaded and / or installed on device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of the methods of this disclosure described above can be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to perform the methods described above by any other suitable means (e.g., by means of firmware).

[0172] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0173] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0174] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0175] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0176] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0177] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0178] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0179] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An interest prediction method, comprising: Based on a pre-constructed heterogeneous graph, the feature information of the target user and the associated feature information of the target user are obtained; The heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes. The heterogeneous graph includes multiple user information nodes and query statement category nodes corresponding to the query statements used by the users of each user information. Two different user information nodes with the same query statement category in the heterogeneous graph are connected through the same query statement category node. Two different user information nodes with social relationships are connected; the user information node includes the user's identifier and the user's feature information; based on the target user's feature information and the target user's associated feature information, a pre-trained interest prediction model for a specified domain is used to predict the target user's interest in information in the specified domain.

2. The method according to claim 1, wherein, Based on a pre-constructed heterogeneous graph, the feature information of the target user and the associated feature information of the target user are obtained, including: obtaining the feature information of the target user from the nodes corresponding to the target user in the heterogeneous graph; and obtaining the associated feature information of the target user from at least two-order neighbor nodes in the heterogeneous graph that have an association relationship with the node corresponding to the target user.

3. The method according to claim 2, wherein, Obtaining the association feature information of the target user from at least two levels of neighboring nodes in the heterogeneous graph that are associated with the node corresponding to the target user includes: obtaining multiple first-order nodes that are nearest neighbors to the node corresponding to the target user from the heterogeneous graph; obtaining multiple second-order nodes that are nearest neighbors to each of the first-order nodes from the heterogeneous graph; and obtaining the feature information of the associated users of the target user from the multiple first-order nodes and the multiple second-order nodes, respectively, as the association feature information of the target user.

4. The method according to claim 3, wherein, Obtaining multiple first-order nodes that are nearest neighbors to the node corresponding to the target user from the heterogeneous graph includes: obtaining the multiple first-order nodes that are nearest neighbors to the node corresponding to the target user from the heterogeneous graph according to a preset filtering strategy; obtaining multiple second-order nodes that are nearest neighbors to each of the first-order nodes from the heterogeneous graph, thereby obtaining multiple second-order nodes in total, includes: obtaining the multiple second-order nodes that are nearest neighbors to each of the first-order nodes from the heterogeneous graph according to the preset filtering strategy.

5. The method according to claim 4, wherein, The preset filtering strategy includes: obtaining the plurality of first-order nodes from the heterogeneous graph, and obtaining the plurality of second-order nodes from the heterogeneous graph; wherein the number of the plurality of first-order nodes is not greater than a first preset threshold, and the number of the plurality of second-order nodes is not greater than a second preset threshold; or the sum of the number of the plurality of first-order nodes and the number of the plurality of second-order nodes is not greater than a third preset threshold; or randomly obtaining the plurality of first-order nodes from the heterogeneous graph according to a first preset ratio, and randomly obtaining the plurality of second-order nodes from the heterogeneous graph according to a second preset ratio.

6. The method according to claim 3, wherein, Obtaining the characteristic information of the associated users of the target user from the plurality of first-order nodes and the plurality of second-order nodes as the associated characteristic information of the target user includes: removing nodes of the query statement category from the plurality of first-order nodes and the plurality of second-order nodes; and obtaining the characteristic information of the associated users of the target user from the remaining nodes after removal as the associated characteristic information of the target user.

7. The method according to any one of claims 1-6, wherein, Before obtaining the feature information of the target user and the associated feature information of the target user based on the pre-constructed heterogeneous graph, the method further includes: constructing the heterogeneous graph based on the collected multiple user information and the query statements used by the users corresponding to each user information.

8. The method according to claim 7, wherein, Based on the collected user information and the query statements used by the users corresponding to each user information, the heterogeneous graph is constructed, including: for each query statement used by the users, using a pre-trained feature representation model to obtain the feature representation of the query statement; identifying the category of the query statement based on the feature representation of other query statements of known categories and the feature representation of the query statement; based on the multiple user information and the query statement category corresponding to each user information, each user information and each query statement category are respectively used as nodes; two different user information nodes with the same query statement category are connected through the same query statement category node to obtain a heterogeneous graph based on user information and query statement category; in the heterogeneous graph, based on the social relationships between different user information in the multiple user information, two user information nodes with social relationships are connected.

9. A training method for an interest prediction model, comprising: Obtain the sample user identifier and the sample user's tag from the log information of the specified domain; Based on the sample user identifier and the pre-constructed heterogeneous graph, the feature information of the sample user and the associated feature information of the sample user are obtained; the heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes. The heterogeneous graph includes multiple user information nodes and query statement category nodes corresponding to the query statements used by the users of each user information; two different user information nodes with the same query statement category in the heterogeneous graph are connected through the same query statement category node. Two different user information nodes with social relationships are connected; the user information node includes the user's identifier and the user's characteristic information; The interest prediction model for the specified domain is trained based on the feature information of the sample users, the associated feature information of the sample users, and the tags of the sample users.

10. The method according to claim 9, wherein, Obtaining sample user identifiers and tags from log information in a specified domain includes: obtaining positive sample user identifiers with operational behavior from the log information in the specified domain; and configuring tags for the positive sample users; and obtaining negative sample user identifiers without operational behavior from the log information in the specified domain; and configuring tags for the negative sample users.

11. The method according to claim 9, wherein, Based on the sample user identifier and the pre-constructed heterogeneous graph, the feature information of the sample user and the associated feature information of the sample user are obtained, including: obtaining the feature information of the sample user from the nodes corresponding to the sample user in the heterogeneous graph; and obtaining the associated feature information of the sample user from at least two-order neighbor nodes in the heterogeneous graph that have an association relationship with the node corresponding to the sample user.

12. An interest prediction device, comprising: The feature acquisition module is used to acquire the feature information of the target user and the associated feature information of the target user based on a pre-constructed heterogeneous graph; The heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes. The heterogeneous graph includes multiple user information nodes and query statement category nodes corresponding to the query statements used by the users of each user information node. Two different user information nodes with the same query statement category in the heterogeneous graph are connected through the same query statement category node. Two different user information nodes with social relationships are connected; the user information node includes the user's identifier and the user's characteristic information; The prediction module is used to predict the target user's interest in information in the specified domain by using a pre-trained interest prediction model for the specified domain, based on the target user's feature information and the target user's associated feature information.

13. The apparatus according to claim 12, wherein, The feature acquisition module includes: a first acquisition unit, configured to acquire feature information of the target user from the nodes corresponding to the target user in the heterogeneous graph; and a second acquisition unit, configured to acquire associated feature information of the target user from at least two-order neighbor nodes of the heterogeneous graph that have an association relationship with the node corresponding to the target user.

14. The apparatus according to claim 13, wherein, The second acquisition unit is configured to: acquire, from the heterogeneous graph, a plurality of first-order nodes that are the nearest neighbors of the node corresponding to the target user; From the heterogeneous graph, the second-order nodes that are nearest neighbors to each first-order node are obtained, resulting in a total of multiple second-order nodes; The feature information of the associated users of the target user is obtained from the plurality of first-order nodes and the plurality of second-order nodes respectively, and is used as the associated feature information of the target user.

15. The apparatus according to claim 14, wherein, The second acquisition unit is configured to: acquire, according to a preset filtering strategy, the plurality of first-order nodes that are the nearest neighbors of the node corresponding to the target user from the heterogeneous graph; From the heterogeneous graph, the nearest second-order node to each first-order node is obtained, resulting in a plurality of second-order nodes. This includes: according to the preset filtering strategy, the nearest second-order node to each first-order node is obtained from the heterogeneous graph, resulting in a plurality of second-order nodes.

16. The apparatus according to claim 15, wherein, The preset filtering strategy includes: obtaining the plurality of first-order nodes from the heterogeneous graph, and obtaining the plurality of second-order nodes from the heterogeneous graph; wherein the number of the plurality of first-order nodes is not greater than a first preset threshold, and the number of the plurality of second-order nodes is not greater than a second preset threshold; or the sum of the number of the plurality of first-order nodes and the number of the plurality of second-order nodes is not greater than a third preset threshold; or randomly obtaining the plurality of first-order nodes from the heterogeneous graph according to a first preset ratio, and randomly obtaining the plurality of second-order nodes from the heterogeneous graph according to a second preset ratio.

17. The apparatus according to claim 14, wherein, The second acquisition unit is used to: remove nodes of the query statement category from the plurality of first-order nodes and the plurality of second-order nodes; From the remaining nodes after removal, obtain the feature information of the associated users of the target user, and use it as the associated feature information of the target user.

18. The apparatus according to any one of claims 12-17, wherein, The device further includes a construction module, used to construct the heterogeneous graph based on the collected user information and the query statements used by the users corresponding to each user information.

19. The apparatus according to claim 18, wherein, The construction module is configured to: for each query statement used by the user, use a pre-trained feature representation model to obtain the feature representation of the query statement; and identify the category of the query statement based on the feature representations of other query statements of known categories and the feature representation of the query statement. Based on the multiple user information and the query statement category corresponding to each user information, each user information and each query statement category are respectively used as nodes; Two different user information nodes with the same query statement category are connected through nodes of the same query statement category to obtain a heterogeneous graph based on user information and query statement category; In the heterogeneous graph, based on the social relationships between different user information in the multiple user information, two user information nodes with social relationships are connected.

20. A training device for an interest prediction model, comprising: The sample acquisition module is used to obtain the sample user identifier and the tag of the sample user from the log information of the specified domain; The feature acquisition module is used to acquire the feature information of the sample user and the associated feature information of the sample user based on the sample user identifier and a pre-constructed heterogeneous graph. The heterogeneous graph is constructed based on two types of nodes: user information nodes and query statement category nodes. The heterogeneous graph includes multiple user information nodes and query statement category nodes corresponding to the query statements used by the users of each user information. Two different user information nodes with the same query statement category in the heterogeneous graph are connected through the same query statement category node. Two different user information nodes with social relationships are connected; the user information node includes the user's identifier and the user's characteristic information; The training module is used to train the interest prediction model for the specified domain based on the feature information of the sample users, the associated feature information of the sample users, and the labels of the sample users.

21. The apparatus according to claim 20, wherein, The sample acquisition module is used to: obtain positive sample user identifiers with operational behaviors from the log information in the specified domain; and configure the tags of the positive sample users; And obtain negative sample user identifiers with no operational behavior from the log information in the specified domain; And configure the labels for the negative sample users.

22. The apparatus according to claim 20, wherein, The feature acquisition module is used to: acquire feature information of the sample user from the node corresponding to the sample user in the heterogeneous graph; and acquire associated feature information of the sample user from at least two neighboring nodes of the heterogeneous graph that have an association relationship with the node corresponding to the sample user.

23. An electronic device, comprising: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8 or 9-11.

24. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-8 or 9-11.

25. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-8 or 9-11.

Images