Information recommendation method and device and electronic equipment

Abstract

The invention discloses an information recommendation method and device and electronic equipment. The method comprises the steps that a plurality of second-degree relation pairs, recalled through a plurality of recall ways, of target users are obtained, and the second-degree relation pairs comprise users having second-degree relations with the target users; performing duplicate removal processing on the plurality of second-degree relation pairs to obtain a plurality of target second-degree relation pairs; determining a target confidence coefficient and a target recall path of the target two-degree relation pair according to the confidence coefficient of at least one recall path corresponding to the target two-degree relation pair; and performing information recommendation on the target user based on the target confidence and the target recall path of the plurality of target two-degree relation pairs. According to the method and the device, the efficiency and the accuracy of information spreading are improved, and more personalized and accurate information recommendation service is provided for the target user.

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and in particular, relates to an information recommendation method, device and electronic equipment. Background Art

[0002] With the popularization of the Internet and the development of social networks, people are increasingly relying on social networks for information exchange and dissemination. Traditional social relationship mining usually only focuses on the user's first-degree relationship, that is, the relationship between the user and his direct friends. For example, in the social push scenario of Weibo, the content that the user is interested in is usually pushed to the user based on the user's first-degree relationship, for example, the materials of the blogger followed by the user are pushed to the user based on the follow relationship between the user and the blogger.

[0003] However, the first-degree relationships of some users are not rich. For example, some users follow fewer bloggers, so the content that can be pushed to these users is relatively small, which reduces the efficiency and accuracy of information dissemination. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide an information recommendation method, device, electronic device and storage medium to solve the problem that traditional social relationship mining only focuses on the first-degree relationships of users, reducing the efficiency and accuracy of information dissemination.

[0005] To achieve the above objectives, the present application embodiment adopts the following technical solutions: In a first aspect, an embodiment of the present application provides an information recommendation method, comprising: obtaining multiple second-degree relationship pairs of target users recalled through multiple recall paths, the second-degree relationship pairs including users having a second-degree relationship with the target user; performing deduplication processing on the multiple second-degree relationship pairs to obtain multiple target second-degree relationship pairs; determining a target confidence and a target recall path of the target second-degree relationship pairs according to the confidence of at least one recall path corresponding to the target second-degree relationship pairs; and recommending information to the target user based on the target confidence and the target recall path of the multiple target second-degree relationship pairs.

[0006] In a second aspect, an embodiment of the present application provides an information recommendation device, comprising: an acquisition module, used to acquire multiple second-degree relationship pairs of target users recalled through multiple recall paths, wherein the second-degree relationship pairs include users having a second-degree relationship with the target users; a deduplication module, used to perform deduplication processing on the multiple second-degree relationship pairs to obtain multiple target second-degree relationship pairs; a determination module, used to determine a target confidence and a target recall path of the target second-degree relationship pairs according to the confidence of at least one recall path corresponding to the target second-degree relationship pairs; and a recommendation module, used to recommend information to the target users based on the target confidence and the target recall path of the multiple target second-degree relationship pairs.

[0007] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the embodiment of the first aspect of the present application, or implements the steps of the method described in the embodiment of the second aspect of the present application.

[0008] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the embodiment of the first aspect of the present application are implemented.

[0009] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: When recommending information to a target user, the embodiment of the present application obtains multiple second-degree relationship pairs of the target user recalled through multiple recall paths, the second-degree relationship pairs include users who have a second-degree relationship with the target user, and performs deduplication processing on the multiple second-degree relationship pairs to obtain multiple target second-degree relationship pairs. According to the confidence of at least one recall path corresponding to the target second-degree relationship pair, the target confidence and target recall path of the target second-degree relationship pair are determined, and information is recommended to the target user based on the target confidence and target recall path of the multiple target second-degree relationship pairs. The embodiment of the present application recalls multiple second-degree relationship pairs of the target user through multiple recall paths, and re-determines the target confidence and target recall path of the multiple target second-degree relationship pairs after deduplication processing, and then recommends information to the target user, thereby realizing accurate mining of potential second-degree relationships of the target user in the social network, and the second-degree relationship is richer than the first-degree relationship. Even if the target user follows fewer bloggers, the content that can be pushed to the user is relatively more, which improves the efficiency and accuracy of information dissemination and provides a more personalized and accurate information recommendation service for the target user. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flowchart of an information recommendation method provided for an embodiment of the present application; Figure 2 A flow chart of blogger tags provided for one embodiment of the present application; Figure 3 A schematic diagram of a user attention relationship provided for an embodiment of the present application; Figure 4 A flowchart of an information recommendation method provided for another embodiment of the present application; Figure 5 A flowchart of an information recommendation method provided for another embodiment of the present application; Figure 6 A schematic diagram of the structure of an FM model provided for one embodiment of the present application; Figure 7 A schematic diagram of the structure of a Deep FM model provided for one embodiment of the present application; Figure 8 A flowchart of an information recommendation method provided for another embodiment of the present application; Fig. 9 A flowchart of an information recommendation method provided for another embodiment of the present application; Fig.10 A schematic diagram of the overall flow of an information recommendation method provided for another embodiment of the present application; Fig.11 A schematic diagram of the structure of an information recommendation device provided by an embodiment of the present application; Fig.12 A schematic diagram of the structure of an electronic device provided for one embodiment of the present application. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0012] The terms "first", "second", etc. in this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here. In addition, the "and / or" in this application represents at least one of the connected objects, and the character " / " generally represents that the front and back associated objects are in an "or" relationship. It should be noted that the data involved in this application are all obtained on the premise of obtaining user authorization.

[0013] With the popularization of the Internet and the development of social networks, people are increasingly relying on social networks for information exchange and dissemination. Traditional social relationship mining usually only focuses on the user's first-degree relationship, that is, the relationship between the user and the direct friend. For example, in the social push scenario of Weibo, the content of interest to the user is usually pushed to the user based on the user's first-degree relationship. For example, the materials of the blogger followed by the user are pushed to the user based on the follow relationship between the user and the blogger. However, the first-degree relationships of some users are not rich. For example, some users follow fewer bloggers, so the content that can be pushed to the user is relatively small, which reduces the efficiency and accuracy of information dissemination. To this end, the present application proposes an information recommendation method, device, electronic device and storage medium to solve the problem that traditional social relationship mining only focuses on the user's first-degree relationship, which reduces the efficiency and accuracy of information dissemination.

[0014] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0015] Figure 1 A flowchart of an information recommendation method provided by an embodiment of the present application is shown below. Figure 1 As shown, the information recommendation method of the embodiment of the present application may specifically include the following steps: S101, obtaining a plurality of second-degree relationship pairs of target users recalled through a plurality of recall paths, wherein the second-degree relationship pairs include users having a second-degree relationship with the target users.

[0016] In an embodiment of the present application, the execution subject of the information recommendation method of the embodiment of the present application is an information recommendation device, which can be set in an electronic device. The electronic device can be a terminal device or a server. Among them, the terminal device can be a mobile phone, a tablet computer, a desktop computer, a portable notebook, a car-mounted device, etc.; the server can be an independent server or a server cluster composed of multiple servers. The electronic device has at least one social platform, such as a Weibo platform. There are multiple services on the social platform, such as push services, sharing services, etc.

[0017] Target user, that is, the user whose second-degree relationship is currently to be mined.

[0018] The second-degree relationship, the simplest and most direct second-degree relationship is the friend of a friend. Taking the social push scenario of Weibo as an example, user A and user B follow each other, and user B and user C follow each other, then there is a second-degree relationship between user A and user C.

[0019] A second-degree relationship pair is a user that has a second-degree relationship with a user. In the embodiment of the present application, the user in the second-degree relationship pair is a user that has a second-degree relationship with the target user. Still taking the above example, assuming that user A is the target user, user C is a user that has a second-degree relationship with user A, and the second-degree relationship pair of user A is user C.

[0020] Data of the target user in the social network can be collected, such as the target user's basic information, friend list, interaction records and other data, to mine the target user's potential second-degree relationship pairs. Second-degree relationship pairs can be mined and recalled through multiple recall paths, and the confidence of second-degree relationship pairs mined through different recall paths will also be different. Among them, these multiple recall paths may specifically include but are not limited to multiple of the following paths: 1) a path for recall based on users following each other; 2) a path for recall based on user interactive behavior; 3) a path based on users sharing outside the site; 4) a path based on user tag data; 5) a path based on users following authors following authors; 6) a path based on similar authors of authors followed by users; and, 7) a path based on user intimacy. The mining process of each recall path is described in detail below: 1) Mining second-degree relationship pairs based on the way users follow each other for recall.

[0021] For example, in the use scenario of Weibo, if two users follow each other, then these two users are very likely to know each other in real life. Therefore, the second-degree relationship of users can be mined through the relationship pairs of mutual attention. If user A and user B follow each other, user B and user C follow each other, and user A does not follow user C, then user C is the second-degree relationship mined by user A through the way of mutual attention. User A and user C are each other's second-degree relationship pair.

[0022] The second-degree relationships mined in this way are the most direct and intuitive. However, in the context of Weibo, there are relatively few pairs of people who follow each other, so other methods are needed to deeply mine the second-degree relationships of users.

[0023] 2) Mining second-degree relationship pairs based on the recall approach of user interaction behavior.

[0024] For example, in the use scenario of Weibo, the user's behavior in the Weibo stream, such as forwarding, commenting, liking, etc., usually means that the user is interested in the content of this Weibo. If two users have frequent interactive behaviors in a short period of time, such as tagging each other, then the two users form a potential second-degree relationship.

[0025] It should be noted here that in the embodiment of the present application, the user's interactive behavior is mainly a unilateral behavior. For example, if user A clicks on user B's Weibo multiple times in the Weibo stream to read and comment, user A recalls user B as a second-degree relationship pair.

[0026] In actual application, only the data of forwarding, commenting, and liking can be used. On the one hand, because forwarding, commenting, and liking are strong interactive behaviors, the second-degree relationship pairs recalled by this method can have a higher confidence. On the other hand, other weaker interactive behaviors are more numerous and more messy, which will bring a large amount of calculation in data processing, but the benefits are low, so they can be omitted.

[0027] 3) Mining second-degree relationship pairs based on users’ off-site sharing channels.

[0028] For example, in the use scenario of Weibo, off-site sharing means that user A shares a Weibo link to other platforms, such as WeChat, QQ, etc. Other users open this Weibo through the link, so the user who opens the Weibo through the link and user A form a second-degree relationship.

[0029] For example, if user B clicks on a Weibo link shared by user A through WeChat, then a second-degree relationship pair is formed between user A and user B.

[0030] 4) Mining second-degree relationship pairs based on user tag data.

[0031] On social platforms, many users will label their friends, such as classmates, family members, colleagues, etc. These users who are labeled with the same label by a certain user may form a potential second-degree relationship with each other. For example, if user A labels user B and user C as classmates, then user B and user C are very likely to be classmates, that is, there is a potential second-degree relationship between user B and user C.

[0032] In actual application, only the three labels of classmates, colleagues and family members can be used to mine second-degree relationships. On the one hand, these labels are more common and have a higher confidence level. On the other hand, these labels do not need to be specially screened and can be used directly.

[0033] 5) Mining second-degree relationship pairs based on the ways in which users follow authors.

[0034] For example, in the usage scenario of Weibo, if users are regarded as nodes and follow-up relationships as directed edges, a social network can be naturally constructed. Suppose user A follows user B, and user B follows user C. In this social network, the distance between user A and user C is 2, forming a relatively significant second-degree relationship.

[0035] However, a large number of second-degree relationship pairs will be screened out in this way. For example, if user A follows 100 users and there are 100 first-degree relationship pairs, then 10,000 second-degree relationship pairs can be mined in this way. Not only will there be an exponential increase in the amount of computation, but the confidence of the second-degree relationship pairs is also difficult to guarantee. Therefore, some filtering is still needed.

[0036] In order to make the second-degree relationship more accurate, if multiple bloggers followed by user A also follow user B, it can be said to a certain extent that these bloggers followed by user A have common interests with user B, and the second-degree relationship between user A and user B is also more reliable. At the same time, limiting the number of people following at the same time can also avoid the problem of magnitude explosion in the calculation process.

[0037] Based on data statistics, in actual application, we can set that when user A has more than 3 bloggers who follow the same user B at the same time, the two users (i.e. user A and user B) will be recalled as a second-degree relationship pair.

[0038] It should be noted here that the second-degree relationship pairs recalled through this channel will have a certain degree of duplication with the second-degree relationship pairs recalled through mutual attention in 1). The essence of this channel of recall is to supplement the mutual attention recall to a certain extent. The repeated recalled second-degree relationship pairs will also be processed in the subsequent process.

[0039] 6) Mining second-degree relationship pairs based on the paths of similar authors that users follow.

[0040] For example, in the usage scenario of Weibo, there are many bloggers who are similar, e.g. Figure 2 The two bloggers in the figure are obviously both entertainment and comedy bloggers, and their microblogs are similar in content. Figure 2 The blogger's popular memes in the blog show that user A is more interested in funny content, so he is likely to Figure 2 If user A is interested in the blogger's funny jokes, then the second-degree relationship pair of blogger's funny jokes can be recalled for user A.

[0041] In this case, the problem that needs to be solved is how to find bloggers similar to a blogger. The simplest way is to search through some tags, such as Figure 2 The two bloggers in the example are both funny and humorous bloggers. However, there may be many bloggers with the same tag. User A follows Figure 2 For the blogger hot memes in , it is impossible for us to recall all bloggers with the label of funny and humorous bloggers. We can only select a few bloggers with the most similar hot memes to the bloggers from a series of bloggers with the label of funny and humorous bloggers to recall the second-degree relationship pairs.

[0042] Specifically, first, the biased random walk Node2vec algorithm is used to generate a blogger embedding vector for each blogger, and the embedding similarity calculation is used to screen similar bloggers. The random walk algorithm of Node2vec is as follows: (1) where ∂ pq ( t,x ) represents the transition probability from blogger t to blogger x , that is, the probability of selecting the next blogger as t when choosing a path, x , d tx represents the distance between blogger t and blogger x , and p and q are two set hyperparameters used to control the random walk direction of the blogger t . When p > max(q, 1), the blogger t tends to walk towards a blogger farther away when walking, and when p < max(q, 1), the blogger t tends to return to the previous blogger when walking.

[0043] It can be set that p = 1 and q = 0.2, so that the random walk sequence tends to walk between adjacent bloggers nearby, which emphasizes the homogeneity between bloggers more and can better reflect the social relationship between users.

[0044] The embedding of each blogger can be specifically generated through the following process: Taking the attention relationship graph between multiple bloggers A to blogger I shown in Figure 3 as an example, where A follows B, C, D, and E (the AB edge represents that A follows B or B follows A), B follows A, G, H, and I,....

[0045] First, each blogger initializes an embedding: E A ~E I , and then multiple paths are randomly selected for the training of the embedding. The training method is the same as the word2vec algorithm for generating word vectors. Initialize a vector for each node, and then train to reduce the loss and make the loss converge to finally obtain the embedding of each blogger. Among them, the process of generating training samples based on randomly selected paths is as follows: Randomly select a blogger, such as blogger A, and start a random walk from blogger A, d AB = 1, d AC = 1,.... d ​AG =2, d AA = 0, ..., we get a random walk path: A→B→B→H→B→I, and the corresponding training sample is [E A , E B , E B , E H , E B ] → E A , the previous [E A , E B , E B , E H , E B ] is a walking sequence, and the following E A To predict the target, similarly, multiple training samples can be obtained, such as [E B , E A , E C , E G , E B ] → E H wait.

[0046] After generating embeddings for each blogger, we need to calculate the similarity between bloggers with the same tag according to the blogger's tag, and then filter out bloggers similar to each blogger. However, in actual applications, there are many bloggers with some tags, and calculating them pairwise will bring huge computational complexity. For example, if there are tens of thousands of bloggers with the tag of handsome men and beautiful women, the computational complexity will be in the hundreds of millions. Therefore, it is necessary to divide these bloggers into more fine-grained categories.

[0047] Specifically, a clustering algorithm, such as the k-means clustering algorithm, can be used to cluster bloggers with the same tag to achieve a more fine-grained division. The clustering process is as follows: Step 1: Randomly select 100 bloggers from bloggers of a certain tag as the initial cluster centers; Step 2: Calculate the distance between each blogger and each cluster center, and assign each blogger to the cluster closest to it. Step 3: After all bloggers are assigned, 100 cluster centers are updated. The cluster center is defined as the mean of all bloggers in the cluster in each dimension. Step 4, compare the 100 cluster centers after the update with the 100 cluster centers before the update. If the cluster centers have changed, go to step 2, otherwise go to step 5; Step 5: When the cluster center no longer changes, stop and output the clustering results, and then organize the required information, such as the category to which each blogger belongs, for subsequent statistics and analysis.

[0048] It should be noted here that the distance calculation in step 2 of the above clustering process can be achieved by calculating the cosine similarity. The formula of cosine similarity is as follows: (2) in, similarity represents cosine similarity, A∙B represents the scalar product of vectors A and B, and ||A|| represents the modulus of vector A. The cosine similarity is also used in the subsequent calculation of similar bloggers.

[0049] After clustering is completed, we only need to calculate the similarity between bloggers in each category to calculate similar bloggers for each blogger. The following Table 1 is an example of similarity calculation between four bloggers in the field of humor (hot memes, funny memes, silly memes, and human humorous behaviors): Table 1 Examples of similarity between bloggers

[0050] After calculating similar bloggers for each blogger, among the similar bloggers selected, up to 5 bloggers with the most recent interactions can be selected for the target user, and then each blogger can select up to 5 similar bloggers for recall, and at most 25 similar bloggers who may be of interest can be recalled as second-degree relationship pairs for the target user.

[0051] 7) Mining second-degree relationship pairs based on user intimacy.

[0052] In the above-mentioned approach 2) of mining second-degree relationship pairs based on user interactive behavior, second-degree relationships are recalled for users with strong interactive behaviors such as forwarding, commenting, and liking. However, the interactive data such as forwarding, commenting, and liking are relatively sparse, and fewer second-degree relationship pairs can be recalled. Therefore, it is necessary to process the interactive data between users in order to make full use of all interactive data between users to recall second-degree relationships.

[0053] In the embodiment of the present application, it is necessary to first quantify all the interactive behaviors of users in order to compare the overall intensity of the interaction between users, i.e., intimacy. Then, all the quantified user behaviors are input into the intimacy calculation formula to calculate the intimacy between users. Among them, the interactive behaviors may specifically include but are not limited to the following ten interactive behaviors: forwarding, commenting, liking, sharing, opening push, tagging, clicking, collecting, private messaging, and visiting homepage.

[0054] Specifically, we first need to normalize each type of behavior. For example, for forwarding behavior, we can count all the user's forwarding times within 7 days and the number of forwardings to a specified blogger, and then count the proportion of the blogger in the user's forwarding times. For example, user A forwarded 10 Weibo posts within 7 days, 6 of which were posted by user B. Then user B's proportion in user A's forwarding times is 6 / 10=0.6, that is, the normalized value of the forwarding behavior is 0.6, which is used to subsequently calculate the intimacy between user A and user B.

[0055] After normalizing all interactive behaviors, the intimacy between two users is calculated using the following formula: (3) Among them, y represents intimacy, ω i Representative i The weight of the interaction behavior, x i Representative i The normalized value of each interactive behavior. The intimacy calculated by this formula ranges from 0 to 1, which is convenient for subsequent sorting calculations.

[0056] In actual application, based on the intimacy calculated by the above formula, up to 10 bloggers with the highest intimacy can be recalled for each user as the recalled second-degree relationship pairs.

[0057] S102, performing deduplication processing on multiple second-degree relationship pairs to obtain multiple target second-degree relationship pairs.

[0058] In the embodiment of the present application, a potential second-degree relationship pair may be simultaneously excavated and recalled by multiple channels. Therefore, there may be duplicate second-degree relationship pairs among the multiple second-degree relationship pairs obtained in step S101. It is necessary to perform deduplication processing on the multiple second-degree relationship pairs obtained in step S101 to obtain multiple non-duplicate second-degree relationship pairs, which are recorded as target second-degree relationship pairs. For example, for target user A, the second-degree relationship pairs recalled through channel 1) are B, C, and D, the second-degree relationship pairs recalled through channel 2) are C, D, and E, and the second-degree relationship pairs recalled through channel 3) ..., then the multiple second-degree relationship pairs obtained by S101 are: B, C, D, C, D, E, ..., and after deduplication processing, the multiple target second-degree relationship pairs obtained are: B, C, D, E, ....

[0059] S103: Determine a target confidence and a target recall path of the target second-degree relationship pair according to the confidence of at least one recall path corresponding to the target second-degree relationship pair.

[0060] In the embodiment of the present application, the confidence levels of the second-degree relationship pairs recalled by different recall paths are different. The confidence level of each recall path can be obtained offline in advance through the following steps: use a small amount of traffic for a period of time, and count the click rates of the second-degree relationship pairs recalled by various recall paths when issuing materials separately, and give different confidence scores (i.e., confidence levels) to each recall path according to the click rate.

[0061] In addition, a title needs to be assigned to each recall path, which is used as the title when issuing the materials of the second-degree relationship pair corresponding to the recall path. The following Table 2 shows the titles and confidences corresponding to different recall paths.

[0062] Table 2 The titles and confidences corresponding to different recall channels

[0063] Since the confidence of the second-degree relationship pairs recalled by different recall pathways is different, and there may be multiple recall pathways for a second-degree relationship pair, it is necessary to re-determine the target confidence and target recall pathway of each target second-degree relationship pair based on at least one recall pathway and the corresponding confidence corresponding to each target second-degree relationship pair: the sum of the confidences of each recall pathway corresponding to the target second-degree relationship pair is determined as the target confidence of the target second-degree relationship pair. The recall pathway with the highest confidence corresponding to the target second-degree relationship pair is determined as the final recall pathway of the target second-degree relationship pair, i.e., the target recall pathway.

[0064] For example, if the second-degree relationship has only one recall path for user B: Path 1), then the target confidence of user B is determined to be 0.7, and the target recall path is Path 1). If the second-degree relationship has two recall paths for user C: Path 1) and Path 2), then the target confidence of user C is determined to be 0.7+0.4=1.1, and the target recall path is Path 1) corresponding to 0.7.

[0065] S104: Recommend information to the target user based on the target confidence and target recall path of the plurality of target second-degree relationship pairs.

[0066] In the embodiment of the present application, information recommendation to the target user may specifically include but is not limited to pushing materials to the target user and / or recommending interested users.

[0067] Among them, when pushing materials to target users, such as Figure 4 As shown, the above step S104 may specifically include the following steps: S1041, sorting multiple materials to be pushed associated with multiple target second-degree relationship pairs according to the target recall path of the target confidence of multiple target second-degree relationship pairs.

[0068] In the embodiment of the present application, unlike the dispatch of materials in a first-degree relationship, the dispatch of materials in a second-degree relationship must first be sorted internally, and then the sorted multiple materials are sent downstream to let the downstream decide whether to dispatch them.

[0069] Assuming that the target second-degree relationship pair determined in step S103 is user B and user C, user B has materials 1 and 2 to be pushed, and user C has materials 3 and 4 to be pushed, then according to the target confidence and target recall path of user B and the target confidence and target recall path of user C, material 1, material 2, material 3 and material 4 are sorted.

[0070] In the usage scenario of Weibo, materials refer to the blog posts published by bloggers.

[0071] S1042: Push materials to target users based on the sorting results of the multiple materials, and the number of pushed materials corresponding to each target second-degree relationship pair does not exceed a set number.

[0072] In the embodiment of the present application, when there are multiple items to be pushed that are associated with a target second-degree relationship, only the item with the highest ranking will be pushed, and the number of items pushed will not exceed the set number. Still taking the above example, assuming that the ranking results of multiple items are: item 4, item 2, item 1 and item 3, and the set number is 1, then item 4 and item 2 will be pushed to the target user.

[0073] Further, such as Figure 5 As shown, the above step S1041 may specifically include the following steps: S401, according to the target confidence and target recall path of multiple target second-degree relationship pairs, a sorting model is used to obtain click-through rate scores of multiple materials to be pushed, and each target second-degree relationship pair is associated with at least one material to be pushed.

[0074] The sorting model may specifically include but is not limited to a Factorization Machines (FM) model or a Deep FM model.

[0075] S402, sorting multiple materials to be pushed according to the click rate scores.

[0076] In the embodiment of the present application, the higher the click rate score, the higher the ranking.

[0077] As a first feasible implementation, the above step S401 "obtaining click-through rate scores of multiple materials to be pushed using a ranking model according to target confidences and target recall paths of multiple target second-degree relationship pairs" may specifically include the following steps: The target confidence of the target second-degree relationship pair, the target recall path, the characteristics of the target user, the characteristics of the target second-degree relationship pair, and any material to be pushed associated with the target second-degree relationship pair are input into the sorting model to obtain the click rate score of the material to be pushed output by the sorting model.

[0078] In an embodiment of the present application, for a material to be pushed, the target confidence of the target second-degree relationship pair corresponding to the material to be pushed, the target recall path, the characteristics of the target user, the characteristics of the target second-degree relationship pair and the material to be pushed are used as input features of the sorting model, and the sorting model outputs the click-through rate score of the material to be pushed.

[0079] When the sorting model is a FM model, the structure of the FM model is as follows: Figure 6 As shown, the expression corresponding to the FM model is as follows: (4) in, y is the click rate score of the material to be pushed. ω 0 is a constant term, ω i For the i The weight corresponding to the input feature of the dimension, n is the number of dimensions of the input feature, which is 5 in the embodiment of the present application, ω ij For the i The input features of the dimensions and j The weights corresponding to the combined features of the input features of dimensions.

[0080] As the number of target second-degree relationship pairs increases, the accuracy of the FM model in sorting is insufficient, so the FM model can be upgraded to the Deep FM model, such as Figure 7 As shown in the figure, the Deep FM model adds several layers of nonlinear deep neural networks (DNN) based on the FM model, which makes the sorting model have better fitting ability.

[0081] The Deep FM model consists of two parts: FM and DNN. The two parts share input features. The expression corresponding to the Deep FM model is as follows: y’ = sigmoid ( y FM + y DNN )(5) in, y’ is the click rate score of the material to be pushed obtained using the Deep FM model. sigmoid is the logistic regression activation function, yFM The click rate score of the materials to be pushed output by the FM part. y DNN It is the click rate score of the material to be pushed output by the DNN part.

[0082] in, y FM The corresponding expressions are as follows:

[0083] in, w is a matrix including the weights corresponding to the input features of each dimension, x is a matrix containing input features of various dimensions, V i For the i The vector output by the embedding layer corresponding to the input features of dimensions, V j For the j The vector output by the embedding layer corresponding to the input features of dimensions, x j1 For the j1 The input features are of dimensions, x j2 For the j2 The input features are of dimensions, d is the number of dimensions of the input features.

[0084] in, y DNN The corresponding expressions are as follows: a (l+1) =σ( W (1) a (1) + b (1) ) in, y DNN Output of the last embedding layer a , a (1) For the l The vector output by the embedding layer, σ is the logistic regression activation ( sigmoid )function, σ=1 / (1+e -x ) ，W is a matrix including the weights corresponding to the input features of each dimension, b is the preset bias parameter.

[0085] As a second feasible implementation method, Figure 8As shown, the above step S401 "obtaining click-through rate scores of multiple materials to be pushed using a ranking model according to target confidences and target recall paths of multiple target second-degree relationship pairs" may specifically include the following steps: S701, input the target confidence of the target second-degree relationship pair, the target recall path, the characteristics of the target user, the characteristics of the target second-degree relationship pair, and any to-be-pushed material associated with the target second-degree relationship pair into the sorting model to obtain the first click rate score of the material output by the sorting model.

[0086] S702, calculating the product of the first click rate score and the preset weighting coefficient corresponding to the confidence of the target recall path corresponding to the material to be pushed, to obtain the click rate score of the material to be pushed, and the confidence of the target recall path and the weighting coefficient are positively correlated.

[0087] In the embodiment of the present application, the target recall path is input into the ranking model as a dimensional feature to calculate the click-through rate score. However, as the ranking model features are continuously added, the weight of the target recall path in the ranking model will be diluted, and the target recall path is a very important feature. The importance of this feature cannot be reflected by the ranking model alone in predicting the click-through rate score.

[0088] Generally, the higher the click rate of a material sent through a recall channel, the higher the confidence of the recall, so the materials of the recall channel should be sent to users first. Based on the above considerations, after using the sorting model to obtain the first click rate score of the material to be pushed, appropriate score weighting is also used to promote the priority of sending the materials to be pushed through the target recall channel. Specifically: a corresponding weighting coefficient is set for each recall channel in advance. The higher the confidence of the recall channel, the larger the corresponding weighting coefficient.

[0089] Further, such as Fig. 9 As shown, based on the consideration of the importance of the target recall path, the above step S401 "obtaining the click-through rate scores of multiple materials to be pushed using a ranking model according to the target confidence and target recall path of multiple target second-degree relationship pairs" may specifically include the following steps: S801, determining a ranking model corresponding to the confidence of the target recall path according to the confidence of the target second-degree relationship pair corresponding to the target recall path.

[0090] In an embodiment of the present application, multiple confidence intervals are pre-set, and a sorting model is trained for each confidence interval. According to the confidence of the target recall path of the target second-degree relationship pair corresponding to the material to be pushed, the confidence interval in which the confidence is located is determined, and then the sorting model corresponding to the confidence interval is determined as the sorting model corresponding to the material to be pushed.

[0091] Among them, the number of ranking models can be two: a high click-through rate ranking model (the confidence interval corresponding to the high click-through rate ranking model is [n, 1), for example, the corresponding confidence interval is [0.5, 1)) and a low click-through rate ranking model (the confidence interval corresponding to the low click-through rate ranking model is (0, n), for example, the corresponding confidence interval is (0, 0.5)).

[0092] Specifically, the most ideal way is to train a ranking model for each recall separately, so that the push to users can be the most accurate. However, introducing multiple ranking models will consume more resources when calculating online; some recall channels have insufficient training samples due to the small amount of click data sent down, making it difficult for the ranking model to be accurately trained, and the accuracy is poor; the click-through rate scores given by each recall channel need to be manually adjusted. If the weights are adjusted according to the historical click-through rate, the click-through rate may decrease if the high-scoring channels are sent down more, and the click-through rate of the low-scoring channels with fewer sends will increase, which will cause oscillations online and is not conducive to the stability of the ranking model. Considering the above, two ranking models can be used, and the click-through rates of these recall channels are also mainly distributed in two intervals.

[0093] S802, according to the target confidence and target recall path of the target second-degree relationship pair, a corresponding sorting model is used to obtain the click rate score of each to-be-pushed material associated with the target second-degree relationship pair.

[0094] In an embodiment of the present application, according to the target confidence and target recall path of the target second-degree relationship pair corresponding to the material to be pushed, the sorting model corresponding to the material to be pushed is used to obtain the click rate score of the material to be pushed.

[0095] To clearly illustrate the information recommendation method of the embodiment of the present application, Fig.10 The overall process of the information recommendation method of the embodiment of the present application is described in detail. Fig.10 As shown, the information recommendation method of the embodiment of the present application includes the following steps: obtaining multiple second-degree relationship pairs of target users recalled through multiple recall paths; deduplicating the multiple second-degree relationship pairs to obtain multiple target second-degree relationship pairs; determining the target confidence and target recall path of the target second-degree relationship pair according to the confidence of at least one recall path corresponding to the target second-degree relationship pair; for the materials to be pushed in the target recall path with high confidence, a high click-through rate sorting model is used to obtain a first click-through rate score; for the materials to be pushed in the target recall path with low confidence, a low click-through rate sorting model is used to obtain a first click-through rate score; calculating the product of the first click-through rate score and a preset weighting coefficient to obtain the click-through rate score of the material to be pushed; and sorting and issuing the multiple materials to be pushed according to the click-through rate scores.

[0096] In summary, the information recommendation method of the embodiment of the present application, when recommending information to a target user, obtains multiple second-degree relationship pairs of the target user recalled through multiple recall paths, the second-degree relationship pairs include users who have a second-degree relationship with the target user, deduplicates the multiple second-degree relationship pairs to obtain multiple target second-degree relationship pairs, determines the target confidence and target recall path of the target second-degree relationship pairs according to the confidence of at least one recall path corresponding to the target second-degree relationship pairs, and recommends information to the target user based on the target confidence and target recall path of the multiple target second-degree relationship pairs. The embodiment of the present application recalls multiple second-degree relationship pairs of the target user through multiple recall paths, and re-determines the target confidence and target recall path of the multiple target second-degree relationship pairs after deduplication processing, and then recommends information to the target user, realizing accurate mining of potential second-degree relationships of the target user in the social network, and the second-degree relationship is richer than the first-degree relationship. Even if the target user follows fewer bloggers, the content that can be pushed to the user is relatively more. When pushing materials, since the number of materials pushed by each target second-degree relationship pair does not exceed the set number, even if some bloggers followed by the target user have a high frequency of posting, the relatively balanced push of the materials to be pushed by each blogger can be achieved, avoiding the content pushed to the user being relatively single, improving the efficiency and accuracy of information dissemination, and providing the target user with a more personalized and accurate information recommendation service. The product of the first click-through rate score of the material to be pushed output by the sorting model and the corresponding weighting coefficient is calculated to obtain the click-through rate score of the material to be pushed, realizing the appropriate weighting of the materials to be pushed in the recall path with high confidence, making the click-through rate score more accurate, and further improving the accuracy of information dissemination. The number of sorting models is two, which avoids consuming more resources due to too many sorting models and inaccurate sorting model training due to insufficient training samples, and improves the stability of the sorting model.

[0097] Fig.11 This is a schematic diagram of the structure of an information recommendation device provided by an embodiment of the present application. Fig.11 As shown, the information recommendation device 1000 of the embodiment of the present application may specifically include: an acquisition module 1001, a deduplication module 1002, a determination module 1003 and a recommendation module 1004. Among them: The acquisition module 1001 is used to acquire multiple second-degree relationship pairs of the target user recalled through multiple recall paths, where the second-degree relationship pairs include users that have second-degree relationships with the target user.

[0098] The deduplication module 1002 is used to perform deduplication processing on multiple second-degree relationship pairs to obtain multiple target second-degree relationship pairs.

[0099] The determination module 1003 is used to determine a target confidence and a target recall path of the target second-degree relationship pair according to the confidence of at least one recall path corresponding to the target second-degree relationship pair.

[0100] The recommendation module 1004 is used for target confidence and target recall path based on multiple target second-degree relationship pairs, Recommend information to target users.

[0101] In the embodiment of the present application, the specific process of each module and unit in the information recommendation device of the embodiment of the present application realizing its function can refer to the relevant description in the above-mentioned information recommendation method embodiment, which will not be repeated here.

[0102] In summary, the information recommendation device of the embodiment of the present application, when recommending information to a target user, recalls multiple second-degree relationship pairs of the target user through multiple recall paths, and re-determines the target confidence and target recall path of the multiple target second-degree relationship pairs after deduplication processing, and then recommends information to the target user, thereby achieving accurate mining of potential second-degree relationships of the target user in the social network. Second-degree relationships are richer than first-degree relationships. Even if the target user follows fewer bloggers, the content that can be pushed to the user is relatively more. When pushing materials, since the number of materials pushed for each target second-degree relationship pair does not exceed the set number, even if some bloggers followed by the target user have a high posting frequency, relatively balanced push of materials to be pushed by each blogger can be achieved, thereby avoiding relatively single content pushed to the user, improving the efficiency and accuracy of information dissemination, and providing more personalized and accurate information recommendation services for the target user. The product of the first click rate score of the material to be pushed output by the sorting model and the corresponding weighting coefficient is calculated to obtain the click rate score of the material to be pushed, which realizes the appropriate weighting of the materials to be pushed in the recall path with high confidence, making the click rate score more accurate and further improving the accuracy of information dissemination. The number of sorting models is two, which avoids the consumption of more resources due to too many sorting models and the inaccurate sorting model training caused by insufficient training samples, and improves the stability of the sorting model.

[0103] The present application also provides an electronic device. Fig.12 As shown, the electronic device 1100 includes: a processor 1101, a memory 1102, and a program or instruction stored in the memory 1102 and executable on the processor 1101. When the program or instruction is executed by the processor 1101, the steps of the information recommendation method in any of the above embodiments are implemented.

[0104] The electronic device of the embodiment of the present application, when recommending information to the target user, recalls multiple second-degree relationship pairs of the target user through multiple recall paths, and re-determines the target confidence and target recall path of the multiple target second-degree relationship pairs after deduplication processing, and then recommends information to the target user, realizing accurate mining of potential second-degree relationships of the target user in the social network, and the second-degree relationship is richer than the first-degree relationship, even if the target user follows fewer bloggers, the content that can be pushed to the user is relatively more, and when pushing materials, since the number of materials pushed by each target second-degree relationship pair does not exceed the set number, even if some bloggers followed by the target user have a high frequency of posting, the relatively balanced push of the materials to be pushed by each blogger can be achieved, avoiding the content pushed to the user being relatively single, improving the efficiency and accuracy of information dissemination, and providing the target user with a more personalized and accurate information recommendation service. The product of the first click-through rate score of the material to be pushed output by the sorting model and the corresponding weighting coefficient is calculated to obtain the click-through rate score of the material to be pushed, realizing the appropriate weighting of the material to be pushed in the recall path with high confidence, making the click-through rate score more accurate, and further improving the accuracy of information dissemination. The number of sorting models is two, which avoids consuming more resources due to too many sorting models and inaccurate sorting model training due to insufficient training samples, and improves the stability of the sorting model.

[0105] An embodiment of the present application also proposes a readable storage medium, on which one or more computer programs are stored. The one or more computer programs include instructions. When the program or instructions are executed by a processor in an electronic device that includes multiple applications, the processor in the electronic device can execute the various steps of the above-mentioned information recommendation method embodiment.

[0106] The readable storage medium of the embodiment of the present application, when recommending information to the target user, recalls multiple second-degree relationship pairs of the target user through multiple recall paths, and re-determines the target confidence and target recall path of the multiple target second-degree relationship pairs after deduplication processing, and then recommends information to the target user, realizing accurate mining of potential second-degree relationships of the target user in the social network, and the second-degree relationship is richer than the first-degree relationship, even if the target user follows fewer bloggers, the content that can be pushed to the user is relatively more, and when pushing materials, since the number of materials pushed by each target second-degree relationship pair does not exceed the set number, even if the target user follows some bloggers with a high frequency of posting, it is possible to achieve a relatively balanced push of the materials to be pushed by each blogger, avoiding the content pushed to the user being relatively single, improving the efficiency and accuracy of information dissemination, and providing the target user with a more personalized and accurate information recommendation service. The product of the first click-through rate score of the material to be pushed output by the sorting model and the corresponding weighting coefficient is calculated to obtain the click-through rate score of the material to be pushed, realizing the appropriate weighting of the materials to be pushed in the recall path with high confidence, making the click-through rate score more accurate, and further improving the accuracy of information dissemination. The number of sorting models is two, which avoids consuming more resources due to too many sorting models and inaccurate sorting model training due to insufficient training samples, and improves the stability of the sorting model.

[0107] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0108] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0109] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0110] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0111] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0113] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0114] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0115] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0116] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0117] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0118] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0119] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. An information recommendation method, characterized in that: include: Acquire multiple second-degree relationship pairs of target users recalled through multiple recall paths, wherein the second-degree relationship pairs include users having second-degree relationships with the target users; Performing deduplication processing on the multiple second-degree relationship pairs to obtain multiple target second-degree relationship pairs; Determining a target confidence and a target recall path of the target two-degree relationship pair according to the confidence of at least one recall path corresponding to the target two-degree relationship pair; Based on the target confidences and target recall paths of the multiple target second-degree relationship pairs, information is recommended to the target user.

2. The method according to claim 1, characterized in that: The recommending information to the target user based on the target confidence and target recall path of the plurality of target second-degree relationship pairs includes: Sorting the multiple materials to be pushed associated with the multiple target second-degree relationship pairs according to the target confidences and target recall paths of the multiple target second-degree relationship pairs; According to the sorting results of the multiple materials to be pushed, materials are pushed to the target user, and the number of pushed materials corresponding to each target second-degree relationship pair does not exceed a set number.

3. The method according to claim 1, characterized in that The step of determining a target confidence and a target recall path of the target second-degree relationship pair according to the confidence of at least one recall path corresponding to the target second-degree relationship pair comprises: Determining the sum of the confidences of the recall paths corresponding to the target second-degree relationship pair as the target confidence of the target second-degree relationship pair; The recall path with the highest confidence corresponding to the target second-degree relationship pair is determined as the target recall path of the target second-degree relationship pair.

4. The method according to claim 2, characterized in that: The step of sorting the multiple materials to be pushed associated with the multiple target second-degree relationship pairs according to the target confidences and target recall paths of the multiple target second-degree relationship pairs includes: According to the target confidence and target recall path of the multiple target second-degree relationship pairs, a sorting model is used to obtain click-through rate scores of the multiple materials to be pushed; each target second-degree relationship pair is associated with at least one material to be pushed; The multiple materials to be pushed are sorted according to the click rate scores.

5. The method according to claim 4, characterized in that The step of obtaining click-through rate scores of the plurality of materials to be pushed by using a sorting model according to the target confidences and target recall paths of the plurality of target second-degree relationship pairs includes: The target confidence and target recall path of the target second-degree relationship pair, the characteristics of the target user, the characteristics of the target second-degree relationship pair, and any material to be pushed associated with the target second-degree relationship pair are input into the sorting model to obtain the click rate score of the material to be pushed output by the sorting model.

6. The method according to claim 4, characterized in that The step of obtaining click-through rate scores of the plurality of materials to be pushed by using a sorting model according to the target confidences and target recall paths of the plurality of target second-degree relationship pairs includes: Inputting the target confidence and target recall path of the target second-degree relationship pair, the characteristics of the target user, the characteristics of the target second-degree relationship pair, and any to-be-pushed material associated with the target second-degree relationship pair into the sorting model, and obtaining the first click rate score of the to-be-pushed material output by the sorting model; The product of the first click rate score and the preset weighting coefficient corresponding to the confidence of the target recall path corresponding to the material to be pushed is calculated to obtain the click rate score of the material to be pushed, and the confidence of the target recall path and the weighting coefficient are positively correlated.

7. The method according to claim 4, characterized in that The step of obtaining click-through rate scores of the plurality of materials to be pushed by using a sorting model according to the target confidences and target recall paths of the plurality of target second-degree relationship pairs includes: Determine a ranking model corresponding to the confidence of the target recall path according to the confidence of the target second-degree relationship; According to the target confidence and target recall path of the target second-degree relationship pair, a corresponding sorting model is used to obtain the click rate score of each to-be-pushed material associated with the target second-degree relationship pair.

8. The method according to claim 7, characterized in that There are two ranking models: a high click-through rate ranking model and a low click-through rate ranking model. The confidence interval corresponding to the high click-through rate ranking model is [n, 1), and the confidence interval corresponding to the low click-through rate ranking model is (0, n).

9. An information recommendation device, characterized in that: include: An acquisition module, used for acquiring a plurality of second-degree relationship pairs of target users recalled through a plurality of recall paths, wherein the second-degree relationship pairs include users having a second-degree relationship with the target user; A deduplication module, used for performing deduplication processing on the plurality of second-degree relationship pairs to obtain a plurality of target second-degree relationship pairs; A determination module, configured to determine a target confidence and a target recall path of the target second-degree relationship pair according to the confidence of at least one recall path corresponding to the target second-degree relationship pair; The recommendation module is used to recommend information to the target user based on the target confidence and target recall path of the multiple target second-degree relationship pairs.

10. An electronic device, characterized in that: The method comprises a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method according to any one of claims 1 to 8.

Images