Article Recommendation Method, Device, Equipment and Storage Medium Based on MHT Algorithm

Abstract

The embodiments of the present application provide an article recommendation method, device, equipment, and storage medium based on the MHT algorithm, which relate to the field of network media in artificial intelligence technology. The method includes: receiving a recommendation request for requesting T recommended articles, where T > 1; in response to the recommendation request, selecting N candidate samples from M collected article samples, where M ≥ N ≥ 1; based on the scores of each article sample, selecting K article samples from the N candidate samples and determining the K article samples as K root nodes, where N ≥ K ≥ 1; based on the K root nodes, using the MHT algorithm to determine the T recommended articles: outputting the T recommended articles. This method can improve the user experience, enhance the flexibility and controllability of recommendations, ensure the integrity of trajectory traversal, reduce the time complexity of traversal, be beneficial to improving the effectiveness of the MHT model, and also be beneficial to introducing more complex models.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of network media in artificial intelligence technology, and more specifically, to an article recommendation method, device, equipment and storage medium based on the MHT algorithm. Background Art

[0002] In recent years, with the rapid development of information technology, various information has grown explosively, and users cannot quickly obtain the information that is truly meaningful to them from the vast amount of information. Therefore, personalized article recommendation has become an important research topic.

[0003] Generally, an article recommendation system is used to recommend articles that users are interested in. The rearrangement module in the article recommendation system is used to balance the diversity of articles and the matching degree between the articles and the user interests through some strategies. The rearrangement module generally uses a streaming processing method and a batch processing method to select and scatter the scores used to represent the matching degree between the user interests and the articles of multiple articles. The streaming processing method collects the sorted articles based on rules and cooperates with a degradation strategy to ensure the number of collected articles. The batch processing method first collects according to the collection strategy and then fine-tunes according to the scattering strategy. However, the streaming processing method has serious collection timeliness problems. In addition, by cooperating with the degradation strategy to ensure the number of collected articles, the experience is affected. The batch processing method is likely to cause the subsequent rules to break the previous rules, resulting in the problem of poor interpretability of recommendations, and it is very easy to have an unreasonable order, seriously affecting the user experience effect. Summary of the Invention

[0004] The embodiments of the present application provide an article recommendation method, device, equipment and storage medium based on the MHT algorithm, which can improve the user experience effect, improve the flexibility and controllability of recommendations, ensure the integrity of trajectory traversal, reduce the time complexity of traversal, be beneficial to improving the timeliness of the MHT model, and also be beneficial to introducing more complex models.

[0005] On the one hand, provided is a

[0006] Receive a recommendation request for requesting T recommended articles, where T > 1;

[0007] In response to the recommendation request, select N candidate samples from M collected article samples, and each of the M article samples has a corresponding score used to represent the matching degree between the user interest and the article sample, where M ≥ N ≥ 1;

[0008] Based on the score of each article sample, select K article samples from the N candidate samples and determine the K article samples as K root nodes, where N ≥ K ≥ 1;

[0009] Based on the K root nodes, determine the T recommended articles in the following manner:

[0010] Based on each parent node, among the remaining article samples in the M article samples except for the article samples already selected in the trajectory where the parent node is located, select the top X article samples with a higher benefit score relative to the parent node as the X child nodes of the parent node; use each of the X child nodes as a new parent node to continue extending child nodes. When the level where the leaf nodes are located is T, determine the T article samples on the target trajectory with the highest benefit score among the trajectories where all leaf nodes are located as the T recommended articles;

[0011] Output the T recommended articles.

[0012] On the other hand, there is provided a

[0013] Receiving unit, configured to receive a recommendation request for requesting T recommended articles, where T ≥ 1;

[0014] Selecting unit, configured to, in response to the recommendation request, select N candidate samples from the collected M article samples, and each article sample in the M article samples has a corresponding score for characterizing the matching degree between the user interest and the article sample, where M ≥ N ≥ 1;

[0015] Determining unit, configured to:

[0016] Based on the scores of each article sample, select K article samples from the N candidate samples and determine the K article samples as K root nodes, where N ≥ K ≥ 1;

[0017] Based on the K root nodes, determine the T recommended articles in the following manner:

[0018] Based on each parent node, among the remaining article samples in the M article samples except for the article samples already selected in the trajectory where the parent node is located, select the top X article samples with a higher benefit score relative to the parent node as the X child nodes of the parent node; use each of the X child nodes as a new parent node to continue extending child nodes. When the level where the leaf nodes are located is T, determine the T article samples on the target trajectory with the highest benefit score among the trajectories where all leaf nodes are located as the T recommended articles;

[0019] Output unit, configured to output the T recommended articles.

[0020] On the other hand, the present application provides an electronic device, including:

[0021] A processor, adapted to implement computer instructions; and,

[0022] A computer-readable storage medium stores computer instructions, and the computer instructions are suitable for being loaded and executed by a processor to perform the above-mentioned article recommendation method based on the MHT algorithm.

[0023] On the other hand, an embodiment of the present application provides a computer-readable storage medium. When the computer instructions stored in the computer-readable storage medium are read and executed by a processor of a computer device, the computer device is caused to perform the above-mentioned article recommendation method based on the MHT algorithm.

[0024] On the other hand, an embodiment of the present application provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the above-mentioned article recommendation method based on the MHT algorithm.

[0025] In the embodiment of the present application, the T recommended articles are determined based on the MHT algorithm. Compared with the serial processing method of batch processing, all conditions can be considered each time of collection, avoiding unreasonable sequences, such as adjacent homogeneous content, which reduces the user's experience of obtaining information. Furthermore, the capabilities of the recall module and the sorting module can be ensured, and the user experience effect can be improved. In addition, compared with the degradation strategy of stream processing, the final recommended articles are selected through a quantitative scoring method, and the articles finally recommended to the user are more flexible and controllable. In addition, based on K article samples selected from N candidate samples, the MHT algorithm is used to collect T recommended articles, that is, K article samples (i.e., seed nodes) selected from N candidate samples are used as root nodes to extend child nodes, which can ensure the integrity of trajectory traversal, reduce the time complexity of traversal, is beneficial to improving the effectiveness of the MHT model, and is also beneficial to introducing more complex models; for example, even if a more complex model for calculating revenue scores is introduced, and K article samples selected from N candidate samples are used as root nodes to extend child nodes, it can also avoid too long traversal time on the basis of ensuring the integrity of trajectory traversal. In short, the solution provided by the embodiment of the present application can improve the user experience effect, improve the flexibility and controllability of recommendations, ensure the integrity of trajectory traversal, reduce the time complexity of traversal, is beneficial to improving the effectiveness of the MHT model, and is also beneficial to introducing more complex models. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is an example of the system framework provided by an embodiment of the present application.

[0027] Figure 2 is a schematic flowchart of the article recommendation method based on the MHT algorithm provided by an embodiment of the present application.

[0028] Figure 3 It is an example of the article recommendation method based on the MHT algorithm provided by the embodiments of the present application.

[0029] Figure 4 It is an example of the article recommendation device based on the MHT algorithm provided by the embodiments of the present application.

[0030] Figure 5 It is a schematic block diagram of an electronic device provided by the embodiments of the present application. Detailed implementation manners

[0031] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0032] The solution provided by the present application may be related to artificial intelligence technology.

[0033] Among them, artificial intelligence (AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results of theory, methods, technologies, and application systems. In other words, artificial intelligence is a comprehensive technology of computer science. It attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning, and decision-making.

[0034] It should be understood that artificial intelligence technology is an interdisciplinary subject, involving a wide range of fields, including both hardware-level technologies and software-level technologies. Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technologies mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0035] With the research and progress of artificial intelligence technology, artificial intelligence technology has been studied and applied in multiple fields. For example, common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, driverless, autonomous driving, drones, robots, smart healthcare, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0036] Embodiments of this application may relate to computer vision (CV) technology in artificial intelligence technology. Computer vision is a science that studies how to enable machines to "see". More specifically, it refers to using cameras and computers to replace the human eye to perform machine vision such as target recognition, tracking, and measurement on targets, and further performing graphics processing to make the computer-processed images more suitable for human eye observation or transmission to instrument detection. As a scientific discipline, computer vision studies related theories and technologies and attempts to establish an artificial intelligence system that can obtain information from images or multi-dimensional data. Computer vision technology generally includes technologies such as image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, and also includes common biometric recognition technologies such as face recognition and fingerprint recognition.

[0037] Embodiments of this application may also relate to network media technology in artificial intelligence technology. Different from the working methods adopted by traditional audio and video devices, network media relies on the technologies and devices provided by information technology (IT) equipment developers to transmit, store, and process audio and video signals. The traditional serial digital interface (SDI) transmission method lacks true network switching characteristics, and a large amount of work is required to create some of the network functions provided by Ethernet and Internet Protocol (IP) using SDI. Therefore, network media technology in the video industry has emerged. Network media, like traditional media such as television, newspapers, and radio, is a channel for spreading information, a tool for communicating and spreading information, and an information carrier. More specifically, embodiments of this application relate to content consultation or article recommendation technology in network media technology.

[0038] Figure 1 It is a schematic block diagram of the system framework 100 provided by embodiments of this application.

[0039] As Figure 1As shown in the figure, the system framework 100 may include an article recommendation system 110 and an article client 120. The article recommendation system 110 may be a news recommendation system. The article client 120 may be any electronic device with a display function or a display interface. The article recommendation system 110 may be implemented by any electronic device with data processing capabilities. For example, the electronic device may be implemented as a server. The server may be an independent physical server, or a server cluster or a distributed system composed of multiple physical servers. It may also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, and big data and artificial intelligence platforms. The servers may be directly or indirectly connected by wired or wireless communication methods, and this application does not make any restrictions here.

[0040] As Figure 1 shown in the figure, the article recommendation system 110 generally includes four modules, namely, a user portrait module 111, a recall module 112, a ranking (rank) module 113, and a rearrangement (shuffling) module 114. The user portrait module 111 calculates the long-term and short-term interests of the user based on the user's historical behavior, providing basic information for recall and ranking. The recall module 112 mines articles that the user is interested in from a vast amount of articles from different perspectives. The ranking module 113 predicts the click-through rate for the recalled articles based on user information and basic article information. In order to avoid the content of the articles recommended to the user being too single, the rearrangement module 114 collects the articles after the ranking module 113 according to the preset restriction rules and finally presents them to the user. The user browses the articles recommended by the client, and for the article titles and thumbnail images of the article illustrations seen, clicks and further reads according to their own interests. The rearrangement module 114, as the last module in the article recommendation system 110 and also the module closest to the user in the article recommendation system 110, directly interacts with the user and also directly affects the final recommendation result. The rearrangement module 114 is used to ensure the diversity of the displayed articles while preferentially selecting articles that the user may be interested in. Article diversity generally includes article classification, article theme, article style, etc. A poor rearrangement module 114 will affect the user's access to article diversity and information benefits, and at the same time limit the effects of the previous modules, affecting the final recommendation effect, and ultimately resulting in a poor user experience.

[0041] As Figure 1 shown in the figure, the article client 120 may include a display module 121, which is used to display the recommended articles output by the article recommendation system 110. The display module 121 may also be used to count and report the user's browsing behavior to the article recommendation system 110.

[0042] Regarding the working mode of the rearrangement module 114, it is mainly divided into two modes: the streaming processing mode and the batch processing mode.

[0043] The main processing flow of the streaming processing mode is as follows:

[0044] First, all the collected articles are sorted according to the scores predicted by the sorting module. Then, strategies to ensure diversity and experience are formulated, such as controlling the number of first-level categories / second-level categories / tag numbers, and the arrangement rules of article media, such as the ratio of videos to pictures and texts. Next, articles are collected from front to back according to the formulated rules until the number of articles requested by the system this time is collected. Generally, the rules for collecting articles contain multiple rule sets. In the first round of collection, for the empty slots where no articles are collected, a downgraded rule is used for article collection. For example, according to the first-round rules, when traversing all article sets, for the fifth slot, the collection of articles for the fifth slot is skipped and the collection of articles for the sixth slot is continued. In the second round of collection, a looser rule is used to collect the fifth slot again until all the articles requested this time are collected. Finally, the article results are sent to the news client.

[0045] However, in the streaming processing mode, every time an article is collected, it is judged whether there is a conflict with the previous ones. If the requested number cannot be collected, downgraded processing is performed. Every time an article is collected in the streaming processing mode, all articles need to be traversed once. The collection time is strongly correlated with the number of collected articles and the total number of articles, and there are serious collection timeliness problems. In addition, in the streaming processing mode, it is possible that no article can be collected at a certain position, which involves using a downgraded strategy for collection. This will abandon some scattering rules, resulting in a poor effect of the finally collected articles. In other words, in order to meet the requested number of collections, the streaming processing mode generally configures a downgraded strategy, that is, the rearrangement standard will be lowered in the second or more rounds of collection, which will lead to unreasonable displays and affect the experience.

[0046] The main process of the batch processing mode is as follows:

[0047] First, all the collected articles are sorted according to the scores predicted by the sorting module. Then, collection strategies and slot scattering strategies to ensure diversity and experience are formulated. The collection strategies are generally some hard rules, such as the number of articles of the same first-level category collected this time cannot exceed the specified number. The scattering strategy is generally the rule for adjacent slots. For example, in order to ensure the information gain obtained by users, the same second-level category cannot be consecutive. After collecting the number of articles requested this time according to the set collection strategy, sequential fine-tuning is performed according to the slot scattering strategy, such as first ensuring that the same first-level category is not consecutive, and then ensuring that the second-level category is not consecutive. Finally, the fine-tuned article results are sent to the news client.

[0048] However, in the batch processing method, a collection strategy is first adopted to collect a batch of articles, and then a slot scattering strategy is used to serially fine-tune the unreasonable order. The fine-tuning in batch processing is generally carried out by serially executing various slot scattering strategies to achieve fine-tuning of the unreasonable order. That is to say, the first strategy is satisfied first, and then fine-tuning is carried out according to the second strategy. This serial processing is likely to cause the subsequent rules to break the previous rules. Since batch processing needs to fine-tune the unreasonable order within an adjustable range based on various slot scattering strategies as much as possible, it may lead to the situation that the final article recommendation order does not meet the rule with the highest priority, and there is also the problem of poor interpretability. It is very easy to have unreasonable orders, such as adjacent homogeneous content, which reduces the user's experience of obtaining information. Although it ensures that the articles with higher rankings are in the front, it is very easy to have unreasonable orders, such as adjacent homogeneous content, which reduces the user's experience of obtaining information, and then reduces the capabilities of the recall module 112 and the ranking module 113, seriously affecting the user experience effect.

[0049] To solve these problems, this paper proposes a rearrangement method based on the Multiple Hypothesis Tracking (MHT) algorithm. MHT is a classic algorithm for solving object tracking or an algorithm applied in the field of visual tracking. The MHT algorithm retains all hypotheses of the true target and allows them to continue to be transmitted to eliminate the uncertainty of the current scan cycle from the subsequent observation data. In other words, the MHT algorithm will retain all current possibilities and make a final decision when the future uncertainty is determined.

[0050] For the convenience of understanding the solution of this application, the following explains the relevant terms involved in this application.

[0051] News recommendation: Using the principles and methods of the recommendation system, quantitatively mining and expressing the existing and potential interest points of users, and recommending news and information that users are interested in to users.

[0052] Tree: It is a very special data structure, which is a set with a hierarchical relationship composed of n (n≥1) finite nodes. This data structure is like an upside-down tree, that is to say, the root is upward and the leaves are downward.

[0053] Binary tree: In computer science, a binary tree is a tree structure in which each node has at most two subtrees. Usually, the subtrees are called "left subtree" and "right subtree". Binary trees are often used to implement binary search trees and binary heaps. The characteristic of this kind of tree is that the number of nodes on each layer is the maximum number of nodes.

[0054] Full binary tree: If a binary tree has the same structure as the first m nodes of a complete binary tree, such a binary tree is called a full binary tree. In a full binary tree, all levels except the last one are full, or a full binary tree is missing a consecutive number of nodes on the right. The depth of a full binary tree with n nodes is log 2 (n + 1). A full binary tree with depth k has at least 2^k - 1 nodes and at most 2^k - 1 nodes.

[0055] Complete binary tree: A complete binary tree belongs to a type of full binary tree. In a complete binary tree, every node except the leaf nodes has left and right child nodes. That is, a binary tree with depth k and 2^k - 1 nodes is called a complete binary tree.

[0056] Degree of a node: The number of subtrees contained in a node is called the degree of that node.

[0057] Leaf node: A node with degree 0 is called a leaf node.

[0058] Branch node: A node with a non-zero degree.

[0059] Parent node: If a node has child nodes, then this node is called the parent node of the child nodes.

[0060] Child node: If a node has a parent node, then this node is called the child node of the parent node.

[0061] Sibling nodes: Nodes with the same parent node are called sibling nodes.

[0062] Degree of a tree: In a tree, the maximum degree of a node is called the degree of the tree.

[0063] Level of a node: Starting from the root, the root is defined as the first level, the children of the root are the second level, and so on.

[0064] Height or depth of a tree: The maximum level of nodes in a tree.

[0065] Cousin nodes: Nodes whose parents are on the same level are called cousin nodes.

[0066] Ancestors of a node: All nodes on the branches from the root to that node;

[0067] Descendants: Any node in the subtree rooted at a certain node is called a descendant of that node.

[0068] Forest: A set of m (m >= 0) non-intersecting trees is called a forest.

[0069] Figure 2FIG. 200 is a schematic flowchart of an article recommendation method 200 based on the MHT algorithm provided by an embodiment of the present application. It should be noted that the solution provided by the embodiment of the present application can be executed by an article recommendation system. The article recommendation system can be implemented by any electronic device with data processing capabilities. For example, the electronic device can be implemented as a server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, and big data and artificial intelligence platforms. The servers can be directly or indirectly connected through wired or wireless communication methods, and the present application does not limit this. Again, for the sake of avoiding description, the following will take the article recommendation system as an example for illustration.

[0070] As Figure 2 shown, the method 200 may include:

[0071] S210, receiving a recommendation request for requesting T recommended articles, where T > 1;

[0072] S220, in response to the recommendation request, selecting N candidate samples from M collected article samples, where each of the M article samples has a corresponding score for characterizing the matching degree between the user interest and the article sample, and M ≥ N ≥ 1;

[0073] S230, based on the scores of each article sample, selecting K article samples from the N candidate samples and determining the K article samples as K root nodes, where N ≥ K ≥ 1;

[0074] S240, based on the K root nodes, determining the T recommended articles in the following manner:

[0075] Based on each parent node, among the remaining article samples in the M article samples except the article samples already selected in the trajectory where the parent node is located, selecting the top X article samples with a higher benefit score relative to the parent node as the X child nodes of the parent node; taking each of the X child nodes as a new parent node and continuing to extend the child nodes until the level where the leaf nodes are located is T, and determining the T article samples on the target trajectory with the largest benefit score in the trajectories where all leaf nodes are located as the T recommended articles;

[0076] S250, outputting the T recommended articles.

[0077] In short, the article recommendation system selects N candidate samples from M collected article samples, and then selects K article samples from these N candidate samples. Based on these K article samples as K root nodes, T recommended articles to be finally recommended are determined based on the MHT algorithm. Based on this, the article recommendation system can select the articles to be finally recommended in the way of assuming trajectories based on a selected candidate set.

[0078] For example, first, based on the set repetition rule, several candidate sets slightly larger than the number of articles requested this time are selected from all article samples. Then, K article samples are selected as the initial seed nodes for forest extension, that is, these K article samples are used as the root nodes of the assumed trajectories, and then the extension starts from these root nodes. During the extension, based on the collected article samples, the article samples that bring the greatest benefit are selected. For example, by designing the scoring rules for selecting branch nodes during each extension, based on these rules, child nodes are extended for each branch, and the article samples that bring the greatest benefit are selected each time. Finally, when the number of articles requested this time is collected, all current trajectories are sorted, and then the one with the highest score is selected as the final result and passed to the news client.

[0079] In the embodiment of the present application, determining the T recommended articles based on the MHT algorithm can, compared with the serial processing method of batch processing, consider all conditions each time of collection, avoid unreasonable orders, such as adjacent homogeneous content, which reduces the user experience of obtaining information, and thus can ensure the capabilities of the recall module and the sorting module, improving the user experience effect. In addition, compared with the degradation strategy of stream processing, the finally recommended articles are more flexible and controllable by selecting the finally recommended articles through a quantitative scoring method. In addition, based on the K article samples selected from the N candidate samples, T recommended articles are collected using the MHT algorithm, that is, K article samples (i.e., seed nodes) selected from the N candidate samples are used as root nodes to extend child nodes, which can ensure the integrity of trajectory traversal, reduce the time complexity of traversal, is beneficial to improving the effectiveness of the MHT model, and is also beneficial to introducing more complex models; for example, even if a more complex model for calculating benefit scores is introduced, using K article samples selected from the N candidate samples as root nodes to extend child nodes can, on the basis of ensuring the integrity of trajectory traversal, avoid excessive traversal time. In short, the solution provided in the embodiment of the present application can improve the user experience effect, improve the flexibility and controllability of recommendation, ensure the integrity of trajectory traversal, reduce the time complexity of traversal, is beneficial to improving the effectiveness of the MHT model, and is also beneficial to introducing more complex models.

[0080] In some embodiments, each of the X child nodes is used as a new parent node to continue extending child nodes. In one implementation, the level where the X child nodes are located is the target level, and the number of child nodes that can be used as the new parent node in the target level is a preset maximum number Y≥K; when the number of child nodes in the target level is greater than or equal to Y, Y child nodes are selected from the child nodes in the target level; each of the Y child nodes is used as the new parent node to continue extending child nodes. For the extension of the tree structure, since the number of leaf nodes grows exponentially, by setting the maximum number of nodes on each layer and pruning or deleting the branch trajectories when the maximum number is reached, the effectiveness can be guaranteed, that is, when the number of assumed trajectories is greater than the set maximum, the scores of all current trajectories are evaluated, and pruning is performed based on the score ranking, which can avoid the explosive increase in time and space complexity caused by too many assumed trajectories. In one implementation, the Y trajectories with the top benefit scores among all the trajectories where the child nodes in the target level are located are determined; the child nodes on the Y trajectories in the target level are determined as the Y child nodes.

[0081] In some embodiments, before selecting the top X article samples with respect to the parent node as the X child nodes of the parent node from the remaining article samples in the M article samples except the article samples already selected in the trajectory where the parent node is located for each parent node, the benefit score of the first article sample in the remaining article samples with respect to the parent node is determined in the following manner: obtaining a first score for characterizing the matching degree between the first article sample and the user interest; determining the time delay score of the first article sample based on the level where the X child nodes are located; obtaining a second score for characterizing the relationship between the first article sample and the article sample on the parent node; the difference between the value obtained by multiplying the first score by the time delay score and the second score is determined as the benefit score of the first article sample with respect to the parent node. In one implementation, the level where the X child nodes are located is multiplied by a preset delay coefficient to obtain a first delay score; the difference between 1 and the first delay score is determined as the time delay score of the first article sample. In one implementation, before obtaining the second score for characterizing the relationship between the first article sample and the article sample on the parent node, a binary relationship is stored, and the binary relationship includes the scores for characterizing the relationship between any two article samples in the M article samples corresponding to any two article samples; based on this, the score corresponding to the two article samples formed by the first article sample and the article sample on the parent node can be determined as the second score by querying the binary relationship.

[0082] Starting from multiple initial seed nodes, a search forest of multiple hypothetical trajectories is established. During the extension process, the node that brings the greatest benefit can be selected by designing an extension function. The extension function takes into account bonus items and penalty items. The bonus items are used to characterize the relevance between the user and the article. When this article breaks the scattering rule, points will be deducted, which forms the penalty items. The bonus items include two parts. One part is used to characterize the relevance or matching degree between the user and the article. For example, the score of the sorting module can be directly used. The other part is the delay item. On the basis of considering the display delay, it can ensure that the articles with higher scores predicted by the sorting module are preferentially displayed. Articles with higher scores ranked in the front will have higher scores. The penalty items are used to characterize the breaking of the scattering rule, that is, the relationship between articles. By considering 4 types of scattering relationships, namely first-level / second-level classification, tag, display style (big picture / small picture / three pictures / no picture), and medium (picture text / video), corresponding scores will be increased for each broken relationship. In this application, the binary relationships of all articles can be cached in advance to form a distance (Dist) matrix. By querying the distance matrix, the selection of articles can be realized, which can increase the processing speed.

[0083] In one implementation, the above extension function can be expressed as a formula:

[0084] s t = S(u, A t ) * (1 - rt) – Dist(A t-1 , A t ).

[0085] Among them, s t represents the score of the node on the t-th layer, S(u, A t ) represents the relevance or matching degree used to characterize the user u and the article A t , r represents the delay coefficient, and Dist(A t-1 , A t ) is the score used to characterize the relationship between the article sample A t-1 and the article sample A t .

[0086] In some embodiments, before determining the T article samples on the target trajectory with the largest benefit score among the trajectories where all leaf nodes are located as the T recommended articles, in the following manner, the benefit score of the first trajectory among the trajectories where all leaf nodes are located is determined: Obtain the third score used to characterize the matching degree between the article sample on the root node of the first trajectory and the user interest; Determine the benefit score of each node on the first trajectory except the root node relative to the parent node; The sum of the third score and the cumulative sum of the benefit scores of the nodes on the first trajectory except the root node relative to the parent node is determined as the benefit score of the first trajectory.

[0087] For example, by setting the score of the branch trajectory, the sum of the scores calculated during each extension is used as the final trajectory score. When the number of nodes in the trajectory reaches the number of articles requested this time, by evaluating the scores of all trajectories, the optimal trajectory is selected as the result of the final rearrangement module, and then the article list is transmitted to the user client for display. For example, the score of the branch trajectory can be determined by the following formula:

[0088]

[0089] where S represents the score of the branch trajectory, and s i represents the score of the node on the i-th layer of the branch trajectory. T is the height or depth of the tree, that is, the maximum level of the nodes.

[0090] In some embodiments, N candidate samples are selected from the M collected article samples. In one implementation, the N candidate samples that meet the repetition rules are selected from the M article samples; wherein, the repetition rules include at least one of the following rules: the number of video article samples among the N candidate samples is less than or equal to the first quantity, the number of first-level classified articles among the N candidate samples is less than or equal to the second quantity, and the number of second-level classified article samples among the N candidate samples is less than or equal to the third quantity.

[0091] First, obtain the index information of the M article samples, including the first-level classification information, second-level classification information, display style information, etc. of the article samples. Sort the article samples using the scores of all the article samples obtained from the sorting module. Assume that the strategy for the first-round collection is mainly used to ensure the diversity of the article samples, such as ensuring the maximum number of collected videos, the maximum number of first-level classified articles, the maximum number of second-level classified articles, etc., so as to select N candidate samples from the M collected article samples. Collecting article samples using the MHT algorithm based on the selected candidate set can reduce the time complexity of traversal, is beneficial to improving the effectiveness of the MHT algorithm, and is also beneficial to introducing more complex models.

[0092] In some embodiments, N≥T.

[0093] In some embodiments, the structure formed by the trajectories where all the leaf nodes are located is a binary tree structure.

[0094] Figure 3 This is an example of the article recommendation method based on the MHT algorithm provided by the embodiments of the present application.

[0095] Such as Figure 3As shown in the figure, assume that 3 nodes are selected as seed nodes in the first layer to start the extension. At the (i - 1)-th layer, there are 4 nodes at this time. For each node, 2 nodes are selected as child nodes based on the extension function. Assume that the set maximum number of nodes is reached at this time. Then, based on the scores of each trajectory at this time, a set number of nodes are selected for the next extension, and the remaining ones are pruned. Assume that the number of requested articles this time is T. When reaching the T-th layer, evaluate all the trajectories at this time, and select the one with the highest score as the final result.

[0096] In summary, for the solution provided by the embodiments of the present application, by reconstructing the collection and scattering methods of the rearrangement module, the circumscription of N candidate samples reduces the time consumption of the entire module. The tree-shaped search mode with multiple child nodes realizes an approximate full permutation search method, and the pruning strategy ensures high operation efficiency. The setting of the extension function takes into account both the relevance between the user and the article and the relationship between articles. The finally obtained optimal trajectory comprehensively considers the overall situation of the entire trajectory, can better mine the articles that the user is interested in, improve the information acquisition of the user when reading articles, and improve the user experience.

[0097] The preferred embodiments of the present application have been described in detail above with reference to the accompanying drawings. However, the present application is not limited to the specific details in the above embodiments. Within the technical concept scope of the present application, various simple modifications can be made to the technical solutions of the present application, and these simple modifications all belong to the protection scope of the present application. For example, for the various specific technical features described in the above specific embodiments, they can be combined in any appropriate manner without contradiction. To avoid unnecessary repetition, the present application does not separately explain various possible combination methods. Also, for example, any combination can be made between various different embodiments of the present application as long as it does not violate the idea of the present application, and it should also be regarded as the content disclosed by the present application.

[0098] For example, in the article search stage, a tree-shaped search structure is used. Using other search methods can also achieve the same function and should be included in the protection scope of this article. Again, for example, this article takes the article recommendation scenario as an example. For other recommendation scenarios, the methods of this article can also be used and should also be included in the protection scope. For example, the video recommendation scenario or even the shopping recommendation scenario.

[0099] It should also be understood that in various method embodiments of the present application, the magnitudes of the serial numbers of the above processes do not mean the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0100] The method provided by the embodiments of the present application has been described above. Next, the device provided by the embodiments of the present application will be described.

[0101] Figure 4It is a schematic block diagram of an article recommendation device 300 based on the MHT algorithm provided by an embodiment of the present application.

[0102] As Figure 4 shown, the article recommendation device 300 may include:

[0103] A receiving unit 310, configured to receive a recommendation request for requesting T recommended articles, where T≥1;

[0104] A selection unit 320, configured to, in response to the recommendation request, select N candidate samples from M collected article samples, where each of the M article samples has a corresponding score for characterizing the matching degree between user interests and the article sample, and M≥N≥1;

[0105] A determination unit 330, configured to:

[0106] Based on the scores of each article sample, select K article samples from the N candidate samples and determine the K article samples as K root nodes, where N≥K≥1;

[0107] Based on the K root nodes, determine the T recommended articles in the following manner:

[0108] Based on each parent node, among the remaining article samples in the M article samples except the article samples already selected in the track where the parent node is located, select X article samples with the top-ranked benefit scores relative to the parent node as the X child nodes of the parent node; use each of the X child nodes as a new parent node to continue extending child nodes until the level where the leaf nodes are located is T, and determine the T article samples on the target track with the largest benefit scores in the tracks where all leaf nodes are located as the T recommended articles;

[0109] An output unit 340, configured to output the T recommended articles.

[0110] In some embodiments, the level where the X child nodes are located is the target level, and the number of child nodes that can be used as the child nodes of the new parent node in the target level is a preset maximum number Y≥K; where the determination unit 330 is specifically configured to:

[0111] When the number of child nodes in the target level is greater than or equal to Y, select Y child nodes from the child nodes in the target level; use each of the Y child nodes as the new parent node to continue extending child nodes.

[0112] In some embodiments, the determination unit 330 is specifically configured to:

[0113] Determine the top Y tracks with the largest benefit scores in the tracks where all child nodes in the target level are located;

[0114] Determine the child nodes on the Y trajectories in the target level as the Y child nodes.

[0115] In some embodiments, before the determining unit 330 selects, for each parent node, X article samples with relatively high benefit scores relative to the parent node as the X child nodes of the parent node from the remaining article samples in the M article samples except for the article samples already selected in the trajectory where the parent node is located, the determining unit 330 is further configured to:

[0116] Determine the benefit score of the first article sample in the remaining article samples relative to the parent node in the following manner:

[0117] Obtain a first score for characterizing the matching degree between the first article sample and the user interest;

[0118] Based on the level where the X child nodes are located, determine the time delay score of the first article sample;

[0119] Obtain a second score for characterizing the relationship between the first article sample and the article sample on the parent node;

[0120] Determine the difference between the value obtained by multiplying the first score by the time delay score and the second score as the benefit score of the first article sample relative to the parent node.

[0121] In some embodiments, the determining unit 330 is specifically configured to;

[0122] Multiply the level where the X child nodes are located by a preset delay coefficient to obtain a first delay score;

[0123] Determine the difference between 1 and the first delay score as the time delay score of the first article sample.

[0124] In some embodiments, the determining unit 330 is specifically configured to:

[0125] Store a binary relationship, where the binary relationship includes scores for characterizing the relationship between any two article samples in the M article samples;

[0126] By querying the binary relationship, determine the score corresponding to the two article samples formed by the first article sample and the article sample on the parent node as the second score.

[0127] In some embodiments, before the determining unit 330 determines the T article samples on the target trajectory with the largest benefit score in the trajectories where all leaf nodes are located as the T recommended articles, the determining unit 330 is further configured to:

[0128] Determine the revenue score of the first trajectory in the trajectory where all these leaf nodes are located in the following manner:

[0129] Obtain a third score used to characterize the matching degree between the article sample on the root node of the first trajectory and the user interest;

[0130] Determine the revenue score of each node other than the root node on the first trajectory relative to its parent node;

[0131] Determine the sum of the third score and the cumulative sum of the revenue scores of the nodes other than the root node on the first trajectory relative to their parent nodes as the revenue score of the first trajectory.

[0132] In some embodiments, the selection unit 320 is specifically configured to:

[0133] Select the N candidate samples that meet the repetition rule from the M article samples;

[0134] Wherein, the repetition rule includes at least one of the following rules:

[0135] The number of video-type article samples among the N candidate samples is less than or equal to a first number, the number of first-level classification articles among the N candidate samples is less than or equal to a second number, and the number of second-level classification article samples among the N candidate samples is less than or equal to a third number.

[0136] In some embodiments, N≥T.

[0137] In some embodiments, the structure formed by the trajectories where all these leaf nodes are located is a binary tree structure.

[0138] It should be understood that the device embodiments and the method embodiments can correspond to each other, and similar descriptions can refer to the method embodiments. To avoid repetition, they are not elaborated here. Specifically, the device 300 can correspond to the corresponding main body in the method 200 of the embodiments of the present application, and each unit in the device 300 is respectively for implementing the corresponding processes in the method 200. For the sake of brevity, they are not elaborated here.

[0139] It should also be understood that each unit in the video processing apparatus according to the embodiments of the present application can be separately or wholly combined into one or several other units to form, or some of the units can be further split into multiple smaller units with more specific functions to form, which can achieve the same operations without affecting the realization of the technical effects of the embodiments of the present application. The above units are divided based on logical functions. In practical applications, the function of one unit can also be realized by multiple units, or the functions of multiple units can be realized by one unit. In other embodiments of the present application, the video processing apparatus can also include other units. In practical applications, these functions can also be assisted by other units and can be realized by the cooperation of multiple units. According to another embodiment of the present application, the video processing apparatus according to the embodiments of the present application can be constructed by running a computer program (including program code) capable of executing the respective steps involved in the corresponding method on a general computing device of a general computer including processing elements and storage elements such as a central processing unit (CPU), a random access storage medium (RAM), a read-only storage medium (ROM), etc., and the video processing method according to the embodiments of the present application can be realized. The computer program can be recorded on, for example, a computer-readable storage medium, loaded into an electronic device through the computer-readable storage medium, and run therein to implement the corresponding method according to the embodiments of the present application.

[0140] In other words, the units described above can be implemented in the form of hardware, can also be implemented by instructions in the form of software, or can be implemented in a form combining software and hardware. Specifically, the respective steps of the method embodiments in the embodiments of the present application can be completed by the integrated logic circuit in the hardware in the processor and / or instructions in the form of software. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as being executed and completed by a hardware decoding processor, or executed and completed by a combination of the hardware and software in the decoding processor. Optionally, the software can be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps in the above method embodiments.

[0141] Figure 5 It is a schematic structural diagram of the electronic device 400 provided by the embodiments of the present application.

[0142] As Figure 5As shown, the electronic device 400 includes at least a processor 410 and a computer-readable storage medium 420. Among them, the processor 410 and the computer-readable storage medium 420 can be connected by a bus or other means. The computer-readable storage medium 420 is used to store a computer program 421, and the computer program 421 includes computer instructions. The processor 410 is used to execute the computer instructions stored in the computer-readable storage medium 420. The processor 410 is the computing core and control core of the electronic device 400, and it is suitable for implementing one or more computer instructions, specifically for loading and executing one or more computer instructions to implement the corresponding method flow or corresponding function.

[0143] As an example, the processor 410 can also be referred to as a Central Processing Unit (CPU). The processor 410 may include, but is not limited to: a general-purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and so on.

[0144] As an example, the computer-readable storage medium 420 can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory; optionally, it can also be at least one computer-readable storage medium located far from the aforementioned processor 410. Specifically, the computer-readable storage medium 420 includes, but is not limited to: volatile memory and / or non-volatile memory. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synch link dynamic random access memory (SLDRAM), and direct rambus random access memory (DR RAM).

[0145] As Figure 5 shown, the electronic device 400 may further include a transceiver 430. Among them, the processor 410 can control the transceiver 430 to communicate with other devices. Specifically, it can send information or data to other devices, or receive information or data sent by other devices. The transceiver 430 can include a transmitter and a receiver. The transceiver 430 may further include an antenna, and the number of antennas can be one or more.

[0146] It should be noted that the various components in the electronic device 400 are connected through a bus system. Among them, the bus system may include, in addition to the data bus, a power bus, a control bus, a status signal bus, etc. The embodiments of the present application do not make specific limitations on this.

[0147] In one implementation, the electronic device 400 can be Figure 4The article recommendation device 200 based on the MHT algorithm shown; computer instructions are stored in the computer-readable storage medium 420; the computer instructions stored in the computer-readable storage medium 420 are loaded and executed by the processor 410 to implement Figure 4 the corresponding steps in the method embodiments shown; in specific implementation, the computer instructions in the computer-readable storage medium 420 are loaded and executed by the processor 410 to perform the corresponding steps. To avoid repetition, they are not elaborated here.

[0148] According to another aspect of the present application, an embodiment of the present application further provides a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in the electronic device 400 and is used to store programs and data. For example, the computer-readable storage medium 420. It can be understood that the computer-readable storage medium 420 here can include both the built-in storage medium in the electronic device 400 and, of course, the extended storage medium supported by the electronic device 400. The computer-readable storage medium provides a storage space, and this storage space stores the operating system of the electronic device 400. And, one or more computer instructions suitable for being loaded and executed by the processor 410 are stored in this storage space, and these computer instructions can be one or more computer programs 421 (including program codes).

[0149] According to another aspect of the present application, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions, and these computer instructions are stored in a computer-readable storage medium. For example, the computer program 421. At this time, the electronic device 400 can be a computer. The processor 410 reads the computer instructions from the computer-readable storage medium 420, and the processor 410 executes the computer instructions, so that the computer executes the video processing method provided in the above various optional manners.

[0150] In other words, when implemented using software, it can be implemented in the form of a computer program product in whole or in part. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes of the embodiments of the present application are run in whole or in part, or the functions of the embodiments of the present application are implemented. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from a website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wirelessly (such as infrared, wireless, microwave, etc.).

[0151] Those of ordinary skill in the art can realize that the units and process steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0152] Finally, it should be noted that the above description is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An article recommendation method based on the Multiple Hypothesis Tracking (MHT) algorithm, characterized in that, it includes: Receiving a recommendation request for requesting T recommended articles, where T > 1; In response to the recommendation request, selecting N candidate samples from M collected article samples, and each of the M article samples has a corresponding score for characterizing the matching degree between the user interest and the article sample, where M ≥ N ≥ 1; Based on the scores of each article sample, selecting K article samples from the N candidate samples and determining the K article samples as K root nodes, where N ≥ K ≥ 1; Based on the K root nodes, determining the T recommended articles in the following manner: Based on each parent node, among the remaining article samples in the M article samples except the article samples already selected in the trajectory where the parent node is located, selecting the top X article samples with a higher benefit score relative to the parent node as the X child nodes of the parent node; taking each of the X child nodes as a new parent node and continuing to extend the child nodes until the level where the leaf node is located is T, and determining the T article samples on the target trajectory with the maximum benefit score in the trajectories where all leaf nodes are located as the T recommended articles; Outputting the T recommended articles; Before the step of, based on each parent node, among the remaining article samples in the M article samples except the article samples already selected in the trajectory where the parent node is located, selecting the top X article samples with a higher benefit score relative to the parent node as the X child nodes of the parent node, the method further includes: Determining the benefit score of the first article sample in the remaining article samples relative to the parent node in the following manner: Obtaining a first score for characterizing the matching degree between the first article sample and the user interest; Based on the level where the X child nodes are located, determining the time delay score of the first article sample; Obtaining a second score for characterizing the relationship between the first article sample and the article sample on the parent node; Determining the difference between the value obtained by multiplying the first score by the time delay score and the second score as the benefit score of the first article sample relative to the parent node; Before the step of determining the T article samples on the target trajectory with the maximum benefit score in the trajectories where all leaf nodes are located as the T recommended articles, the method further includes: Determining the benefit score of the first trajectory among the trajectories where all leaf nodes are located in the following manner: Obtaining a third score for characterizing the matching degree between the article sample on the root node of the first trajectory and the user interest; Determining the benefit score of each node except the root node on the first trajectory relative to the parent node; Determining the sum of the third score and the cumulative sum of the benefit scores of the nodes except the root node on the first trajectory relative to the parent node as the benefit score of the first trajectory.

2. The method according to claim 1, characterized in that, The level where the X child nodes are located is the target level, and the number of child nodes in the target level that can be used as the new parent node is a preset maximum number Y≥K; Among them, the step of using each of the X child nodes as a new parent node to continue extending child nodes includes: When the number of child nodes in the target level is greater than or equal to Y, select Y child nodes from the child nodes in the target level; use each of the Y child nodes as the new parent node to continue extending child nodes.

3. The method according to claim 2, wherein, The step of selecting Y child nodes from the child nodes in the target level includes: Determine the top Y trajectories with the highest benefit scores among the trajectories where all child nodes in the target level are located; Determine the child nodes on the Y trajectories in the target level as the Y child nodes.

4. The method according to claim 1, wherein, The step of determining the time delay score of the first article sample based on the level where the X child nodes are located includes: Multiply the level where the X child nodes are located by a preset delay coefficient to obtain a first delay score; Determine the difference between 1 and the first delay score as the time delay score of the first article sample.

5. The method according to claim 1, wherein, Before obtaining the second score for characterizing the relationship between the first article sample and the article sample on the parent node, the method further includes: Store a binary relationship, where the binary relationship includes the scores for characterizing the relationship between any two article samples among the M article samples; Among them, the step of obtaining the second score for characterizing the relationship between the first article sample and the article sample on the parent node includes: By querying the binary relationship, determine the score corresponding to the two article samples formed by the first article sample and the article sample on the parent node as the second score.

6. The method according to claim 1, wherein, The step of selecting N candidate samples from the collected M article samples includes: Select the N candidate samples that meet the repetition rule from the M article samples; Among them, the repetition rule includes at least one of the following rules: The number of video-type article samples among the N candidate samples is less than or equal to a first number, the number of first-level classification articles among the N candidate samples is less than or equal to a second number, and the number of second-level classification article samples among the N candidate samples is less than or equal to a third number.

7. The method according to claim 1, wherein, N≥T.

8. The method according to any one of claims 1 to 7, wherein, The structure formed by the trajectories where all leaf nodes are located is a binary tree structure.

9. An article recommendation device based on the Multiple Hypothesis Tracking (MHT) algorithm, wherein, It includes: A receiving unit, configured to receive a recommendation request for requesting T recommended articles, where T≥1; A selection unit, configured to select N candidate samples from the collected M article samples in response to the recommendation request, where each of the M article samples has a corresponding score for characterizing the matching degree between the user interest and the article sample, and M≥N≥1; A determination unit, configured to: Based on the scores of each article sample, select K article samples from the N candidate samples and determine the K article samples as K root nodes, where N≥K≥1; Based on the K root nodes, determine the T recommended articles in the following manner: Based on each parent node, among the remaining article samples in the M article samples except the article samples already selected in the trajectory where the parent node is located, select X article samples with the top-ranked benefit scores relative to the parent node as the X child nodes of the parent node; use each of the X child nodes as a new parent node to continue extending child nodes until the level where the leaf nodes are located is T, and determine the T article samples on the target trajectory with the maximum benefit score in the trajectories where all leaf nodes are located as the T recommended articles; An output unit, configured to output the T recommended articles; Before the step of, based on each parent node, selecting X article samples with the top-ranked benefit scores relative to the parent node as the X child nodes of the parent node among the remaining article samples in the M article samples except the article samples already selected in the trajectory where the parent node is located, the determination unit is further configured to: Determine the benefit score of the first article sample in the remaining article samples relative to the parent node in the following manner: Obtain a first score for characterizing the matching degree between the first article sample and the user interest; Based on the level where the X child nodes are located, determine the time delay score of the first article sample; Obtain a second score for characterizing the relationship between the first article sample and the article sample on the parent node; Determine the difference between the value obtained by multiplying the first score by the time delay score and the second score as the benefit score of the first article sample relative to the parent node; Before the step of determining the T article samples on the target trajectory with the maximum benefit score in the trajectories where all leaf nodes are located as the T recommended articles, the determination unit is further configured to: Determine the benefit score of the first trajectory among the trajectories where all leaf nodes are located in the following manner: Obtain a third score for characterizing the matching degree between the article sample on the root node of the first trajectory and the user interest; Determine the benefit score of each node except the root node on the first trajectory relative to the parent node; Determine the sum of the third score and the cumulative sum of the benefit scores of the nodes except the root node on the first trajectory relative to the parent node as the benefit score of the first trajectory.

10. An electronic device, characterized in that it includes: a processor, adapted to execute a computer program; a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by the processor, the method according to any one of claims 1 to 8 is implemented.

11. A computer-readable storage medium, characterized in that, for storing a computer program, the computer program causing a computer to execute the method according to any one of claims 1 to 8.

Images