Graph-based index construction method, article recommendation method, computing device, computer readable storage medium and computer program product

Abstract

The embodiment of the invention provides a graph-based index construction method. The graph-based index construction method comprises the following steps: acquiring an initial index graph comprising user nodes and article nodes; based on the connection relationship between each user node and each article node in the initial index map, generating a reference index map including the article nodes, the reference index map including the connection probability between the article nodes; and updating the reference index map according to the connection probability between the article nodes, and obtaining a target index map corresponding to the reference index map. According to the method, the number of connection relations can be reduced on the premise of ensuring relatively high connectivity among the item nodes in the index map, so that the index size (memory use) is reduced to improve the scalability. And the accuracy and the efficiency of searching the article nodes are balanced.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technology, and in particular to a graph-based index construction method, an item recommendation method, a computing device, a computer-readable storage medium, and a computer program product. Background Art

[0002] With the development of computer technology, the rapid growth of available information has posed a significant challenge to modern recommendation systems. Current recommendation systems typically adopt a multi-stage, funnel-like architecture, consisting of a recall phase and a ranking phase. The recall phase typically uses vector retrieval to quickly retrieve items from a larger corpus. The ranking phase then sorts the recalled items based on user preferences and features. However, this approach, based on vector retrieval, consumes significant resources. Furthermore, if there is a loss of precision in the recall phase, errors will be amplified in the subsequent ranking phase, affecting the final recommendation results.

[0003] Based on this, researchers proposed a single-stage recommendation system based on bipartite graph retrieval. This system, implemented using a neural network model and indexing structure, constructs a user-item bipartite graph. This bipartite graph is then used for retrieval and recommendation, enabling the direct retrieval of top-k lists from larger corpora. However, in actual online searches, this bipartite graph still exhibits high search complexity and computational overhead. The challenge for researchers in this single-stage recommendation system is to further reduce this complexity and computational overhead. Summary of the Invention

[0004] In view of this, embodiments of this specification provide a graph-based index construction method. One or more embodiments of this specification also relate to an item recommendation method, a computing device, a computer-readable storage medium, and a computer program to address technical deficiencies in the prior art.

[0005] According to a first aspect of an embodiment of this specification, a method for constructing a graph-based index is provided, comprising:

[0006] Obtain an initial index graph including user nodes and item nodes;

[0007] generating a reference index graph including item nodes based on the connection relationship between each user node and each item node in the initial index graph, wherein the reference index graph includes connection probabilities between item nodes;

[0008] The reference index graph is updated according to the connection probability between each item node to obtain a target index graph corresponding to the reference index graph.

[0009] According to a second aspect of the embodiments of this specification, a method for recommending items is provided, including:

[0010] Obtain reference item information corresponding to the target user;

[0011] Determining an initial item node and at least one to-be-recommended item node corresponding to the initial item node in a target index graph based on the reference item information, wherein the target index graph is constructed using the graph-based index construction method described above;

[0012] At least one target recommended item is determined in each of the item nodes to be recommended according to the connection probability between the initial item node and each of the item nodes to be recommended.

[0013] According to a third aspect of an embodiment of this specification, a computing device is provided, including:

[0014] memory and processor;

[0015] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the above-mentioned graph-based index construction method or item recommendation method are implemented.

[0016] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores computer-executable instructions, which, when executed by a processor, implement the steps of the above-mentioned graph-based index construction method or item recommendation method.

[0017] According to a fifth aspect of the embodiments of this specification, a computer program product is provided, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned graph-based index construction method or item recommendation method.

[0018] The graph-based index construction method provided in this specification extracts the connection relationships and connection probabilities between item nodes from a heterogeneous bipartite graph structure including user nodes and item nodes. By repeatedly updating the connection probabilities between item nodes and updating the reference index graph based on the connection probabilities between item nodes, the number of out-degrees of each item node in the index graph is reduced, thereby reducing the average out-degree of the graph as a whole. While ensuring high connectivity between item nodes in the index graph, the number of connection relationships can be reduced, thereby reducing the index size (memory usage) and improving scalability. A balance is achieved between the accuracy and efficiency of searching for item nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flowchart of a graph-based index construction method provided by one embodiment of this specification;

[0020] Figure 2 is a schematic diagram of an initial index map provided by an embodiment of this specification;

[0021] Figure 3 This is a schematic diagram of the connection relationship between various item nodes provided in one embodiment of this specification;

[0022] Figure 4 is a schematic diagram of a target index map provided by one embodiment of this specification;

[0023] Figure 5 This is a flowchart of an item recommendation method provided by one embodiment of this specification;

[0024] Figure 6 This is a schematic diagram of the structure of a graph-based index construction device provided by an embodiment of this specification;

[0025] Figure 7 This is a schematic diagram of the structure of an item recommendation device provided by one embodiment of this specification;

[0026] Figure 8 This is an architectural diagram of an item recommendation system provided by one embodiment of this specification;

[0027] Figure 9 This is a structural block diagram of a computing device provided by one embodiment of this specification. DETAILED DESCRIPTION

[0028] The following description sets forth many specific details to facilitate a thorough understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0029] The terms used in one or more embodiments of this specification are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of this specification. The singular forms "a," "the," and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0030] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0031] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0032] First, the terms involved in one or more embodiments of this specification are explained.

[0033] Personalized recommendation system: Personalized recommendation system is an advanced business intelligence platform based on massive data mining to help e-commerce websites provide their customers with fully personalized decision support and information services.

[0034] Recall Layer: This layer is primarily used to narrow the scope of product calculations, selecting products of interest to users from a pool of millions. Using simple models and algorithms, this pool is reduced to a few hundred or even dozens. This allows users to receive rapid product feedback with millisecond latency.

[0035] Ranking layer: The ranking layer aims to achieve accurate ranking results. It is the key to the effectiveness of recommendation systems and the core of applications such as deep learning. The recall layer accurately ranks hundreds of items, assigning each item a score based on rules and sorting them from high to low. Due to the required accuracy, the ranking layer model is generally more complex and requires more features.

[0036] Bipartite graphs: Bipartite graphs, also known as bipartite graphs, are a special model in graph theory. Let G = (V, E) be an undirected graph. If the vertices V can be split into two disjoint subsets (A, B), and each edge (i, j) in the graph connects two vertices i and j belonging to these two different vertex sets (i in A, j in B), then the graph G is called a bipartite graph.

[0037] Vector retrieval: Vector retrieval is to retrieve K vectors close to the query vector in a given vector dataset according to a certain metric (K-Nearest Neighbor, KNN). However, due to the high computational complexity of KNN, we usually only focus on the approximate nearest neighbor (ANN) problem.

[0038] Graph retrieval: Based on the characteristics of the graph structure, it walks and searches on the graph, and retrieves the K nodes most relevant to the query node according to the distance or correlation between the nodes.

[0039] NN-Distance: This extends traditional ANN search to neural network models. Any neural network function can be designed as a ranking function. The ranking function's score is used as a metric, known as Neural Network-Based Distance (NN-Distance).

[0040] In this specification, a graph-based index construction method is provided. This specification also involves an item recommendation method, a computing device, a computer-readable storage medium, and a computer program, which are described in detail one by one in the following embodiments.

[0041] See also Figure 1 , Figure 1 A flowchart of a graph-based index construction method provided by an embodiment of this specification is shown, which specifically includes the following steps:

[0042] Step 102: Obtain an initial index graph including user nodes and item nodes.

[0043] Among them, the initial index graph specifically refers to a directed index graph including user nodes and item nodes. The method provided in the embodiment of this specification is based on a single-stage recommendation system (SSR, SingleShotRecommendation). The single-stage recommendation system breaks the limitation of the multi-stage "funnel" structure of the traditional recommendation system. In the SSR recommendation system, the recommendation step includes one stage, namely the retrieval process. The retrieval process involves a scoring model for outputting the correlation score between users and items and a graph-based index. The graph here is the initial index graph including user nodes and item nodes. The scoring model can better model the complex relationship between users and items. The graph-based index measures the distance between users and items by the correlation score between users and items. Through one stage, the final list of the top K recommended items can be retrieved from millions of items. This reduces the complexity of the recommendation system, saves a lot of computing resources and training costs, achieves higher search efficiency and recommendation accuracy, and ensures the recommendation effect of the recommendation system.

[0044] In one or more embodiments of this specification, items include recommended objects in various recommendation scenarios, including but not limited to various tangible or intangible items, such as commodities, advertisements, software products, information, data, etc. Users can be understood as any person, organization, enterprise, etc. who requires recommendation services. In this specification, users and items are not specifically defined and are subject to actual application.

[0045] In a specific embodiment provided in this specification, obtaining an initial index graph including user nodes and item nodes includes:

[0046] Randomly select user nodes or item nodes and insert them into the graph;

[0047] If a user node is selected, a preset number of item nodes with the highest correlation with the user node are selected from the item node set and inserted into the graph;

[0048] If an item node is selected, a preset number of user nodes with the highest correlation with the item node are selected from the user node set and inserted into the graph;

[0049] Connecting the selected user node with some or all of the preset number of item nodes that have the highest correlation with the user node, and connecting the selected item node with some or all of the preset number of user nodes that have the highest correlation with the item node;

[0050] Until all user nodes and item nodes are inserted into the graph and links are established, an initial index graph including user nodes and item nodes is obtained.

[0051] Furthermore, the method further comprises:

[0052] Determine the relevance score between users and items based on a pre-trained scoring model;

[0053] A preset number of item nodes with the highest correlation with the user node and a preset number of user nodes with the highest correlation with the item node are determined according to the correlation scores between the user and the item.

[0054] In practical applications, an initial index graph can be pre-created and then retrieved. During the creation of the initial index graph, which includes user nodes and item nodes, multiple user nodes and multiple item nodes are first determined, and each user node and item node are sequentially inserted into the initial index graph. The predetermined number of item nodes with the highest correlations with the user nodes, and the predetermined number of user nodes with the highest correlations with the item nodes, are determined based on the correlation scores between users and items output by a pre-trained scoring model.

[0055] Select user nodes or item nodes from multiple user nodes and multiple item nodes in turn, and insert the item nodes or user nodes into the initial index graph according to the association score, and repeat this action until all user nodes and item nodes have been inserted into the initial index graph and connections are created between each user node and item node.

[0056] See also Figure 2 , Figure 2 A schematic diagram of an initial index map provided by an embodiment of this specification is shown. Figure 2 As shown, nodes 0-6 are item nodes, and nodes AD are user nodes.

[0057] Step 104: Based on the connection relationship between each user node and each item node in the initial index graph, generate a reference index graph including item nodes, wherein the reference index graph includes connection probabilities between item nodes.

[0058] After determining the initial index graph, a reference index graph consisting solely of item nodes is generated based on the connection relationships between each user node and each item node. Specifically, in the method provided herein, user nodes are removed from the initial index graph, leaving only the user nodes. This generates a reference index graph consisting solely of item nodes, and the reference index graph includes the connection probabilities of the connection relationships between item nodes.

[0059] Specifically, the reference index graph includes the connection relationships between item nodes and the corresponding connection probabilities for each connection relationship. The connection relationships between item nodes can be determined based on the initial index graph. Based on the connection relationships between nodes, the corresponding connection probabilities are calculated.

[0060] Specifically, in a specific embodiment provided in this specification, based on the connection relationship between each user node and each item node in the initial index graph, generating a reference index graph including item nodes includes:

[0061] Randomly selecting a target item node, a reference user node corresponding to the target item node, and a reference item node corresponding to the reference user node in the initial index graph;

[0062] Connecting the target item node and each reference item node, and calculating the connection probability between each item node based on the connection relationship between the target item node and each reference item node;

[0063] Each item node in the initial index graph is traversed until a connection relationship is established between the item nodes, thereby obtaining a reference index graph including the item nodes.

[0064] In this embodiment, a target item node is randomly selected from the initial index graph. Specifically, the target item node refers to any item node. After the target item node is determined, the corresponding reference user node and the corresponding reference item node are also determined. The target item node and the reference item node are then connected.

[0065] See also Figure 3 , Figure 3 A schematic diagram showing the connection relationship between various item nodes provided in an embodiment of this specification is shown. Figure 3 The connection relationship diagram shown is based on Figure 2 As shown in the schematic diagram of the initial index graph, for example, item node 0 is the target item node. It is connected to user A and user B, which are the reference user nodes. User A is connected to item nodes 1 and 2, and user B is connected to item nodes 3 and 4. Item nodes 1, 2, 3, and 4 are the reference item nodes. The target item node is connected to each reference item node, i.e., item node 0 is connected to item nodes 1, 2, 3, and 4.

[0066] For item node 1, it is connected to user node A and user node C, user node A is connected to item node 2, and user node C is connected to item node 5 and item node 6. Then item node 1 is connected to item node 2, item node 5, and item node 6.

[0067] By analogy, we can Figure 2 The schematic diagram of the initial index map shown in Figure 3 The connection relationship diagram between the item nodes shown in Figure 3 The schematic diagram shown only includes the connection relationships between the item nodes, and the connection probability corresponding to each connection relationship has not been calculated.

[0068] After establishing connections between the target item node and each reference item node, the connection probabilities between the item nodes are calculated based on the connection relationships between the target item node and each reference item node. After calculating the connection probabilities between the item nodes, a reference index map is obtained. The reference index map specifically refers to the transitional item index map between the initial index map and the final target index map. The reference index map can be understood as an item index map that has not yet been fully updated.

[0069] In a specific embodiment provided herein, after obtaining the connection relationship between the target item and the reference item node, the connection relationship between the target item and each reference item node is further calculated. Specifically, the connection probability between each item node is calculated based on the connection relationship between the target item node and each reference item node, including:

[0070] Determine the target item node and the target reference item node;

[0071] Determine whether the target item node has adjacent item nodes;

[0072] If not, determining the probability between the target item node and the target reference item node is 1;

[0073] If so, the connection probability between the target item node and the target reference item node is determined according to the adjacent item nodes and the target reference item node.

[0074] In practical applications, to calculate the connection probability between a target item node and each reference item node, one must first determine a target reference item node. It is also necessary to determine whether the target item node currently has any neighboring item nodes. In the methods provided herein, a neighboring item node specifically refers to a reference item node that has a connection relationship with the target item node and for which the connection probability corresponding to the connection relationship has been determined.

[0075] See also Figure 3 For example, consider item node 0 as the target item node. Its corresponding reference item nodes are item node 1, item node 2, item node 3, and item node 4. In this case, item node 0 only has connections with each reference item node, but the connection probabilities for each connection have not yet been determined. For ease of understanding, Table 1 below represents item node 0 establishing connections with item nodes 1, 2, 3, and 4, but without determining the connection probabilities.

[0076] Table 1

[0077] Item Node 1 Item Node 2 Item Node 3 Item Node 4 Item Node 0

[0078] Item node 1 is determined as the target reference item node. Whether item node 0 has any adjacent item nodes is determined using Table 1. Since no connection probability has yet been determined between the reference item node and item node 0, the connection probability between item node 1 and item node 0 can be determined to be 1. Table 1 is then updated to produce Table 2.

[0079] Table 2

[0080] Item Node 1 Item Node 2 Item Node 3 Item Node 4 Item Node 0 1

[0081] Item node 2 is selected as the target reference item node again. Now, we determine whether item node 0 has any adjacent item nodes. Table 2 shows that item node 1 is the adjacent item node. In this case, we cannot directly determine the connection probability between item node 2 and item node 0 to be 1. Instead, we need to further calculate the connection probability between item node 0 and item node 2 based on item node 2 and item node 1.

[0082] Furthermore, determining a connection probability between a target item node and the target reference item node based on the adjacent item nodes and the target reference item node includes:

[0083] Determining whether the target reference item node is a neighbor node of the adjacent item node;

[0084] If not, determining the probability between the target item node and the target reference item node is 1;

[0085] If so, the connection probability between the target item node and the target reference item node is calculated based on the connection probability between the adjacent item node and the target item node.

[0086] In the method provided in this specification, the index graphs are all directed graphs. If node b can be reached from node a, then node b can be considered to be a neighbor node of node a.

[0087] Furthermore, after determining the adjacent item nodes, it is necessary to determine whether the target reference item node is a neighbor node of the adjacent item node. The method provided in this specification, in order to reduce the connection relationship between each item node, tries to ensure that the nodes that were originally reachable in one hop are reachable in two hops on the updated index graph. For example, for node 1, node 2 and node 3, node 1 is connected to node 2, node 1 is connected to node 3, and node 2 is connected to node 3. Node 3 can be reached directly from node 1, or node 1 can first reach node 2 and then reach node 3. The probability of node 1 reaching node 3 directly is reduced.

[0088] Based on this, in the process of determining the connection probability between the target item node and the target reference item node according to the adjacent item node and the target reference item node, it is necessary to first determine whether the target reference item node is a neighbor node of the adjacent item node, that is, to determine whether there is a connection relationship from the adjacent item node to the target reference item node.

[0089] If the target reference item node is not a neighbor node of the adjacent item node, it means that the target reference item node cannot be reached from the target item node through the adjacent item nodes. Therefore, the probability between the target item node and the target reference item node can be determined to be 1.

[0090] If the target reference item node is a neighbor node of an adjacent item node, it means that the target reference item node can be reached from the target item node through the adjacent item node. In this case, the connection probability between the target item node and the target reference item node needs to be calculated based on the connection probability between the adjacent item node and the target item node.

[0091] Furthermore, calculating the connection probability between the target item node and the target reference item node based on the connection probability between the adjacent item nodes and the target item node includes:

[0092] Obtaining an initial connection probability between the target item node and the target reference item node, a first connection probability between the adjacent item node and the target item node, and a second connection probability between the target reference item node and the adjacent item node;

[0093] Calculating a reference connection probability between the target item node and the target reference item node according to the first connection probability and the second connection probability;

[0094] The connection probability between the target item node and the target reference item node is determined according to the initial connection probability and the reference connection probability.

[0095] The initial connection probability between the target item node and the target reference item node specifically refers to the preset probability value between the two item nodes. If the two are directly connected, the initial connection probability between the two is 1. The first connection probability between the adjacent item node and the target item node specifically refers to the connection probability of the two item nodes that has been determined in the calculation of the aforementioned steps. The second connection probability between the target reference item node and the adjacent item node specifically refers to the connection probability between the two when the target reference item node is a neighbor node of the adjacent item node. In a specific embodiment provided in this specification, if the target reference item node is a neighbor node of the adjacent item node, the second connection probability between the two is 0.5. In actual applications, the second connection probability can be set according to actual conditions, and its value range is between 0-1.

[0096] After obtaining the first connection probability and the second connection probability, the reference connection probability between the target item node and the target reference item node can be calculated. Then, the connection probability between the target item node and the target reference item node is selected from the initial connection probability and the reference connection probability between the two.

[0097] In the method provided in this specification, the connection probability between the target item node and the target reference item node is calculated by the following formula 1.

[0098] wi.j =min(w i.j , 1-w i.j-1 w j-1.j ) Formula 1

[0099] Among them, w on the right side of the equation i.j is the connection probability between the target item node i and the target reference item node j (i.e., the initial connection probability), w i.j-1 represents the connection probability between the target item node i and the adjacent item node j-1 (i.e., the first connection probability), w j-1.j Represents the connection probability between the adjacent item node j-1 and the target reference item node j (i.e., the second connection probability). i.j-1 w j-1.j represents the reference connection probability between the target item node and the target reference item node. i.j is the connection probability between the target item node and the target reference item node j. It selects the minimum value between the initial connection probability and the reference connection probability.

[0100] See also Figure 3 As shown in Table 2 above, item node 2 is used as the target reference item node for explanation. In this embodiment, the second connection probability is 0.5 as an example. Item node 1 is an adjacent item node, and item node 2 is directly connected to item node 0. The initial connection probability between the two is 1, and the connection probability between item node 0 and item node 1 (i.e., the first connection probability) is 1. The reference connection probability between item node 0 and item node 2 can be calculated as 0.5 (calculated by 1-1*0.5) using Formula 1 above. Then, by selecting the smaller connection probability from the initial connection probability of 1 and the reference connection probability of 0.5, it can be determined that the connection probability between item node 0 and item node 2 is 0.5. Based on this, Table 2 above is updated to obtain Table 3 below.

[0101] Table 3

[0102] Item Node 1 Item Node 2 Item Node 3 Item Node 4 Item Node 0 1 0.5

[0103] The following takes item node 3 as an example for further explanation.

[0104] Item node 3 is determined as the target reference item node. Item node 0 has adjacent item nodes 1 and 2. Item node 1 is first used as the adjacent item node for calculation. Item node 3 is not a neighbor of item node 1. Item node 1 is then used as the adjacent item node for calculation, and the connection probability between item node 3 and item node 0 is 1.

[0105] Item node 2 is considered an adjacent item node, and item node 3 is a neighbor of item node 2. The connection probability between item node 0 and item node 2 (i.e., the first connection probability) is 0.5, and the connection probability between item node 3 and item node 2 (i.e., the second connection probability) is 0.5. Using formula 1 above, we can calculate the reference connection probability between item node 0 and item node 3 to be 0.75 (calculated by 1-0.5*0.5).

[0106] The connection probability calculated with item node 1 as the adjacent item node is 1, and the connection probability calculated with item node 2 as the adjacent item node is 0.75. The smallest connection probability, 0.75, is selected as the connection probability between item node 0 and item node 3, thereby updating Table 3 above. Similarly, the connection probability between item node 4 and item node 0 is updated to obtain the following Table 4.

[0107] Table 4

[0108] Item Node 1 Item Node 2 Item Node 3 Item Node 4 Item Node 0 1 0.5 0.75 0.75

[0109] Based on the same calculation method, the connection probability between each item node in the reference index graph is calculated to obtain the final reference index graph.

[0110] Step 106: Update the reference index graph according to the connection probability between each item node, and obtain a target index graph corresponding to the reference index graph.

[0111] In practical applications, the connection probabilities between item nodes calculated using the above calculation method may result in different connection probabilities due to differences in the order in which the item nodes are determined. To mitigate the imbalance in connection probability calculation caused by the order in which the target item nodes are determined, the method provided in this specification updates the reference index map multiple times based on the connection probabilities between the item nodes to obtain the final target index map.

[0112] Furthermore, in another specific embodiment provided in this specification, the method further includes:

[0113] Determine the preset update rounds;

[0114] The reference index map is updated according to the preset update round to obtain a target index map.

[0115] In practical applications, when updating the reference index graph, a preset update round is required. Specifically, each update round is composed of updating the connection probability of all item nodes in the reference index graph as target item nodes. After the preset update rounds, the target index graph is obtained.

[0116] Specifically, updating the reference index graph according to the connection probability between each item node to obtain the target index graph corresponding to the reference index graph includes:

[0117] Determining at least one connection probability to be deleted from each connection probability of the reference index graph;

[0118] Determine the connected item node group corresponding to each connection probability to be deleted;

[0119] Delete the connection relationship corresponding to each connected item node group to obtain the target index graph.

[0120] See also Figure 4 , Figure 4 A schematic diagram of a target index map provided by an embodiment of this specification is shown. The target index map is based on Figure 3 The connection relationship diagram between the item nodes shown is obtained after updating.

[0121] The method provided in this specification aims to reduce the out-degree of the index graph, that is, to reduce the number of choices for each item node. Therefore, it is preferred to delete connection relationships whose connection probability is less than a preset threshold.

[0122] After each update of the reference index map, the connection probabilities whose connection probabilities are less than a preset probability threshold are selected from the connection probabilities to be deleted. For example, if the preset probability threshold is 0.5, then after the reference index map is updated, the connection probabilities less than 0.5 are determined to be the connection probabilities to be deleted.

[0123] Then, the connected item node group corresponding to the connection probability to be deleted is determined. The connected item node group specifically refers to the two item nodes corresponding to the connection probability. Then, the connection relationship corresponding to the connected item node group is deleted, thereby reducing the number of connection relationships in the reference index graph. Under the premise of ensuring a large connectivity between the item nodes in the index graph, the number of connection relationships can be reduced, thereby reducing the index size (memory usage) to improve scalability.

[0124] Specifically, the process of obtaining the reference index map from the initial index map is the process of creating the reference index map. After the reference index map is generated, the connection relationship in the reference index map is not deleted. The reference index map must be updated first, and the connection relationship in the updated reference index map must be deleted.

[0125] Furthermore, in practical applications, due to the large number of item nodes, the connection probabilities on the item nodes can be sorted based on the item nodes, and the top-k connection relationships with high connection probabilities are saved, and the connection relationships outside the top-k connection relationships are deleted.

[0126] The graph-based index construction method provided in this specification extracts the connection relationships and connection probabilities between item nodes from a heterogeneous bipartite graph structure including user nodes and item nodes. By repeatedly updating the connection probabilities between item nodes and updating the reference index graph based on the connection probabilities between item nodes, the number of out-degrees of each item node in the index graph is reduced, thereby reducing the average out-degree of the graph as a whole. While ensuring high connectivity between item nodes in the index graph, the number of connection relationships can be reduced, thereby reducing the index size (memory usage) and improving scalability. A balance is achieved between the accuracy and efficiency of searching for item nodes.

[0127] See also Figure 5 , Figure 5 A flowchart of an item recommendation method provided in one embodiment of this specification is shown, specifically including:

[0128] Step 502: Obtain reference item information corresponding to the target user.

[0129] The target user specifically refers to the user for whom item recommendations are requested, and can be a natural person, institution, or organization. The reference item information corresponding to the target user specifically refers to item information related to the target user. In practice, this reference item information can include information about items purchased or viewed by the target user, or information about items purchased or viewed by other users related to the target user. The methods provided herein do not specify the specific content of the reference item information; it will be determined based on the actual application.

[0130] Step 504: Determine an initial item node and at least one to-be-recommended item node corresponding to the initial item node in a target index graph based on the reference item information, wherein the target index graph is constructed using the graph-based index construction method.

[0131] After determining the reference item information, an initial item node corresponding to the reference item information is determined in the target index graph, wherein the target index graph is specifically constructed by the graph-based index construction method in the above embodiment.

[0132] In the target index graph, an initial item node and the nodes to be recommended that can be connected to the initial item node are determined. In practical applications, a single initial item node may be connected to multiple nodes to be recommended. The connection probabilities between the initial item node and the nodes to be recommended vary.

[0133] Step 506: Determine at least one target recommended item in each of the item nodes to be recommended based on the connection probability between the initial item node and each of the item nodes to be recommended.

[0134] After determining the initial item node and the connection probability between the initial item node and each item node to be recommended, the item nodes to be recommended are sorted, and a preset number of recommended item nodes with high connection probability are selected to determine target recommended items for the target user.

[0135] In another specific embodiment provided in this specification, determining at least one target recommended item in each to-be-recommended item node based on the connection probability between the initial item node and each to-be-recommended item node includes:

[0136] sorting the item nodes to be recommended according to the connection probability between the initial item node and the item nodes to be recommended;

[0137] According to the sorting results, recommended items corresponding to a preset number of to-be-recommended item nodes with high connection probabilities are selected as target recommended items.

[0138] In practice, the recommended item nodes are sorted based on the probability of their connection to the initial item node, placing nodes with higher connection probabilities at the front and nodes with lower connection probabilities at the back. Based on the sorting results, items corresponding to a preset number of nodes with higher connection probabilities are selected as the target recommended items.

[0139] And recommend at least one determined target recommended item to the target user.

[0140] The item recommendation method provided by the embodiments of this specification includes obtaining reference item information corresponding to a target user; determining an initial item node and at least one to-be-recommended item node corresponding to the initial item node in a target index graph based on the reference item information, wherein the target index graph is constructed by the above-mentioned graph-based index construction method; and determining at least one target recommended item in each to-be-recommended item node based on the connection probability between the initial item node and each to-be-recommended item node.

[0141] This method can quickly identify recommended items within a target index graph with a reduced number of connections. Based on the target index graph, it ensures both the relevance of recommended items to the target user's reference item information and the efficiency of retrieval within the target index graph. This allows for rapid recommendations of relevant target items based on the user's reference item information.

[0142] Corresponding to the above-mentioned graph-based index construction method embodiment, this specification also provides a graph-based index construction device embodiment, Figure 6 A schematic structural diagram of a graph-based index construction device provided by an embodiment of this specification is shown.

[0143] like Figure 6 As shown, the device includes:

[0144] An acquisition module 602 is configured to acquire an initial index graph including user nodes and item nodes;

[0145] A generation module 604 is configured to generate a reference index graph including item nodes based on the connection relationship between each user node and each item node in the initial index graph, wherein the reference index graph includes connection probabilities between item nodes;

[0146] The updating module 606 is configured to update the reference index graph according to the connection probability between each item node, and obtain a target index graph corresponding to the reference index graph.

[0147] Optionally, the generating module 604 is further configured to:

[0148] Randomly selecting a target item node, a reference user node corresponding to the target item node, and a reference item node corresponding to the reference user node in the initial index graph;

[0149] Connecting the target item node and each reference item node, and calculating the connection probability between each item node based on the connection relationship between the target item node and each reference item node;

[0150] Each item node in the initial index graph is traversed until a connection relationship is established between the item nodes, thereby obtaining a reference index graph including the item nodes.

[0151] Optionally, the generating module 604 is further configured to:

[0152] Determine the target item node and the target reference item node;

[0153] Determine whether the target item node has adjacent item nodes;

[0154] If not, determining the probability between the target item node and the target reference item node is 1;

[0155] If so, the connection probability between the target item node and the target reference item node is determined according to the adjacent item nodes and the target reference item node.

[0156] Optionally, the generating module 604 is further configured to:

[0157] Determining whether the target reference item node is a neighbor node of the adjacent item node;

[0158] If not, determining the probability between the target item node and the target reference item node is 1;

[0159] If so, the connection probability between the target item node and the target reference item node is calculated based on the connection probability between the adjacent item node and the target item node.

[0160] Optionally, the generating module 604 is further configured to:

[0161] Obtaining an initial connection probability between the target item node and the target reference item node, a first connection probability between the adjacent item node and the target item node, and a second connection probability between the target reference item node and the adjacent item node;

[0162] Calculating a reference connection probability between the target item node and the target reference item node according to the first connection probability and the second connection probability;

[0163] The connection probability between the target item node and the target reference item node is determined according to the initial connection probability and the reference connection probability.

[0164] Optionally, the updating module 606 is further configured to:

[0165] Determining at least one connection probability to be deleted from each connection probability of the reference index graph;

[0166] Determine the connected item node group corresponding to each connection probability to be deleted;

[0167] Delete the connection relationship corresponding to each connected item node group to obtain the target index graph.

[0168] Optionally, the updating module 606 is further configured to:

[0169] Determine the preset update rounds;

[0170] The reference index map is updated according to the preset update round to obtain a target index map.

[0171] Optionally, the acquisition module 602 is further configured to:

[0172] Randomly select user nodes or item nodes and insert them into the graph;

[0173] If a user node is selected, a preset number of item nodes with the highest correlation with the user node are selected from the item node set and inserted into the graph;

[0174] If an item node is selected, a preset number of user nodes with the highest correlation with the item node are selected from the user node set and inserted into the graph;

[0175] Connecting the selected user node with some or all of the preset number of item nodes that have the highest correlation with the user node, and connecting the selected item node with some or all of the preset number of user nodes that have the highest correlation with the item node;

[0176] Until all user nodes and item nodes are inserted into the graph and links are established, an initial index graph including user nodes and item nodes is obtained.

[0177] Optionally, the device further includes a correlation score determination module configured to:

[0178] Determine the relevance score between users and items based on a pre-trained scoring model;

[0179] A preset number of item nodes with the highest correlation with the user node and a preset number of user nodes with the highest correlation with the item node are determined according to the correlation scores between the user and the item.

[0180] The graph-based index construction device provided in this specification extracts the connection relationships and connection probabilities between item nodes from a heterogeneous bipartite graph structure including user nodes and item nodes. By repeatedly updating the connection probabilities between item nodes and updating the reference index graph based on the connection probabilities between item nodes, the number of out-degrees of each item node in the index graph is reduced, thereby reducing the average out-degree of the graph as a whole. While ensuring high connectivity between item nodes in the index graph, the number of connection relationships can be reduced, thereby reducing the index size (memory usage) and improving scalability. A balance is achieved between the accuracy and efficiency of searching for item nodes.

[0181] The above is a schematic scheme of a graph-based index construction device of this embodiment. It should be noted that the technical scheme of the graph-based index construction device and the technical scheme of the graph-based index construction method described above are based on the same concept. For details not described in detail in the technical scheme of the graph-based index construction device, please refer to the description of the technical scheme of the graph-based index construction method described above.

[0182] Corresponding to the above-mentioned item recommendation method embodiment, this specification also provides an item recommendation device embodiment. Figure 7 FIG. 1 shows a schematic diagram of the structure of an item recommendation device provided by an embodiment of this specification. Figure 7 As shown, the device includes:

[0183] An acquisition module 702 is configured to acquire reference item information corresponding to a target user;

[0184] a node determination module 704 configured to determine an initial item node and at least one to-be-recommended item node corresponding to the initial item node in a target index graph based on the reference item information, wherein the target index graph is constructed using the graph-based index construction method described above;

[0185] The item determination module 706 is configured to determine at least one target recommended item in each to-be-recommended item node based on the connection probability between the initial item node and each to-be-recommended item node.

[0186] Optionally, the item determination module 706 is further configured to:

[0187] sorting the item nodes to be recommended according to the connection probability between the initial item node and the item nodes to be recommended;

[0188] According to the sorting results, recommended items corresponding to a preset number of to-be-recommended item nodes with high connection probabilities are selected as target recommended items.

[0189] Optionally, the device further includes a recommendation module configured to:

[0190] Recommending the at least one target recommended item to the target user.

[0191] This device can quickly identify recommended items within a target index graph with a reduced number of connections. Based on the target index graph, it ensures both the relevance of recommended items to the target user's reference item information and the efficiency of retrieval within the target index graph. It can quickly recommend relevant target items to the user based on their reference item information.

[0192] The above is a schematic diagram of an item recommendation device according to this embodiment. It should be noted that the technical solution of this item recommendation device and the technical solution of the aforementioned item recommendation method are based on the same concept. For details not described in detail in the technical solution of the item recommendation device, please refer to the description of the technical solution of the aforementioned item recommendation method.

[0193] See also Figure 8 , Figure 8 1 shows an architecture diagram of an item recommendation system provided by an embodiment of this specification. The item recommendation system may include a client 100 and a server 200;

[0194] The client 100 is used to send the reference item information corresponding to the target user to the server 200;

[0195] The server 200 is configured to determine, in a target index graph based on the reference item information, an initial item node and at least one to-be-recommended item node corresponding to the initial item node, wherein the target index graph is constructed using the graph-based index construction method described above, and to determine at least one target recommended item in each to-be-recommended item node based on a connection probability between the initial item node and each to-be-recommended item node; and to send the at least one target recommended item to the client 100.

[0196] The client 100 is further configured to receive at least one target recommended item sent by the server 200 .

[0197] The item recommendation system can include multiple clients 100 and a server 200. The clients 100 can be referred to as client-side devices, and the server 200 can be referred to as cloud-side devices. The multiple clients 100 can establish a communication connection through the server 200. In the item recommendation scenario, the server 200 is used to provide item recommendation services between the multiple clients 100. The multiple clients 100 can act as either senders or receivers, communicating through the server 200.

[0198] Users can interact with the server 200 through the client 100 to receive data sent by other clients 100, or send data to other clients 100. In the item recommendation scenario, the user can publish a data stream to the server 200 through the client 100. The server 200 generates at least one target recommended item based on the data stream and pushes the at least one target recommended item to other clients with which communication has been established.

[0199] The client 100 and the server 200 are connected via a network. The network provides a medium for the communication link between the client 100 and the server 200. The network can include various connection types, such as wired or wireless communication links or fiber optic cables. The data transmitted by the client 100 may need to be encoded, transcoded, compressed, or other processing before being released to the server 200.

[0200] The client 100 can be a browser, an APP (Application), or a web application such as an H5 (HyperTextMarkup Language 5, Hypertext Markup Language 5) application, or a light application (also known as a mini-program, a lightweight application) or a cloud application. The client 100 can be based on the software development kit (SDK) of the corresponding service provided by the server 200, such as developed based on the real-time communication (RTC) SDK. The client 100 can be deployed in an electronic device and needs to rely on the device to run or certain APPs in the device to run. For example, the electronic device can have a display screen and support information browsing, such as a personal mobile terminal such as a mobile phone, tablet computer, personal computer, etc. Various other types of applications can also be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0201] The server 200 may include servers that provide various services, such as servers that provide communication services to multiple clients, servers that support background training for models used on clients, and servers that process data sent by clients. It should be noted that the server 200 can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. The server can also be a server in a distributed system, or a server that integrates a blockchain. The server 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, content delivery networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0202] It is worth noting that the item recommendation methods provided in the embodiments of this specification are generally executed by the server. However, in other embodiments of this specification, the client may also have similar functions to the server and thus execute the item recommendation methods provided in the embodiments of this specification. In other embodiments, the item recommendation methods provided in the embodiments of this specification may also be executed jointly by the client and the server.

[0203] Figure 9The block diagram of a computing device 900 according to one embodiment of the present disclosure is shown. Components of the computing device 900 include, but are not limited to, a memory 910 and a processor 920. The processor 920 is connected to the memory 910 via a bus 930, and a database 950 is used to store data.

[0204] The computing device 900 also includes an access device 940 that enables the computing device 900 to communicate via one or more networks 960. Examples of such networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 940 may include one or more of any type of network interface (e.g., a network interface card (NIC)) whether wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, or a near field communication (NFC) interface.

[0205] In one embodiment of the present specification, the above components of the computing device 900 and Figure 9 Other components not shown in the figure may also be connected to each other, for example, via a bus. Figure 9 The computing device structure block diagram shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art may add or replace other components as needed.

[0206] The computing device 900 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook computer, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smartwatch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 900 may also be a mobile or stationary server.

[0207] The processor 920 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-mentioned graph-based index construction method or item recommendation method.

[0208] The above is a schematic diagram of a computing device according to this embodiment. It should be noted that the technical solution of this computing device is based on the same concept as the technical solution of the graph-based index construction method or item recommendation method described above. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the graph-based index construction method or item recommendation method described above.

[0209] An embodiment of the present specification further provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the above-mentioned graph-based index construction method or item recommendation method.

[0210] The above is a schematic diagram of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium is based on the same concept as the technical solution of the graph-based index construction method or item recommendation method described above. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the graph-based index construction method or item recommendation method described above.

[0211] An embodiment of the present specification further provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned graph-based index construction method or item recommendation method.

[0212] The above is an illustrative embodiment of a computer program product. It should be noted that the technical solution of this computer program product shares the same concept as the technical solution of the graph-based index construction method or item recommendation method described above. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the graph-based index construction method or item recommendation method described above.

[0213] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0214] The computer instructions include computer program code, which may be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium may include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunications signal, and a software distribution medium.

[0215] It should be noted that for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0216] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0217] The preferred embodiments disclosed above are intended only to help illustrate this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made based on the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that those skilled in the art can better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A graph-based index construction method, comprising: Obtain an initial index graph including user nodes and item nodes; generating a reference index graph including item nodes based on the connection relationship between each user node and each item node in the initial index graph, wherein the reference index graph includes connection probabilities between item nodes; The reference index graph is updated according to the connection probability between each item node to obtain a target index graph corresponding to the reference index graph.

2. The method of claim 1, wherein generating a reference index graph including item nodes based on the connection relationship between each user node and each item node in the initial index graph comprises: Randomly selecting a target item node, a reference user node corresponding to the target item node, and a reference item node corresponding to the reference user node in the initial index graph; Connecting the target item node and each reference item node, and calculating the connection probability between each item node based on the connection relationship between the target item node and each reference item node; Each item node in the initial index graph is traversed until a connection relationship is established between the item nodes, thereby obtaining a reference index graph including the item nodes.

3. The method of claim 2, wherein calculating the connection probability between each item node based on the connection relationship between the target item node and each reference item node comprises: Determine the target item node and the target reference item node; Determine whether the target item node has adjacent item nodes; If not, determining the probability between the target item node and the target reference item node is 1; If so, the connection probability between the target item node and the target reference item node is determined according to the adjacent item nodes and the target reference item node.

4. The method of claim 3, wherein determining the connection probability between the target item node and the target reference item node based on the adjacent item nodes and the target reference item node comprises: Determining whether the target reference item node is a neighbor node of the adjacent item node; If not, determining the probability between the target item node and the target reference item node is 1; If so, the connection probability between the target item node and the target reference item node is calculated based on the connection probability between the adjacent item node and the target item node.

5. The method of claim 4, wherein the method further comprises calculating the connection probability between the target item node and the target reference item node based on the connection probability between the adjacent item nodes and the target item node, comprising: Obtaining an initial connection probability between the target item node and the target reference item node, a first connection probability between the adjacent item node and the target item node, and a second connection probability between the target reference item node and the adjacent item node; Calculating a reference connection probability between the target item node and the target reference item node according to the first connection probability and the second connection probability; The connection probability between the target item node and the target reference item node is determined according to the initial connection probability and the reference connection probability.

6. The method of claim 1, wherein updating the reference index graph based on the connection probability between each item node to obtain a target index graph corresponding to the reference index graph comprises: Determining at least one connection probability to be deleted from each connection probability of the reference index graph; Determine the connected item node group corresponding to each connection probability to be deleted; Delete the connection relationship corresponding to each connected item node group to obtain the target index graph.

7. The method of claim 6, further comprising: Determine the preset update rounds; The reference index map is updated according to the preset update round to obtain a target index map.

8. The method of claim 1, wherein obtaining an initial index graph including user nodes and item nodes comprises: Randomly select user nodes or item nodes and insert them into the graph; If a user node is selected, a preset number of item nodes with the highest correlation with the user node are selected from the item node set and inserted into the graph; If an item node is selected, a preset number of user nodes with the highest correlation with the item node are selected from the user node set and inserted into the graph; Connecting the selected user node with some or all of the preset number of item nodes that have the highest correlation with the user node, and connecting the selected item node with some or all of the preset number of user nodes that have the highest correlation with the item node; Until all user nodes and item nodes are inserted into the graph and links are established, an initial index graph including user nodes and item nodes is obtained.

9. The method of claim 8, further comprising: Determine the relevance score between users and items based on a pre-trained scoring model; A preset number of item nodes with the highest correlation with the user node and a preset number of user nodes with the highest correlation with the item node are determined according to the correlation scores between the user and the item.

10. An item recommendation method, comprising: Obtain reference item information corresponding to the target user; Determining an initial item node and at least one to-be-recommended item node corresponding to the initial item node in a target index graph according to the reference item information, wherein the target index graph is constructed by the method according to any one of claims 1 to 8; At least one target recommended item is determined in each of the item nodes to be recommended according to the connection probability between the initial item node and each of the item nodes to be recommended.

11. The method of claim 10, wherein determining at least one target recommended item in each to-be-recommended item node based on the connection probability between the initial item node and each to-be-recommended item node comprises: sorting the item nodes to be recommended according to the connection probability between the initial item node and the item nodes to be recommended; According to the sorting results, recommended items corresponding to a preset number of to-be-recommended item nodes with high connection probabilities are selected as target recommended items.

12. The method of claim 10, further comprising: Recommending the at least one target recommended item to the target user.

13. A computing device comprising: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 12 are implemented.

14. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 12.

15. A computer program product comprising a computer program / instruction, which implements the steps of the method according to any one of claims 1 to 12 when executed by a processor.

Images