Information processing method and device, electronic equipment and computer readable medium

Abstract

Embodiments of the present disclosure disclose information processing methods, devices, electronic equipment and computer readable media. A specific implementation of the method includes: performing subgraph extraction on a bipartite graph to generate a subgraph set; performing first encoding on a user information set input by each user node of the subgraph and an item information set input by each item node of the subgraph in the subgraph set respectively to generate a first user vector and a first item vector; generating an encoded subgraph; performing second encoding on the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph respectively to generate a second user vector and a second item vector; and replacing the vectors of the nodes in the encoded subgraph with the second user vector set and the second item vector set according to the corresponding positions to generate a replaced subgraph as a word embedded subgraph. The implementation can accurately and efficiently encode and process the user nodes and the item nodes in the bipartite graph.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the technical field of computer technology, and particularly to an information processing method and device, an electronic device and a computer readable medium. BACKGROUND

[0002] At present, various businesses often intuitively present various users and various items involved in the business in the form of a graph model. For the use of the graph model, the commonly used way is: first, the user information of each user and the item information of each item in the graph model are subjected to word embedding. Then, the word-embedded each user vector and each item vector are directly input into a pre-trained graph neural network related to the business, and the output result required by the business can be obtained. This way has the following problems:

[0003] The word-embedded each user vector and each item vector obtained by word embedding the user information of each user and the item information of each item in the graph model cannot better and deeply extract the feature information of each user and each item. In addition, the word-embedded each user vector and each item vector cannot better reflect the association information between the users and the items. SUMMARY

[0004] The summary part of the present disclosure is used to briefly introduce the concepts, which will be described in detail in the specific embodiments part. The summary part of the present disclosure is not intended to identify the key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] Some embodiments of the present disclosure propose an information processing method and device, an electronic device and a computer readable medium to solve the technical problems mentioned in the background part.

[0006] In a first aspect, some embodiments of the present disclosure provide an information processing method, which comprises: performing subgraph extraction on a bipartite graph to generate a set of subgraphs, wherein the bipartite graph represents an association relationship between a set of users and a set of items; performing first encoding on a set of user information input by each user node and a set of item information input by each item node in each subgraph in the set of subgraphs to generate a first user vector and a first item vector, to obtain a set of first user vectors and a set of first item vectors; generating an encoded subgraph according to the set of first user vectors and the set of first item vectors, to obtain a set of encoded subgraphs; performing second encoding on the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph according to the association relationship between the user nodes and the item nodes in the subgraph to generate a second user vector and a second item vector, to obtain a set of second user vectors and a set of second item vectors; and replacing the vectors of the nodes in the encoded subgraph with the set of second user vectors and the set of second item vectors in the corresponding positions to generate a replaced subgraph as a word-embedded subgraph, to obtain a set of word-embedded subgraphs.

[0007] Optionally, the method further comprises: according to the obtained word-embedded subgraphs, performing vector labeling on each node in the bipartite graph to obtain a vector-labeled bipartite graph; taking each item node in the vector-labeled bipartite graph as a leaf node to construct a tree model; and determining a set of items to be recalled in association with a target user node according to the tree model.

[0008] Optionally, the performing first encoding on a set of user information input by each user node and a set of item information input by each item node in each subgraph in the set of subgraphs to generate a first user vector and a first item vector comprises: performing word embedding processing on each user information in the set of user information to generate a first vector, to obtain a set of first vectors; splicing each first vector in the set of first vectors to obtain a spliced vector; performing batch normalization processing on the spliced vector to obtain a normalized vector; and inputting the normalized vector into a first multi-layer activation function layer to obtain the first user vector.

[0009] Optionally, the first user vector and the first item vector are respectively generated by performing first encoding on the set of user information input for each user node and the set of item information input for each item node in the subgraph set, including: performing word embedding processing on each item information in the set of item information to generate a second vector, to obtain a set of second vectors; splicing each second vector in the set of second vectors to obtain a spliced vector; performing batch normalization processing on the spliced vector to obtain a normalized vector; and inputting the normalized vector into a multi-layer second activation function layer to obtain the first item vector, wherein a vector dimension of the first user vector is greater than or equal to a vector dimension of the first item vector.

[0010] Optionally, the second user vector and the second item vector are respectively generated by performing second encoding on the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph according to the association relationship between the user nodes and the item nodes in the subgraph, including: determining a set of first item vectors adjacent to the first user vector according to the encoded subgraph; inputting each first item vector in the set of first item vectors into a third activation function layer to output a third vector, to obtain a set of third vectors; inputting each third vector in the set of third vectors into a first pooling layer to output a fourth vector, to obtain a set of fourth vectors; splicing each fourth vector in the set of fourth vectors and the first user vector to obtain a fifth vector; and inputting the fifth vector into a fourth activation function layer to obtain the second user vector.

[0011] Optionally, the second user vector and the second item vector are respectively generated by performing second encoding on the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph according to the association relationship between the user nodes and the item nodes in the subgraph, including: determining a set of first item vectors adjacent to the first user vector according to the encoded subgraph; inputting each first item vector in the set of first item vectors into a third activation function layer to output a third vector, to obtain a set of third vectors; inputting each third vector in the set of third vectors into a first pooling layer to output a fourth vector, to obtain a set of fourth vectors; splicing each fourth vector in the set of fourth vectors and the first user vector to obtain a fifth vector; and inputting the fifth vector into a fourth activation function layer to obtain the second user vector.

[0012] Optionally, the set of items to be recalled associated with the target user node is determined according to the tree model, including: determining an associated item node from the tree model by using a nearest neighbor search method according to the target user node; and determining the set of items to be recalled according to the item node.

[0013] In a second aspect, some embodiments of the present disclosure provide an information processing apparatus, the apparatus comprising: an extraction unit configured to perform subgraph extraction on a bipartite graph to generate a set of subgraphs, wherein the bipartite graph represents an association between a set of users and a set of items; a first encoding unit configured to perform first encoding on a set of user information input by each user node and a set of item information input by each item node in each subgraph in the set of subgraphs to generate a first user vector and a first item vector, to obtain a set of first user vectors and a set of first item vectors; a generation unit configured to generate an encoded subgraph according to the set of first user vectors and the set of first item vectors, to obtain a set of encoded subgraphs; a second encoding unit configured to perform second encoding on the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph according to the association between the user node and the item node in the subgraph to generate a second user vector and a second item vector, to obtain a set of second user vectors and a set of second item vectors; and a replacement unit configured to replace the vectors of the nodes in the encoded subgraph with the set of second user vectors and the set of second item vectors according to the corresponding positions to generate a replaced subgraph as a word-embedded subgraph, to obtain a set of word-embedded subgraphs.

[0014] Optionally, the apparatus further comprises: vector labeling each node in the bipartite graph according to the obtained each word-embedded subgraph, to obtain a vector-labeled bipartite graph; constructing a tree model with each item node in the vector-labeled bipartite graph as a leaf node; and determining a set of items to be recalled associated with a target user node according to the tree model.

[0015] Optionally, the first encoding unit is further configured to: perform word embedding processing on each user information in the set of user information to generate a first vector, to obtain a set of first vectors; splice each first vector in the set of first vectors, to obtain a spliced vector; perform batch normalization processing on the spliced vector, to obtain a normalized vector; and input the normalized vector into a first multi-layer activation function layer, to obtain the first user vector.

[0016] Optionally, the first encoding unit is further configured to: perform word embedding processing on each of the item information in the set of item information to generate a second vector, to obtain a set of second vectors; splice each of the second vectors in the set of second vectors to obtain a spliced vector; perform batch normalization processing on the spliced vector to obtain a normalized vector; and input the normalized vector into a second multi-layer activation function layer to obtain the first item vector, wherein a vector dimension of the first user vector is greater than or equal to a vector dimension of the first item vector.

[0017] Optionally, the second encoding unit is further configured to: determine, according to the encoded subgraph, a set of first item vectors adjacent to the first user vector; input each of the first item vectors in the set of first item vectors into a third activation function layer to output a third vector, to obtain a set of third vectors; input each of the third vectors in the set of third vectors into a first pooling layer to output a fourth vector, to obtain a set of fourth vectors; splice each of the fourth vectors in the set of fourth vectors with the first user vector to obtain a fifth vector; and input the fifth vector into a fourth activation function layer to obtain the second user vector.

[0018] Optionally, the second encoding unit is further configured to: determine, according to the encoded subgraph, a set of first user vectors adjacent to the first item vector; input each of the first user vectors in the set of adjacent first user vectors into a fifth activation function layer to output a sixth vector, to obtain a set of sixth vectors; input each of the sixth vectors in the set of sixth vectors into a second pooling layer to output a seventh vector, to obtain a set of seventh vectors; splice each of the seventh vectors in the set of seventh vectors with the first item vector to obtain an eighth vector; and input the eighth vector into a sixth activation function layer to obtain the second item vector.

[0019] Optionally, the apparatus further includes: determining, according to a target user node, an associated item node from the tree model using a nearest neighbor search method; and determining, according to the item node, a set of items to be recalled.

[0020] In a third aspect, some embodiments of the present disclosure provide an electronic device, including: one or more processors; a storage device having one or more programs stored thereon, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any of the first aspect.

[0021] In a fourth aspect, some embodiments of the present disclosure provide a computer readable medium having a computer program stored thereon, wherein the program is executed by a processor to implement the method of any of the first aspect.

[0022] The above various embodiments of the present disclosure have the following beneficial effects: the information processing method of some embodiments of the present disclosure can accurately and efficiently encode the user nodes and the item nodes in the bipartite graph. Specifically, the user information of each user and the item information of each item in the graph model are subjected to word embedding, and the obtained word-embedded user vectors and item vectors cannot better and deeply extract the feature information of each user and each item. In addition, the word-embedded user vectors and item vectors cannot better reflect the association information between users and items. Based on this, the information processing method of some embodiments of the present disclosure will first perform subgraph extraction on the bipartite graph to generate a subgraph set. The bipartite graph represents the association relationship between the user set and the item set. Here, the number of nodes in the bipartite graph is generally much larger than the number of nodes in each subgraph. Therefore, in subsequent training, the node size during subsequent model aggregation can be greatly reduced, and the complexity of model training is reduced. Then, the user information set input by each user node of the subgraph in the subgraph set and the item information set input by each item node are respectively subjected to first encoding to extract user feature information and item feature information to generate first user vectors and first item vectors, and a first user vector set and a first item vector set are obtained. Then, according to the first user vector set and the first item vector set, an encoded subgraph is generated, and an encoded subgraph set is obtained. Further, according to the association relationship between the user nodes and the item nodes in the subgraph, the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph are respectively subjected to second encoding to generate second user vectors and second item vectors, and a second user vector set and a second item vector set are obtained. It should be noted that the second encoding can further extract the association relationship between the user nodes and the item nodes in the subgraph. So that each node in the bipartite graph can learn the graph topology structure information at the same time. Finally, the vectors of the nodes in the encoded subgraph are replaced by the second user vector set and the second item vector set according to the corresponding positions to generate a replaced subgraph as a word-embedded subgraph, and a word-embedded subgraph set is obtained. Therefore, the information processing method can accurately and efficiently encode the user nodes and the item nodes in the bipartite graph. BRIEF DESCRIPTION OF DRAWINGS

[0023] The above and other features, advantages, and aspects of embodiments of the present disclosure will become more apparent upon consideration of the following detailed description, taken in conjunction with the accompanying drawings. In the drawings like reference numerals designate like elements throughout the various figures. It should be understood that the drawings are schematic and elements in the drawings are not necessarily to scale.

[0024] Figure 1is a first schematic diagram of one application scenario of the information processing method of some embodiments of the present disclosure;

[0025] Figure 2 is a second schematic diagram of one application scenario of the information processing method of some embodiments of the present disclosure;

[0026] Figure 3 is a third schematic diagram of one application scenario of the information processing method of some embodiments of the present disclosure;

[0027] Figure 4 is a flowchart of some embodiments of the information processing method according to the present disclosure;

[0028] Figure 5 is a flowchart of some other embodiments of the information processing method according to the present disclosure;

[0029] Figure 6 is a structural schematic diagram of some embodiments of the information processing apparatus according to the present disclosure;

[0030] Figure 7 is a structural schematic diagram of an electronic device suitable for being used to implement some embodiments of the present disclosure. DETAILED DESCRIPTION

[0031] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be interpreted as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly and completely understood. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0032] In addition, it should be further noted that only parts related to the present application are shown in the drawings for ease of description. The embodiments in the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0033] It should be noted that the terms "first", "second", etc. mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0034] It should be noted that the adjectives "one", "multiple" mentioned in the present disclosure are illustrative and not limiting, and those skilled in the art should understand that unless otherwise explicitly stated in the context, it should be understood as "one or more".

[0035] The names of messages or information exchanged between the plurality of devices in the embodiments of the present disclosure are only for illustrative purposes, and are not intended to limit the scope of the messages or information.

[0036] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0037] Figures 1-3 is a schematic diagram of one application scenario of an information processing method of some embodiments of the present disclosure.

[0038] As Figures 1-3As shown, the electronic device 101 can first perform subgraph extraction on the bipartite graph 102 to generate a subgraph set 105. Among them, the bipartite graph 102 represents the association between the user set 103 and the item set 104. In this application scenario, the user set 103 can include a first user 1031, a second user 1032, and a third user 1033. The item set 104 can include a first item 1041, a second item 1042, and a third item 1043. The subgraph set 105 includes a subgraph 1051, a subgraph 1052, and a subgraph 1053. The first layer node of the subgraph 1051 is the first user 1031. The second layer node of the subgraph 1051 is the first item 1041 and the second item 1042 associated with the first user 1031 in the first layer node. The third layer node of the subgraph 1051 includes the first user 1031 and the third user 1033 associated with the first item 1041 in the second layer node, and the first user 1031 and the third user 1033 associated with the second item 1042 in the second layer node. The first layer node of the subgraph 1052 is the second user 1032. The second layer node of the subgraph 1052 is the first item 1041 and the third item 1043 associated with the second user 1032 in the first layer node. The third layer node of the subgraph 1052 includes the first user 1031 and the third user 1033 associated with the first item 1041 in the second layer node, and the second user 1032 and the third user 1033 associated with the third item 1043 in the second layer node. The first layer node of the subgraph 1053 is the third user 1033. The second layer node of the subgraph 1053 is the first item 1041 and the second item 1042 associated with the third user 1033 in the first layer node. The third layer node of the subgraph 1053 includes the first user 1031 and the third user 1033 associated with the first item 1041 in the second layer node, and the first user 1031 and the third user 1033 associated with the second item 1042 in the second layer node. Then, the user information set input by each user node of the subgraph in the subgraph set 105 and the item information set input by each item node are respectively encoded for the first time to generate the first user vector and the first item vector, obtaining the first user vector set and the first item vector set. Then, according to the first user vector set and the first item vector set, an encoded subgraph is generated, obtaining an encoded subgraph set. In this application scenario, the encoded subgraph 106 is obtained according to the subgraph 1052. The first user vector corresponding to the second user in the first layer of the encoded subgraph 106 can be (1, 2, 4, 5, 2, 43, 2, 12, 21, 33). The first item vector corresponding to the first item in the second layer of the encoded subgraph 106 can be (6, 2, 5, 2, 9, 4, 1, 3).The first item vector corresponding to the third item in the second layer of the encoded subgraph 106 can be (12, 4, 9, 3, 7, 4, 0, 3). The first user vector corresponding to the first user in the third layer of the encoded subgraph 106 can be (4, 3, 4, 8, 4, 8, 21, 12, 23, 21). The first user vector corresponding to the third user in the third layer of the encoded subgraph 106 can be (12, 2, 12, 5, 2, 4, 2, 2, 0, 3). The first user vector corresponding to the second user in the third layer of the encoded subgraph 106 can be (1, 2, 4, 5, 2, 43, 2, 12, 21, 33). However, according to the association between the user nodes and the item nodes in the subgraph, the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph are encoded for the second time respectively to generate a second user vector and a second item vector, and a second user vector set and a second item vector set are obtained. Finally, the vectors of each node in the encoded subgraph are replaced by the second user vector set and the second item vector set in the corresponding positions to generate a replaced subgraph as a word embedding subgraph, and a word embedding subgraph set is obtained. In this application scenario, the word embedding subgraph 107 is generated according to the encoded subgraph 106. The second user vector corresponding to the second user in the first layer of the word embedding subgraph 107 can be (8, 8, 7, 5, 2, 3, 5, 12, 5, 4). The second item vector corresponding to the first item in the second layer of the word embedding subgraph 107 can be (6, 6, 8, 5, 2, 3, 5, 5, 5, 3). The second item vector corresponding to the third item in the second layer of the word embedding subgraph 107 can be (5, 5, 7, 5, 2, 3, 3, 12, 0, 2). The second user vector corresponding to the first user in the third layer of the word embedding subgraph 107 can be (1, 2, 3, 5, 2, 3, 5, 9, 5, 4). The second user vector corresponding to the third user in the third layer of the word embedding subgraph 107 can be (2, 2, 7, 5, 2, 3, 1, 1, 5, 9). The second user vector corresponding to the second user in the third layer of the word embedding subgraph 107 can be (8, 8, 7, 5, 2, 3, 5, 12, 5, 4).

[0039] It should be noted that the electronic device 101 can be hardware or software. When the electronic device is hardware, it can be implemented as a distributed cluster composed of multiple servers or terminal devices, or as a single server or a single terminal device. When the electronic device is software, it can be installed in the hardware devices listed above. It can be implemented as multiple software or software modules for providing distributed services, or as a single software or software module. No specific limitation is made here.

[0040] It should be understood that Figure 1 , Figure 2 or Figure 3 The number of electronic devices in

[0041] With reference to Figure 4 , a flow 400 of some embodiments of the information processing method according to the present disclosure is shown. The information processing method comprises the following steps:

[0042] Step 401, subgraph extraction is performed on the bipartite graph to generate a subgraph set.

[0043] In some embodiments, the execution subject of the information processing method (for example Figure 1 The electronic device 101 shown) can perform subgraph extraction on the bipartite graph to generate a subgraph set. Wherein, the bipartite graph described above characterizes the association relationship between the user set and the item set. For example, the association relationship between users and items in the bipartite graph described above can be a click behavior relationship. In addition, the connection edge in the bipartite graph described above can be directed from the user to the item. No connection edge will be established between users and items. The connection edge described above can characterize the association relationship between users and items.

[0044] As an example, a two-stage graph sampling strategy can be used to perform subgraph extraction to generate a subgraph set.

[0045] Optionally, the two-stage graph sampling strategy described above can include:

[0046] First, determine a target user node from the user node set in the bipartite graph described above.

[0047] Second, determine the number of nodes in the set of item nodes connected to the target user node in the bipartite graph described above.

[0048] Third, randomly select a first number of item nodes from the set of item nodes connected to the target user node. Wherein, the first number is less than or equal to the number of nodes in the set of connected item nodes.

[0049] Fourth, determine the number of nodes in the set of user nodes connected to each selected item node.

[0050] Fifth, select a second number of user nodes from the set of user nodes connected to each selected item node. Wherein, the second number is less than or equal to the number of nodes in the set of connected user nodes.

[0051] At step 402, the execution subject respectively encodes the user information set input by each user node of the subgraph in the subgraph set and the item information set input by each item node to generate a first user vector and a first item vector, to obtain a first user vector set and a first item vector set.

[0052] In some embodiments, the execution subject respectively encodes the user information set input by each user node of the subgraph in the subgraph set and the item information set input by each item node to generate a first user vector and a first item vector, to obtain a first user vector set and a first item vector set. The user information set input by the user node can be user feature information related to the user, such as age information of the user, gender information of the user, and the like. The item information set input by the item node can be item feature information related to the item, such as category information of the item, brand information of the item, value information of the item, and the like.

[0053] As an example, the execution subject can respectively perform word embedding processing on the user information set input by each user node of the subgraph in the subgraph set and the item information set input by each item node to generate a first user vector and a first item vector, to obtain a first user vector set and a first item vector set. Word embedding can be a method of converting words in text into numerical vectors. The word embedding process can be embedding a high-dimensional space with a dimension of the number of all words into a continuous vector space with a lower dimension, and each word or phrase is mapped to a vector in the real number field. The result of word embedding generates a word vector.

[0054] In some optional implementations of some embodiments, the encoding of the user information set input by each user node of the subgraph in the subgraph set and the item information set input by each item node to generate a first user vector and a first item vector can include the following steps:

[0055] First, word embedding processing is performed on each user information in the user information set to generate a first vector, to obtain a first vector set.

[0056] Second, each first vector in the first vector set is spliced to obtain a spliced vector.

[0057] Third, batch normalization (BN) is performed on the spliced vector to obtain a normalized vector. Here, the batch normalization of the spliced vector can accelerate the training speed in model training and save the regularization process. Further, the training accuracy of the model is improved.

[0058] In the fourth step, the normalized vector is input into a multi-layer first activation function layer to obtain the first user vector. The dimensions of the output vectors between the first activation function layers can be inconsistent. The activation function can include, but is not limited to, at least one of a normalized exponential function (Softmax function) and a rectified linear unit (ReLU). As an example, the multi-layer first activation function layer can be a three-layer first activation function. The output vector dimension of the first layer of the first activation function can be 512x1. The output vector dimension of the second layer of the first activation function can be 128x1. The output vector dimension of the third layer of the first activation function can be 64x1.

[0059] Optionally, the first encoding of the set of user information input to each user node of the subgraph and the set of item information input to each item node of the subgraph to generate the first user vector and the first item vector can include the following steps:

[0060] In the first step, word embedding processing is performed on each item information in the set of item information to generate a second vector, thereby obtaining a set of second vectors.

[0061] In the second step, each second vector in the set of second vectors is spliced to obtain a spliced vector.

[0062] In the third step, batch normalization processing is performed on the spliced vector to obtain a normalized vector.

[0063] In the fourth step, the normalized vector is input into a multi-layer second activation function layer to obtain the first item vector. The vector dimension of the first user vector is greater than or equal to the vector dimension of the first item vector. As an example, the multi-layer second activation function layer can be a three-layer second activation function. The output vector dimension of the first layer of the second activation function can be 128x1. The output vector dimension of the second layer of the second activation function can be 64x1. The output vector dimension of the third layer of the second activation function can be 32x1.

[0064] In step 403, an encoded subgraph is generated based on the set of first user vectors and the set of first item vectors, thereby obtaining a set of encoded subgraphs.

[0065] In some embodiments, the execution subject can generate an encoded subgraph according to the first user vector set and the first item vector set, to obtain an encoded subgraph set. Each node in the encoded subgraph has a corresponding encoded vector. As an example, the execution subject can label the first user vector set and the first item vector set in the corresponding positions in the subgraph to obtain the encoded subgraph.

[0066] At step 404, the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph are respectively encoded for the second time according to the association relationship between the user nodes and the item nodes in the subgraph to generate the second user vector and the second item vector, to obtain a second user vector set and a second item vector set.

[0067] In some embodiments, the execution subject can encode the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph for the second time according to the association relationship between the user nodes and the item nodes in the subgraph to generate the second user vector and the second item vector, to obtain a second user vector set and a second item vector set. As an example, the execution subject can input the encoded subgraph into a pre-trained graph convolutional neural network (GCN) according to the association relationship between the user nodes and the item nodes in the subgraph, to obtain the second user vector set and the second item vector set.

[0068] In some optional implementations of some embodiments, the encoding of the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph for the second time according to the association relationship between the user nodes and the item nodes in the subgraph to generate the second user vector and the second item vector can include the following steps:

[0069] First, determine the first item vector set adjacent to the first user vector according to the encoded subgraph.

[0070] Second, input each first item vector in the first item vector set into a third activation function layer to output a third vector, to obtain a third vector set. As an example, the output vector dimension of the third activation function layer can be 64x1.

[0071] Thirdly, input each third vector in the third vector set into a first pooling layer to output a fourth vector, and obtain a fourth vector set. The first pooling layer can be one of Max Pooling and Average Pooling. The pooling layer can further process the feature mapping result. The pooling layer can statistically aggregate the feature values of a position and its adjacent positions in a plane, and take the aggregated result as the value of the position in the plane. The Max Pooling layer calculates the maximum value in the position and its adjacent matrix region, and takes the maximum value as the value of the position. The Average Pooling layer calculates the average value in the position and its adjacent matrix region, and takes the average value as the value of the position. The use of the pooling layer does not change the depth of the data matrix, but only reduces the height and width to achieve the purpose of dimension reduction. As an example, the dimension of the output vector of the first pooling layer can be 64x1.

[0072] Fourthly, splice each fourth vector in the fourth vector set and the first user vector to obtain a fifth vector.

[0073] Fifthly, input the fifth vector into a fourth activation function layer to obtain the second user vector. As an example, the dimension of the output vector of the fourth activation function layer can be 64x1.

[0074] In some optional implementations of some embodiments, the second encoding of each first user vector corresponding to each user node and each first item vector corresponding to each item node in the encoded subgraph according to the association between the user nodes and the item nodes in the subgraph to generate a second user vector and a second item vector can include the following steps:

[0075] Firstly, determine a first user vector set adjacent to the first item vector according to the encoded subgraph.

[0076] Secondly, input each first user vector in the adjacent first user vector set into a fifth activation function layer to output a sixth vector, and obtain a sixth vector set. As an example, the dimension of the output vector of the fifth activation function layer can be 64x1.

[0077] Thirdly, input each sixth vector in the sixth vector set into a second pooling layer to output a seventh vector, and obtain a seventh vector set. As an example, the dimension of the output vector of the second pooling layer can be 64x1.

[0078] Fourthly, splice each seventh vector in the seventh vector set and the first item vector to obtain an eighth vector.

[0079] In the fifth step, the eighth vector is input into a sixth activation function layer to obtain the second item vector. As an example, the output vector dimension of the sixth activation function layer can be 64x1.

[0080] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: the information processing method of some embodiments of the present disclosure can accurately and efficiently encode the user nodes and item nodes in the bipartite graph. Specifically, the user information of each user and the item information of each item in the graph model are subjected to word embedding, and the obtained word-embedded user vectors and item vectors cannot better and deeply extract the feature information of each user and each item. In addition, the word-embedded user vectors and item vectors cannot better reflect the association information between users and items. Based on this, the information processing method of some embodiments of the present disclosure will first extract subgraphs from the bipartite graph to generate a set of subgraphs. Here, the bipartite graph represents the association relationship between the user set and the item set. At this point, the number of nodes in the bipartite graph is generally much larger than the number of nodes in each subgraph. Therefore, in subsequent training, the node size when aggregating the subsequent model can be greatly reduced, and the complexity of model training can be reduced. Then, the user information set input by each user node and the item information set input by each item node in the subgraph in the set of subgraphs are respectively subjected to first encoding to extract user feature information and item feature information to generate first user vectors and first item vectors, to obtain a set of first user vectors and a set of first item vectors. Then, according to the set of first user vectors and the set of first item vectors, an encoded subgraph is generated to obtain a set of encoded subgraphs. Further, according to the association relationship between the user nodes and the item nodes in the subgraph, the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph are respectively subjected to second encoding to generate second user vectors and second item vectors, to obtain a set of second user vectors and a set of second item vectors. It should be noted that the second encoding can further extract the association relationship between the user nodes and the item nodes in the subgraph. So that each node in the bipartite graph can learn the graph topology structure information at the same time. Finally, the vectors of the nodes in the encoded subgraph are replaced by the set of second user vectors and the set of second item vectors according to the corresponding positions to generate a replaced subgraph as a word-embedded subgraph, to obtain a set of word-embedded subgraphs. Therefore, the information processing method can accurately and efficiently encode the user nodes and item nodes in the bipartite graph.

[0081] With reference to Figure 5 , flow 500 illustrating another embodiment of the information processing method according to the present disclosure is shown. The information processing method comprises the following steps:

[0082] Step 501, subgraph extraction is performed on the bipartite graph to generate a set of subgraphs.

[0083] Step 502, the user information set input by each user node of the subgraph in the set of subgraphs and the item information set input by each item node are respectively encoded for the first time to generate a first user vector and a first item vector, obtaining a first user vector set and a first item vector set.

[0084] Step 503, according to the first user vector set and the first item vector set, an encoded subgraph is generated, obtaining a set of encoded subgraphs.

[0085] Step 504, according to the association relationship between the user nodes and the item nodes in the subgraph, the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph are respectively encoded for the second time to generate a second user vector and a second item vector, obtaining a second user vector set and a second item vector set.

[0086] Step 505, the vectors of the nodes in the encoded subgraph are replaced by the second user vector set and the second item vector set according to the corresponding positions to generate a replaced subgraph as a word-embedded subgraph, obtaining a set of word-embedded subgraphs.

[0087] In some embodiments, the specific implementation of steps 501-505 and the resulting technical effects can refer to steps 401-405 in the corresponding embodiments, which will not be described here. Figure 4 The corresponding embodiments of steps 401-405, which will not be described here.

[0088] Step 506, according to the obtained each word-embedded subgraph, vector labeling is performed on each node on the bipartite graph to obtain a vector-labeled bipartite graph.

[0089] In some embodiments, in response to the absence, the execution subject (for example Figure 1 The electronic device 101 shown in the figure) can perform vector labeling on each node on the bipartite graph according to the obtained each word-embedded subgraph, to obtain a vector-labeled bipartite graph.

[0090] Step 507, each item node in the vector-labeled bipartite graph is taken as a leaf node to construct a tree model.

[0091] In some embodiments, the execution subject takes each item node in the vector-labeled bipartite graph as a leaf node to construct a tree model.

[0092] As an example, the execution subject can first take each item node in the vector-labeled bipartite graph as a leaf node. Then, using a clustering algorithm, the vectors corresponding to each leaf node are clustered to obtain a predetermined number of internal nodes at the next level. Each internal node can be at least one leaf node that is classified into a category. Further, take the vector set at the cluster center in each internal node. Finally, by multiple clustering to determine the internal nodes at the next level and the vector set at the cluster center, the root node of the tree model can be finally obtained.

[0093] At step 508, according to the above tree model, the set of items to be recalled associated with the target user node is determined.

[0094] In some embodiments, the execution subject can determine the set of items to be recalled associated with the target user node according to the above tree model in various ways.

[0095] In some optional implementations of some embodiments, the above determination of the set of items to be recalled associated with the target user node according to the above tree model can include the following steps:

[0096] First, according to the target user node, the method of neighbor search is used to determine the associated item nodes from the above tree model.

[0097] Second, according to the above item nodes, the set of items to be recalled is determined.

[0098] From Figure 5 It can be seen that, compared with Figure 4 the description of some embodiments corresponding to, Figure 5 the flow 500 of the information processing method in some embodiments corresponding to highlights the specific steps of implementing item recall related to the target user by obtaining the vector-embedded subgraph set. Thus, the scheme described in these embodiments is based on the obtained vector-embedded subgraph set to construct a tree model, and further accurately and quickly determines the set of items to be recalled associated with the target user node through the tree model.

[0099] Continuing to refer to Figure 6 , as an implementation of the above method in each figure, the present disclosure provides some embodiments of an information processing device, which device embodiments correspond to Figure 4 the above method embodiments, and the device can be specifically applied to various electronic devices.

[0100] As Figure 6As shown, the information processing apparatus 600 of some embodiments includes an extraction unit 601, a first encoding unit 602, a generation unit 603, a second encoding unit 604, and a replacement unit 605. The extraction unit 601 is configured to perform subgraph extraction on a bipartite graph to generate a set of subgraphs, where the bipartite graph represents the association between a set of users and a set of items. The first encoding unit 602 is configured to respectively perform first encoding on the set of user information input by each user node and the set of item information input by each item node in the set of subgraphs to generate a first user vector and a first item vector, thereby obtaining a set of first user vectors and a set of first item vectors. The generation unit 603 is configured to generate encoded subgraphs according to the set of first user vectors and the set of first item vectors, thereby obtaining a set of encoded subgraphs. The second encoding unit 604 is configured to respectively perform second encoding on the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph according to the association between the user nodes and the item nodes in the subgraph to generate a second user vector and a second item vector, thereby obtaining a set of second user vectors and a set of second item vectors. The replacement unit 605 is configured to replace the vectors of the nodes in the encoded subgraph with the set of second user vectors and the set of second item vectors according to the corresponding positions to generate a replaced subgraph as a word-embedded subgraph, thereby obtaining a set of word-embedded subgraphs.

[0101] In some optional implementations of some embodiments, the apparatus further includes a preprocessing unit and an input unit (not shown in the figure). The preprocessing unit can be further configured to preprocess the question asked by the target user to obtain a processed text in response to the absence. The input unit can be further configured to input the processed text into the pre-trained language generation model to obtain the reply information.

[0102] In some optional implementations of some embodiments, the apparatus further includes a vector labeling unit, a construction unit, and a determination unit (not shown in the figure). The vector labeling unit can be configured to vector label each node in the bipartite graph according to the obtained set of word-embedded subgraphs to obtain a vector-labeled bipartite graph. The construction unit can be configured to construct a tree model with each item node in the vector-labeled bipartite graph as a leaf node. The determination unit can be configured to determine a set of items to be recalled in association with the target user node according to the tree model.

[0103] In some optional implementations of some embodiments, the first encoding unit 602 of the information processing device 600 can be further configured to: perform word embedding processing on each of the user information in the user information set to generate a first vector, to obtain a first vector set; concatenate each first vector in the first vector set to obtain a concatenated vector; perform batch normalization processing on the concatenated vector to obtain a normalized vector; and input the normalized vector into a multi-layer first activation function layer to obtain the first user vector.

[0104] In some optional implementations of some embodiments, the first encoding unit 602 of the information processing device 600 can be further configured to: perform word embedding processing on each of the item information in the item information set to generate a second vector, to obtain a second vector set; concatenate each second vector in the second vector set to obtain a concatenated vector; perform batch normalization processing on the concatenated vector to obtain a normalized vector; and input the normalized vector into a multi-layer second activation function layer to obtain the first item vector, wherein the vector dimension of the first user vector is greater than or equal to the vector dimension of the first item vector.

[0105] In some optional implementations of some embodiments, the second encoding unit 604 of the information processing device 600 can be further configured to: determine, according to the encoded subgraph, a first item vector set adjacent to the first user vector; input each first item vector in the first item vector set into a third activation function layer to output a third vector, to obtain a third vector set; input each third vector in the third vector set into a first pooling layer to output a fourth vector, to obtain a fourth vector set; concatenate each fourth vector in the fourth vector set with the first user vector to obtain a fifth vector; and input the fifth vector into a fourth activation function layer to obtain the second user vector.

[0106] In some optional implementations of some embodiments, the second encoding unit 604 of the information processing device 600 can be further configured to: determine, according to the encoded subgraph, a first user vector set adjacent to the first item vector; input each first user vector in the adjacent first user vector set into a fifth activation function layer to output a sixth vector, to obtain a sixth vector set; input each sixth vector in the sixth vector set into a second pooling layer to output a seventh vector, to obtain a seventh vector set; concatenate each seventh vector in the seventh vector set with the first item vector to obtain an eighth vector; and input the eighth vector into a sixth activation function layer to obtain the second item vector.

[0107] In some optional implementation of some embodiments, the apparatus further comprises an item node determining unit and an item determining unit (not shown in the figures). The item node determining unit can be configured to determine the associated item node from the tree model according to the target user node by using a nearest neighbor search method. The item determining unit can be configured to determine the item set to be recalled according to the item node.

[0108] It can be understood that the units described in the apparatus 600 correspond to the respective steps in the method described above. Therefore, the operations, features and advantages described above for the method also apply to the apparatus 600 and the units included therein, and are not repeated here. Figure 4

[0109] The following refers to Figure 7 which shows a structural schematic diagram of an electronic device 700 suitable for implementing some embodiments of the present disclosure. Figure 7 The electronic device shown is merely an example and should not impose any limitation on the functions and use range of the embodiments of the present disclosure.

[0110] As shown in Figure 7 , the electronic device 700 can include a processing device (such as a central processor, a graphics processor, etc.) 701, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 702 or loaded from a storage device 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the electronic device 700 are also stored. The processing device 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0111] Generally, the following devices can be connected to the I / O interface 705: an input device 706 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 707 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 708 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 709. The communication device 709 can allow the electronic device 700 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 7 The electronic device 700 is shown with various devices, but it should be understood that all the shown devices are not required to be implemented or possessed. More or fewer devices can be alternatively implemented or possessed. Figure 7 Each block shown in the figure can represent a device or multiple devices as needed.

[0112] ​In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to some embodiments of the present disclosure. For example, some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for performing the methods illustrated by the flowcharts. In some such embodiments, the computer program can be downloaded and installed from a network via the communication device 709, or installed from the storage device 708, or installed from the ROM 702. When the computer program is executed by the processing device 701, the above-mentioned functions defined in the methods of some embodiments of the present disclosure are performed.

[0113] It should be noted that the computer readable medium described above in some embodiments of the present disclosure can be a computer readable signal medium or a computer readable storage medium or any combination thereof. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any suitable combination thereof. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In some embodiments of the present disclosure, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus or device. In some embodiments of the present disclosure, the computer readable signal medium can include a data signal carried in a baseband or as part of a carrier wave, in which the computer readable program code is carried. Such a propagated data signal can take any of a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport a program for use by or in connection with an instruction execution system, apparatus or device. The program code contained on the computer readable medium can be transmitted by any suitable medium, including but not limited to wire, cable, RF (radio frequency), etc., or any suitable combination thereof.

[0114] In some embodiments, the client, server, or both can communicate using any known or future developed network protocols such as HTTP (HyperText Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., communication networks). Examples of communication networks include local area networks ("LAN"), wide area networks ("WAN"), the Internet, and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any current or future developed networks.

[0115] The computer readable medium described above can be included in the apparatus described above; or can exist separately therefrom. The computer readable medium described above bears one or more programs, when the one or more programs are executed by the electronic device, cause the electronic device to: perform subgraph extraction on a bipartite graph to generate a subgraph set, wherein the bipartite graph represents an association relationship between a user set and an item set; perform first encoding on a user information set input by each user node and an item information set input by each item node of each subgraph in the subgraph set to generate a first user vector and a first item vector, to obtain a first user vector set and a first item vector set; generate an encoded subgraph according to the first user vector set and the first item vector set, to obtain an encoded subgraph set; perform second encoding on the first user vector corresponding to each user node and the first item vector corresponding to each item node in the encoded subgraph according to the association relationship between the user node and the item node in the subgraph, to generate a second user vector and a second item vector, to obtain a second user vector set and a second item vector set; and replace the vectors of each node in the encoded subgraph with the second user vector set and the second item vector set according to the corresponding positions to generate a replaced subgraph as a word embedded subgraph, to obtain a word embedded subgraph set.

[0116] Computer program code for carrying out operations of some embodiments of the present disclosure can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0117] The flow diagrams and the block diagrams in the drawings are illustrations of architectures, functionalities, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of present disclosure. In this regard, each block in the flow diagrams or block diagrams can represent a module, a procedure, or a part of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or in the reverse order, depending on the functionality involved. It is also noted that each block of the block diagrams and / or flow diagrams and combinations of blocks in the block diagrams and / or flow diagrams can be implemented by special purpose hardware-based systems which perform the specified functions or operations, or combinations of special purpose hardware and

[0118] The units described in some embodiments of the present disclosure can be implemented by software, or can be implemented by hardware. The described units can also be arranged in a processor, for example, a processor can be described as including an extraction unit, a first encoding unit, a generation unit, a second encoding unit, and a replacement unit. Among them, the names of these units do not constitute a limitation to the units themselves in some cases, for example, the first encoding unit can also be described as "a unit that respectively encodes the user information set input by each user node of the subgraph and the item information set input by each item node of the subgraph in the subgraph set to generate a first user vector and a first item vector, and obtains a first user vector set and a first item vector set".

[0119] The functionality described herein above can be performed, at least in part, by one or more hardware logic components. For example, and without limitation, illustrative types of hardware logic components that can be used include Field-programmable Gate Arrays (FPGAs), Program- specific Integrated Circuits (ASICs), Program- specific Standard Products (ASSPs), System-on-a-chip systems (SOCs), Complex Programmable Logic Devices (CPLDs), etc.

[0120] The above description is merely exemplary of the disclosure and the application made use of the principles of the technology. It is to be understood that the application scope of the embodiments of the disclosure is not limited to the specific combinations of technical features described above, and should also cover other technical solutions formed by any combination of the above technical features or equivalent features thereof without departing from the inventive concept. For example, the technical solutions formed by replacing the above features with technical features having similar functions disclosed in the embodiments of the disclosure (but not limited to) with each other.

Claims

1. An information processing method comprising: performing subgraph extraction on a bipartite graph to generate a set of subgraphs, wherein the bipartite graph represents an association relationship between a set of users and a set of items; performing first encoding on a set of user information input by each user node and a set of item information input by each item node of a subgraph in the set of subgraphs to generate a first user vector and a first item vector, respectively, to obtain a set of first user vectors and a set of first item vectors, wherein the first encoding comprises word embedding processing, concatenation, batch normalization processing, and input to a multi-layer activation function layer; generating an encoded subgraph according to the set of first user vectors and the set of first item vectors to obtain a set of encoded subgraphs; performing second encoding on a first user vector corresponding to each user node and a first item vector corresponding to each item node in the encoded subgraph according to an association relationship between the user nodes and the item nodes in the subgraph to generate a second user vector and a second item vector, respectively, to obtain a set of second user vectors and a set of second item vectors, wherein the second encoding comprises determining a set of vectors, input to an activation function layer, input to a pooling layer, concatenation, and input to an activation function layer; replacing vectors of each node in the encoded subgraph with the set of second user vectors and the set of second item vectors according to corresponding positions to generate a replaced subgraph as a word-embedded subgraph to obtain a set of word-embedded subgraphs.

2. The method of claim 1, wherein, The method further comprises: vector labeling each node on the bipartite graph according to each obtained word-embedded subgraph to obtain a vector-labeled bipartite graph; constructing a tree model with each item node in the vector-labeled bipartite graph as a leaf node; determining a set of items to be recalled associated with a target user node according to the tree model.

3. The method of claim 1, wherein, The first encoding on a set of user information input by each user node and a set of item information input by each item node of a subgraph in the set of subgraphs to generate a first user vector and a first item vector, respectively, comprises: performing word embedding processing on each user information in the set of user information to generate a first vector to obtain a set of first vectors; concatenating each first vector in the set of first vectors to obtain a concatenated vector; performing batch normalization processing on the concatenated vector to obtain a normalized vector; inputting the normalized vector to a multi-layer first activation function layer to obtain the first user vector.

4. The method of claim 3, wherein, The first encoding on a set of user information input by each user node and a set of item information input by each item node of a subgraph in the set of subgraphs to generate a first user vector and a first item vector, respectively, comprises: performing word embedding processing on each item information in the set of item information to generate a second vector to obtain a set of second vectors; concatenating each second vector in the set of second vectors to obtain a concatenated vector; performing batch normalization processing on the concatenated vector to obtain a normalized vector; Input the normalized vector into a multi-layer second activation function layer to obtain the first item vector, wherein the vector dimension of the first user vector is greater than or equal to the vector dimension of the first item vector.

5. The method of claim 1, wherein, The second encoding of each first user vector and each first item vector in the encoded subgraph according to the association relationship between the user nodes and the item nodes in the subgraph to generate a second user vector and a second item vector comprises: determining a first user vector set adjacent to the first item vector according to the encoded subgraph; inputting each first user vector in the adjacent first user vector set into a third activation function layer to output a third vector to obtain a third vector set; inputting each third vector in the third vector set into a first pooling layer to output a fourth vector to obtain a fourth vector set; splicing each fourth vector in the fourth vector set with the first user vector to obtain a fifth vector; inputting the fifth vector into a fourth activation function layer to obtain the second user vector.

6. The method of claim 1, wherein, The second encoding of each first user vector and each first item vector in the encoded subgraph according to the association relationship between the user nodes and the item nodes in the subgraph to generate a second user vector and a second item vector comprises: determining a first user vector set adjacent to the first item vector according to the encoded subgraph; inputting each first user vector in the adjacent first user vector set into a third activation function layer to output a third vector to obtain a third vector set; inputting each third vector in the third vector set into a first pooling layer to output a fourth vector to obtain a fourth vector set; splicing each fourth vector in the fourth vector set with the first user vector to obtain a fifth vector; inputting the fifth vector into a fourth activation function layer to obtain the second user vector.

7. The method of claim 2, wherein, The determining of the item set to be recalled associated with the target user node according to the tree model comprises: determining the associated item node from the tree model by using a nearest neighbor search method according to the target user node; determining the item set to be recalled according to the item node.

8. An information processing apparatus, comprising: an extraction unit configured to perform subgraph extraction on a bipartite graph to generate a subgraph set, wherein the bipartite graph represents an association relationship between a user set and an item set; a first encoding unit configured to perform first encoding on a user information set input by each user node and an item information set input by each item node of a subgraph in the subgraph set to generate a first user vector and a first item vector, to obtain a first user vector set and a first item vector set, wherein the first encoding comprises word embedding processing, splicing, batch normalization processing, and input into a multi-layer activation function layer; a generation unit configured to generate an encoded subgraph according to the first user vector set and the first item vector set to obtain an encoded subgraph set; a second encoding unit, configured to perform second encoding on each first user vector corresponding to each user node and each first item vector corresponding to each item node in the encoded subgraph according to the association between the user nodes and the item nodes in the subgraph to generate second user vectors and second item vectors, to obtain a second user vector set and a second item vector set, wherein the second encoding comprises determining a vector set, inputting the vector set into an activation function layer, inputting the vector set into a pooling layer, and inputting the vector set into an activation function layer after splicing; a replacing unit, configured to replace vectors of each node in the encoded subgraph with the second user vector set and the second item vector set according to corresponding positions to generate a replaced subgraph as a word embedding subgraph, to obtain a word embedding subgraph set.

9. An electronic device, comprising: one or more processors; storage configured to store one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the method of any of claims 1-7.

10. A computer readable medium having stored thereon a computer program, wherein, The program is executed by the processor to implement the method of any of claims 1-7.

Images