CN110781407A - User label generation method and device and computer readable storage medium - Google Patents
User label generation method and device and computer readable storage medium Download PDFInfo
- Publication number
- CN110781407A CN110781407A CN201911000656.2A CN201911000656A CN110781407A CN 110781407 A CN110781407 A CN 110781407A CN 201911000656 A CN201911000656 A CN 201911000656A CN 110781407 A CN110781407 A CN 110781407A
- Authority
- CN
- China
- Prior art keywords
- feature vector
- user
- text
- texts
- social
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000003860 storage Methods 0.000 title claims abstract description 13
- 239000013598 vector Substances 0.000 claims abstract description 237
- 230000006399 behavior Effects 0.000 claims abstract description 71
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 33
- 238000012545 processing Methods 0.000 claims abstract description 32
- 230000015654 memory Effects 0.000 claims abstract description 29
- 230000002457 bidirectional effect Effects 0.000 claims abstract description 20
- 230000001939 inductive effect Effects 0.000 claims abstract description 18
- 238000004590 computer program Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 7
- 230000006855 networking Effects 0.000 claims description 5
- 230000003542 behavioural effect Effects 0.000 claims 2
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 11
- 238000013528 artificial neural network Methods 0.000 description 9
- 230000004927 fusion Effects 0.000 description 9
- 230000008569 process Effects 0.000 description 9
- 238000012549 training Methods 0.000 description 8
- 230000004913 activation Effects 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 230000007246 mechanism Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 235000008694 Humulus lupulus Nutrition 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Abstract
The embodiment of the application discloses a user label generation method, a device and a computer readable storage medium, wherein the method comprises the following steps: acquiring a behavior text sequence of a user, and randomly initializing the behavior text sequence by using a bidirectional long-short term memory network (BilSTM) to obtain a first feature vector, wherein the behavior text sequence is a sequence formed by texts which are generated according to a time sequence and used for representing user behaviors; acquiring social network information of the user, and processing the social network information according to an inductive learning algorithm to obtain a second feature vector; obtaining social statistics information of the user, and obtaining a third feature vector according to the social statistics information; and transversely splicing the first feature vector, the second feature vector and the third feature vector to obtain the user label of the user. The embodiment of the application is beneficial to improving the precision of generating the user label.
Description
Technical Field
The application relates to the technical field of artificial intelligence, in particular to a user tag generation method and device and a computer readable storage medium.
Background
At present, a multi-modal information fusion technology is a hotspot of research in the field of artificial intelligence, for example, in the production process of a user tag, data of different modalities, such as images, texts, social networks, geographic positions and social statistics information, need to be organically fused together, so that user behavior characteristics are constructed in a multi-angle and all-around manner, an accurate user tag is given, an accurate predicted value is given, and product operation, personalized recommendation and accurate advertisement delivery are better served.
In addition, social network information plays a role in user tags, and is particularly important for users with sparse features, for example, if most friends of the user are in the building industry or the user joins multiple building industry discussion groups, even if other information of the user is sparse, the user is strongly believed to belong to the building industry.
At present, when the social network information is fused with other information in other modalities, a linear splicing method can be adopted, but for a super-large scale social network scene, the social relationship of each user is complicated, the social network vector of each user cannot be generated, and in addition, because different modality information has different contribution degrees to the user tags to which the users belong, only simple splicing is carried out, and the generated user tags are low in precision.
Disclosure of Invention
The embodiment of the application provides a user tag generation method, a user tag generation device and a computer readable storage medium, and the user tag generation method, the user tag generation device and the computer readable storage medium are used for generating an accurate user tag by fusing multi-modal data.
In a first aspect, an embodiment of the present application provides a user tag generation method, including:
acquiring a behavior text sequence of a user, and randomly initializing the behavior text sequence by using a bidirectional long-short term memory network (BilSTM) to obtain a first feature vector, wherein the behavior text sequence is a sequence formed by texts which are generated according to a time sequence and used for representing user behaviors;
acquiring social network information of the user, and processing the social network information according to an inductive learning algorithm to obtain a second feature vector;
obtaining social statistics information of the user, and obtaining a third feature vector according to the social statistics information;
and transversely splicing the first feature vector, the second feature vector and the third feature vector to obtain the user label of the user.
In a second aspect, an embodiment of the present application provides a user tag generation apparatus, where the user tag generation apparatus includes a processor, a transceiver, and at least one circuit, where the processor and the transceiver are connected through the at least one circuit;
the transceiver is used for acquiring a behavior text sequence, social network information and social statistics information of a user and sending the behavior text sequence, the social network information and the social statistics information to the processor, wherein the behavior text sequence is a sequence formed by texts which are generated according to a time sequence and used for representing user behaviors;
the processor is used for randomly initializing the behavior text sequence by using a bidirectional long-short term memory network (BilSTM) to obtain a first feature vector;
the processor is further used for processing the social network information according to an inductive learning algorithm to obtain a second feature vector;
the processor is further configured to obtain a third feature vector according to the social statistics information;
the processor is further configured to transversely splice the first feature vector, the second feature vector, and the third feature vector to obtain a user tag of the user.
In a third aspect, embodiments of the present application provide an electronic device, including a processor, a memory, a communication interface, and one or more programs, where the one or more programs are stored in the memory and configured to be executed by the processor, and the program includes instructions for performing the steps in the method according to the first aspect.
In a fourth aspect, embodiments of the present application provide a computer-readable storage medium, which stores a computer program, where the computer program makes a computer execute the method according to the first aspect.
In a fifth aspect, embodiments of the present application provide a computer program product comprising a non-transitory computer-readable storage medium storing a computer program, the computer being operable to cause a computer to perform the method according to the first aspect.
The embodiment of the application has the following beneficial effects:
it can be seen that, in the embodiment of the application, the social network information in the social modality is spliced with the feature vectors corresponding to the data information in other modalities to obtain the user splice, and the user label fuses the social network information in the social modality, and the social network information contains rich user features (such as industry information), so that the precision of the user label is improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a schematic diagram of a multi-modal data fusion provided by an embodiment of the present application;
FIG. 2 is a schematic diagram of another multimodal data fusion provided by an embodiment of the present application;
fig. 3A is a schematic view of a user tag generation scenario provided in an embodiment of the present application;
fig. 3B is a schematic flowchart of a user tag generation method according to an embodiment of the present application;
FIG. 3C is a schematic diagram of a sequence of processing actions based on an attention mechanism according to an embodiment of the present application;
FIG. 3D is a diagram illustrating a social graph network according to an embodiment of the present disclosure;
FIG. 3E is a diagram illustrating a social graph network according to an embodiment of the present disclosure;
fig. 4 is a schematic flowchart of another user tag generation method according to an embodiment of the present application;
FIG. 5 is a schematic diagram of a model for generating user tags according to an embodiment of the present application;
fig. 6 is a schematic structural diagram of a user tag generation apparatus according to an embodiment of the present application;
fig. 7 is a block diagram illustrating functional units of a user tag generation apparatus according to an embodiment of the present disclosure.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some, but not all, embodiments of the present application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The terms "first," "second," "third," and "fourth," etc. in the description and claims of this application and in the accompanying drawings are used for distinguishing between different objects and not for describing a particular order. Furthermore, the terms "include" and "have," as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those steps or elements listed, but may alternatively include other steps or elements not listed, or inherent to such process, method, article, or apparatus.
In the user tag generation process, data of different modalities (such as images, texts, social network information, geographic positions, social statistics information and the like) need to be fused, so that user behavior characteristics can be constructed in a multi-angle and all-around manner, and an accurate user tag is given.
Referring to fig. 1, fig. 1 provides a Deep Neural network model (DNN) for fusing a user browsing video sequence, a geographical location, and user social statistics information, and specifically, after a plurality of browsing vectors and a Search behavior sequence Vector of a user are respectively weighted, a video browsing Vector Watch Vector and a Search Vector are generated; and then splicing the Watch Vector and the Search Vector with social statistical information such as address location information, age, sex and the like to obtain a target characteristic Vector, feeding the target characteristic Vector into a forward neural network, and performing a series of subsequent processing (such as nonlinear activation of ReLU) to obtain the user label of the user.
Although the DNN network described above implements fusion of different modality data such as text, image, geographical location information, etc., a scheme for fusing social network information of a user is still lacking; in addition, social network information has a significant role in generating user tags, and is particularly important for users with sparse features, in order to fuse the social network information, a multi-modal fusion scheme based on a graph convolution neural network is provided, specifically referring to fig. 2, as shown in fig. 2, different nodes represent different users, the different users are connected in a side manner to represent the social relationship of each user, a social graph network is obtained, data of other modalities (such as a behavior text sequence, social statistics information and the like) are added to the social graph network as node features, then the graph network is used for training, and when the users include data of multiple modalities, simple linear splicing is performed on the multiple node features. When the social network graph is adopted for training, signals are transmitted among neighbor nodes of each node, so that the social network information of a user is added into the training process, but the node characteristic splicing method is simple, is not friendly to complex modal data (such as image characteristics), and reduces the expandability of multi-modal data fusion; moreover, when the node features are spliced, the contribution degree of each modal data to the user label is not considered, so that the precision of the spliced node features is low; in addition, when the social network of the user is complex, it is difficult to obtain signals of other nodes to the local node, and the feature vector corresponding to the local node cannot be obtained.
The scheme of the application is particularly provided for solving the problems in the multi-modal data fusion.
Referring to fig. 3A, fig. 3A is a schematic view of a user tag generation scenario provided in the embodiment of the present application, including a user terminal 110 and a tag generation apparatus 120;
wherein, the user performs social networking, online entertainment and the like through the user terminal 110, industry information (behavior text sequence) affiliated by the user, social networking information of the user, social statistical information of the user (such as age, gender, academic calendar and the like) can be collected from the user terminal 110, then, the collected behavior text sequence, social network information and social statistics information are sent to the tag generation device 120, the tag generation device 120 invokes a network model matched with each piece of information to process each piece of information, a first feature vector corresponding to the behavior text sequence is obtained, a second eigenvector corresponding to the social networking information, a third eigenvector corresponding to the social statistics information, and finally, and splicing the first feature vector, the second feature vector and the third feature vector to obtain a user label of the user, and carrying out targeted recommendation on the user based on the user label.
Referring to fig. 3B, fig. 3B is a user tag generation method according to an embodiment of the present application, where the method includes, but is not limited to, the following steps:
301: and acquiring a behavior text sequence of the user, and randomly initializing the behavior text sequence by using a bidirectional long-short term memory network to obtain a first feature vector.
The behavior text sequence is a sequence formed by texts which are generated in time sequence and used for representing user behaviors.
Specifically, the behavior text sequence includes N texts, where the N texts are used for representing behaviors of the user at N moments, where each moment corresponds to one text, that is, the texts used for the behaviors of the user at N different moments are combined according to a chronological order to obtain the behavior text sequence.
Then, a bidirectional Short-Term Memory (BilSTM) network is used for carrying out vector coding on each text to obtain a feature vector of each text, and weighting processing is carried out on the feature vector of each text to obtain the first feature vector.
302: and acquiring social network information of the user, and processing the social network information according to an inductive learning algorithm to obtain a second feature vector.
Wherein the social networking information comprises: industry information, social object information, and the like.
Wherein the inductive learning may be a GraphSAGE algorithm.
Optionally, a social graph network with the user is established according to the social network information, and the node where the user is located in the social graph network is processed according to the graphcage algorithm to obtain a second feature vector of the user.
303: and acquiring social statistics information of the user, and acquiring a third feature vector according to the social statistics information.
Optionally, the social statistics information includes P items of information with the user, where the P items of information include but are not limited to: age, gender, school calendar, home address, work place; and carrying out vector coding on each item of information to obtain a sub-vector corresponding to each item of information so as to obtain P sub-vectors, and then carrying out coding on the P sub-vectors so as to obtain a third feature vector.
304: and transversely splicing the first feature vector, the second feature vector and the third feature vector to obtain the user label of the user.
Performing dimensionality splicing, namely transverse splicing, on the first feature vector, the second feature vector and the third feature vector to obtain a target feature vector, performing nonlinear activation on the target feature vector, inputting the target feature vector subjected to nonlinear activation into a softmax classifier, and outputting a user tag of the user.
For example, the first eigenvector x ═ a, b, c ], the second eigenvector y ═ e, f, g, the third eigenvector z ═ h, i, j ], and then the first eigenvector x, the second eigenvector y, and the third eigenvector z are transversely spliced to obtain the target eigenvector w ═ a, b, c, e, f, g, h, i, j. It should be noted that the transverse splices mentioned later are similar to the above, and the transverse splices will not be described in detail.
It can be seen that in the embodiment, data information in each mode is processed independently, and in the processor, behavior texts are processed based on an attention mechanism, so that the first feature vector can reflect industries to which users belong better, and the precision of user labels is improved; in addition, the social network information is processed based on the GraphSAGE algorithm, so that when new social network information is added, the second characteristic vector of the node can be directly obtained without model training again, and further the calculation cost is reduced.
In some possible embodiments, randomly initializing the behavior text sequence using a bidirectional long-short term memory network to obtain a first feature vector may be: extracting keywords of each text in the N texts, randomly initializing the keywords of each text in the N texts by using a BilSTM network to obtain N characteristic vectors, and determining the weight of each text according to the N characteristic vectors; and carrying out weighting processing on the N eigenvectors according to the weight of each text to obtain a first eigenvector.
The random initialization of the keywords is prior art and will not be described.
When the keywords of each text in the N texts are randomly initialized by using the BiLSTM network, specifically, the keywords corresponding to each text are randomly initialized by using the forward LSTM of the BiLSTM network to obtain forward feature vectors corresponding to each text, the keywords corresponding to each text are randomly initialized by using the backward LSTM of the BiLSTM network to obtain backward feature vectors corresponding to each text, and the forward feature vectors and the backward feature vectors corresponding to each text in the N texts are transversely spliced to obtain N feature vectors.
For example, if a keyword corresponding to a certain text is "china", the keyword "china" is randomly initialized using forward LSTM to obtain forward feature vector h1 ═ s1, s2, s3, the keyword "china" is randomly initialized using backward LSTM to obtain backward feature vector h1 ═ m1, m2, m3, then h1 and h2 are transversely spliced to obtain feature vector p ═ s1, s2, s3, m1, m2, m3 of the keyword, and the feature vector of the text is obtained.
The above shows a case that one text includes one keyword, and for a case that the text includes a plurality of keywords, the feature vector p corresponding to each keyword is obtained according to the above random initialization manner, and then, the plurality of feature vectors corresponding to the plurality of keywords are transversely spliced to obtain the feature vector corresponding to the text.
It can be seen that, in the implementation mode, the BilSTM network is adopted to initialize the behavior text, the past and future information of the current state is obtained by utilizing the positive and negative time sequence directions in the text, the semantic features are enriched, and the feature vector obtained after initialization is more in line with the behavior of the user; and moreover, the weight of each text is obtained firstly, the weight reflects the contribution degree of the text to the industry to which the user belongs, and the accuracy of the industry to which the user belongs is represented by further improving the first feature vector by performing weighting processing according to the weight.
Optionally, before extracting keywords from each text in the N texts, denoising the N texts according to a pre-constructed industry dictionary, and filtering out texts irrelevant to the industry from the N texts to obtain T texts; randomly initializing keywords of each behavior text in the T texts by using a BilSTM network to obtain T feature vectors; determining the weight of each text according to the T feature vectors; and carrying out weighting processing on the T characteristic vectors according to the weight of each text to obtain the first characteristic vector.
Wherein, the keywords of each behavior text in the T texts randomly initialized by using the BilSTM network are consistent with the random initialization process and are not described,
it can be seen that in the embodiment, the behavior text sequence is denoised first, so that irrelevant texts are filtered, invalid data are prevented from being processed, and the generation efficiency of the user tag is improved; and moreover, the weight of each text is obtained, the weight reflects the contribution degree of the text to the industry to which the user belongs, and the accuracy of the first feature vector is further improved by performing weighting processing according to the weight.
The following describes in detail the process of calculating the weight of each text and weighting a plurality of feature vectors with reference to fig. 3C, where the feature vectors are T feature vectors of T texts obtained after denoising.
As shown in fig. 3C, after the feature vector of each text is obtained, based on the attention mechanism Attentionmechanism, the weight corresponding to each text is obtained according to the T feature vectors, and then, the T feature vectors are weighted based on the weight corresponding to each text, so as to obtain a first feature vector.
Wherein the first feature vector can be obtained by formula (1).
h
tIs the feature vector of the T-th text in the T texts, and W is a weight matrix, i.e. W ═ W
1,w
2,…,w
T),b
tFor the corresponding bias of the feature vector, α
tAnd C is a first feature vector.
In some possible embodiments, the inductive learning algorithm may be a graphsage algorithm, and the processing the social network information according to the inductive learning algorithm to obtain the second feature vector may be: and processing the social network information according to the graphsage algorithm to obtain a second feature vector. Namely, generating a social graph network according to the social network information; sampling neighbor nodes of a target node according to a preset sampling ratio to obtain R neighbor nodes, wherein the target node is a node corresponding to the user in the social graph network, and the sampling ratio is the ratio of the neighbor nodes to be sampled to all the neighbor nodes; weighting the feature vectors of the R neighbor nodes to obtain a fourth feature vector, wherein the feature vectors of the R neighbor nodes are obtained according to the feature vectors of the neighbor nodes corresponding to the search depth; and splicing the feature vector of the target node and the fourth feature direction to obtain the second feature vector.
The weighting process for the feature vectors of the R neighbor nodes may be performed by aggregating the R neighbor nodes through a pooling function, where the pooling function may include posing, mean, and the like.
It can be seen that, in the embodiment, the GraphSAGE algorithm is used to fuse the social network information of each user, and when a node is newly added in the social graph network, the feature vector of the node can be directly generated, the whole model does not need to be retrained, so that the operation overhead during fusion is reduced, and the generalization performance during fusion is improved.
The following describes the processing procedure of the GraphSAGE algorithm in detail with reference to fig. 3D and 3E.
Referring to fig. 3D, fig. 3D is a social graph network of a user a, a user B, a user C, a user D, a user E, and a user F, where a node a is a target node, and a sampling ratio is assumed to be 1, that is, all neighboring nodes participate in the calculation of the feature vector of the node a, as shown in fig. 3E, the feature vectors of the node B, the node C, and the node D
And
weighting to obtain a fourth feature vector
Then, to
And the feature vector of node A
Splicing to obtain a second feature vector of the node A
Wherein the content of the first and second substances,
and
can be calculated from the feature vectors of the neighboring nodes to node B, node C and node D.
Wherein the preset search depth determines the neighbor depth to the target node. For example, when the search depth is 1, only the feature vector of the neighboring node is needed to calculate the second feature vector, and when the search depth is 2, as shown in fig. 3E, the feature vector of each neighboring node is calculated by the feature vector of the neighboring node, and then the feature vector of the target node is calculated by using the feature vectors of the neighboring nodes.
In some possible embodiments, after obtaining the user tag of the user, the method further includes: and using the user tag of the user to perform personalized recommendation for the user.
Referring to fig. 4, fig. 4 is a schematic diagram of another user tag generation method provided in an embodiment of the present application, where the same contents as those in the embodiment shown in fig. 3B are not described again here, and the method includes, but is not limited to, the following steps:
401: and acquiring a behavior text sequence of a user, and processing the behavior text sequence by using a bidirectional long-short term memory network (BilSTM) random initialization to obtain a first feature vector.
402: and acquiring social network information of the user, and processing the social network information according to an inductive learning algorithm to obtain a second feature vector.
403: and acquiring social statistics information of the user, and acquiring a third feature vector according to the social statistics information.
404: and transversely splicing the first feature vector, the second feature vector and the third feature vector to obtain the user label of the user.
405: and carrying out personalized recommendation on the user according to the label of the user.
The personalized recommendation may be an advertisement recommendation, a video recommendation, a news recommendation, a music recommendation, etc.
It can be seen that in the embodiment, data information in each mode is processed independently, and in the processor, behavior texts are processed based on an attention mechanism, so that the first feature vector can reflect industries to which users belong better, and the precision of user labels is improved; in addition, the social network information is processed based on the GraphSAGE algorithm, so that when new social network information is added, the second characteristic vector of the node can be directly obtained without model training again, and further the calculation cost is reduced; because the generated user label is more accurate, the personalized recommendation can better meet the actual requirements of the user.
It should be noted that, for the specific implementation of the steps of the method shown in fig. 4, reference may be made to the specific implementation of the method described in fig. 3B, and a description thereof is omitted here.
In some possible implementations, the user tag generation method provided in the embodiment of the present application is applied to a tag generation model as shown in fig. 5, where the tag generation model includes a first neural network, a second neural network, and a third neural network.
The first neural network is a bidirectional BilSTM network based on an attention mechanism and is used for processing an input behavior text sequence to obtain a first feature vector (weighted text vector);
the second neural network is obtained based on GraphSAGE algorithm training and is used for processing input social network information to obtain a second feature vector (social network vector), and specifically, the network is obtained by performing unsupervised optimization by using a loss function in formula (2):
J
G(Z
u) Is destination node z
uCorresponding loss, z
vTo be within a specified depth z
uσ is a sigmoid function, Q is the number of negative samples, P
nIs a distribution function of negative samples, v
nIs satisfying P
nThe set of negative examples of the distribution is,
to obey P
nAll negative examples v of the distribution
nThe expectation is that.
It should be noted that, when training the second neural network, a training data set (social graph network) needs to be constructed, and when constructing the social graph network, generally, only the neighboring node whose target node a is adjacent to and has a depth of 2 hops or 3 hops needs to be sampled, so as to reduce the calculation overhead as much as possible without reducing the performance of the algorithm.
The third neural network is a concat network and is used for concat connection of the plurality of sub-vectors in the social statistics dimension to obtain a third feature vector (userprofile vector).
And finally, performing dimensionality splicing on the first feature vector, the second feature vector and the third feature vector based on the label generation model to obtain a target feature vector, performing a series of nonlinear activation, and inputting the target feature vector subjected to the nonlinear activation into a softmax classifier to obtain a label of the user.
Fig. 6 is a schematic structural diagram of a user tag generating apparatus according to an embodiment of the present application, where the apparatus 600 includes a processor, a memory, a communication interface, and one or more programs, where the one or more programs are different from the one or more application programs, and the one or more programs are stored in the memory and configured to be executed by the processor, and the programs include instructions for:
acquiring a behavior text sequence of a user, and randomly initializing the behavior text sequence by using a bidirectional long-short term memory network (BilSTM) to obtain a first feature vector, wherein the behavior text sequence is a sequence formed by texts which are generated according to a time sequence and used for representing user behaviors;
acquiring social network information of the user, and processing the social network information according to an inductive learning algorithm to obtain a second feature vector;
obtaining social statistics information of the user, and obtaining a third feature vector according to the social statistics information;
and transversely splicing the first feature vector, the second feature vector and the third feature vector to obtain the user label of the user.
In some possible embodiments, the behavior text sequence includes N texts, the N texts are used for characterizing the behavior of the user at N moments, the N is an integer greater than 1, and before the behavior text sequence is randomly initialized by using the bidirectional long-short term memory network BiLSTM, the program further includes instructions for:
denoising the N texts according to a pre-constructed industry dictionary to obtain T texts, wherein T is not more than N and is an integer more than or equal to 1;
the randomly initializing the behavior text sequence by using a bidirectional long-short term memory network (BilSTM) to obtain a first feature vector, comprising the following steps of:
randomly initializing each text in the T texts by using the BilSTM to obtain T feature vectors;
determining the weight of each text in the T texts according to the T feature vectors;
and carrying out weighting processing on the T characteristic vectors according to the weight of each text to obtain the first characteristic vector.
In some possible embodiments, the above program is specifically configured to, in terms of randomly initializing each of the T texts using the BiLSTM to obtain T feature vectors, execute the following steps:
extracting a keyword corresponding to each text in the T texts;
randomly initializing keywords corresponding to each text by using the forward LSTM of the BilSTM to obtain a forward feature vector corresponding to each text, and randomly initializing keywords corresponding to each text by using the backward LSTM of the BilSTM to obtain a backward feature vector corresponding to each text;
and transversely splicing the forward characteristic vector and the backward characteristic vector corresponding to each text in the T texts to obtain T characteristic vectors.
In some possible embodiments, the inductive learning algorithm comprises a graphsage algorithm.
In some possible embodiments, the program is specifically configured to execute the following steps in processing the social network information according to an inductive learning algorithm to obtain a second feature vector:
processing the social network information according to the graphsage algorithm to obtain a second feature vector, which specifically includes:
generating a social graph network according to the social network information;
sampling neighbor nodes of a target node to obtain R neighbor nodes, wherein the target node is a node corresponding to the user in the social graph network;
weighting the feature vectors of the R neighbor nodes to obtain a fourth feature vector;
and transversely splicing the feature vector of the target node and the fourth feature vector to obtain the second feature vector.
In some possible embodiments, the social statistics information includes P items of information, and the program is specifically configured to execute the following steps in obtaining a third feature vector according to the social statistics information:
acquiring a sub-vector corresponding to each item of information in the P items of information to obtain P sub-vectors;
and splicing the P sub-vectors to obtain the third feature vector.
Fig. 7 is a block diagram of another user tag generation apparatus provided in an embodiment of the present application, where the apparatus 700 includes a processor 710, a transceiver 720, and at least one circuit 730, and the processor and the transceiver are connected through the at least one circuit, where:
the transceiver 710 is configured to obtain a behavior text sequence, social network information, and social statistics information of a user, and send the behavior text sequence, the social network information, and the social statistics information to the processor 720, where the behavior text sequence is a sequence formed by texts generated according to a time sequence and used for representing behaviors of the user;
a processor 720, configured to randomly initialize the behavior text sequence by using a bidirectional long-short term memory network BiLSTM to obtain a first feature vector;
the processor 720 is further configured to process the social network information according to an inductive learning algorithm to obtain a second feature vector;
the processor 720 is further configured to obtain a third feature vector according to the social statistics information;
the processor 720 is further configured to transversely splice the first feature vector, the second feature vector, and the third feature vector to obtain a user tag of the user.
In some possible embodiments, the behavior text sequence includes N texts, where the N texts are used for characterizing behaviors of the user at N moments, where N is an integer greater than 1, and before the behavior text sequence is randomly initialized using a bidirectional long-short term memory network BiLSTM, the processor 720 is further configured to denoise the N texts according to a pre-constructed industry dictionary to obtain T texts, where T is less than or equal to N, and T is an integer greater than or equal to 1;
in randomly initializing the behavior text sequence using a bidirectional long-short term memory network bilst, to obtain a first feature vector, processor 720 is specifically configured to:
randomly initializing each text in the T texts by using the BilSTM to obtain T feature vectors;
determining the weight of each text in the T texts according to the T feature vectors;
and carrying out weighting processing on the T characteristic vectors according to the weight of each text to obtain the first characteristic vector.
In some possible embodiments, in randomly initializing each of the T texts using the bilst, resulting in T feature vectors, processor 720 is specifically configured to:
extracting a keyword corresponding to each text in the T texts;
randomly initializing keywords corresponding to each text by using the forward LSTM of the BilSTM to obtain a forward feature vector corresponding to each text, and randomly initializing keywords corresponding to each text by using the backward LSTM of the BilSTM to obtain a backward feature vector corresponding to each text;
and transversely splicing the forward characteristic vector and the backward characteristic vector corresponding to each text in the T texts to obtain T characteristic vectors.
In some possible embodiments, the inductive learning algorithm comprises a graphsage algorithm.
In some possible embodiments, in processing the social network information according to an inductive learning algorithm to obtain a second feature vector, the processor 720 is specifically configured to:
processing the social network information according to the graphsage algorithm to obtain a second feature vector, which is specifically used for:
generating a social graph network according to the social network information;
sampling neighbor nodes of a target node to obtain R neighbor nodes, wherein the target node is a node corresponding to the user in the social graph network;
weighting the feature vectors of the R neighbor nodes to obtain a fourth feature vector;
and transversely splicing the feature vector of the target node and the fourth feature vector to obtain the second feature vector.
In some possible embodiments, the social statistics information includes P items of information, and in obtaining the third feature vector according to the social statistics information, the processor 720 is specifically configured to:
acquiring a sub-vector corresponding to each item of information in the P items of information to obtain P sub-vectors;
and splicing the P sub-vectors to obtain the third feature vector.
Embodiments of the present application also provide a computer storage medium, where the computer storage medium stores a computer program, and the computer program is executed by a processor to implement part or all of the steps of any one of the user tag generation methods described in the above method embodiments.
Embodiments of the present application also provide a computer program product comprising a non-transitory computer readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of any of the user tag generation methods as recited in the above method embodiments.
It should be noted that, for simplicity of description, the above-mentioned method embodiments are described as a series of acts or combination of acts, but those skilled in the art will recognize that the present application is not limited by the order of acts described, as some steps may occur in other orders or concurrently depending on the application. Further, those skilled in the art should also appreciate that the embodiments described in the specification are exemplary embodiments and that the acts and modules referred to are not necessarily required in this application.
In the foregoing embodiments, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus may be implemented in other manners. For example, the above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one type of division of logical functions, and there may be other divisions when actually implementing, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted, or not implemented. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of some interfaces, devices or units, and may be an electric or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit may be implemented in the form of hardware, or may be implemented in the form of a software program module.
The integrated units, if implemented in the form of software program modules and sold or used as stand-alone products, may be stored in a computer readable memory. Based on such understanding, the technical solution of the present application may be substantially implemented or a part of or all or part of the technical solution contributing to the prior art may be embodied in the form of a software product stored in a memory, and including several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method described in the embodiments of the present application. And the aforementioned memory comprises: various media capable of storing program codes, such as a usb disk, a Read-Only memory (ROT), a random Access memory (RAT), a mobile hard disk, a magnetic disk, or an optical disk.
Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by associated hardware instructed by a program, which may be stored in a computer-readable memory, which may include: flash memory disks, Read-Only memory (ROT), random Access memory (RAT), magnetic or optical disks, etc.
The foregoing detailed description of the embodiments of the present application has been presented to illustrate the principles and implementations of the present application, and the above description of the embodiments is only provided to help understand the method and the core concept of the present application; meanwhile, for a person skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.
Claims (10)
1. A user tag generation method is characterized by comprising the following steps:
acquiring a behavior text sequence of a user, and randomly initializing the behavior text sequence by using a bidirectional long-short term memory network (BilSTM) to obtain a first feature vector, wherein the behavior text sequence is a sequence formed by texts which are generated according to a time sequence and used for representing user behaviors;
acquiring social network information of the user, and processing the social network information according to an inductive learning algorithm to obtain a second feature vector;
obtaining social statistics information of the user, and obtaining a third feature vector according to the social statistics information;
and transversely splicing the first feature vector, the second feature vector and the third feature vector to obtain the user label of the user.
2. The method of claim 1, wherein the behavior text sequence comprises N texts, the N texts being used for characterizing the behavior of the user at N moments in time, the N being an integer greater than 1, and wherein before the randomly initializing the behavior text sequence using a bidirectional long-short term memory network (bilst), the method further comprises:
denoising the N texts according to a pre-constructed industry dictionary to obtain T texts, wherein T is not more than N and is an integer more than or equal to 1;
the randomly initializing the behavior text sequence by using a bidirectional long-short term memory network (BilSTM) to obtain a first feature vector, comprising the following steps of:
randomly initializing each text in the T texts by using the BilSTM to obtain T feature vectors;
determining the weight of each text in the T texts according to the T feature vectors;
and carrying out weighting processing on the T characteristic vectors according to the weight of each text to obtain the first characteristic vector.
3. The method of claim 2, wherein the randomly initializing each of the T texts using the BiLSTM to obtain T feature vectors comprises:
extracting a keyword corresponding to each text in the T texts;
randomly initializing keywords corresponding to each text by using the forward LSTM of the BilSTM to obtain a forward feature vector corresponding to each text, and randomly initializing keywords corresponding to each text by using the backward LSTM of the BilSTM to obtain a backward feature vector corresponding to each text;
and transversely splicing the forward characteristic vector and the backward characteristic vector corresponding to each text in the T texts to obtain T characteristic vectors.
4. The method of any of claims 1-3, wherein the inductive learning algorithm comprises a graphsage algorithm.
5. The method of claim 4, wherein the processing the social networking information according to an inductive learning algorithm to obtain a second feature vector comprises:
processing the social network information according to the graphsage algorithm to obtain a second feature vector, which specifically includes:
generating a social graph network according to the social network information;
sampling neighbor nodes of a target node to obtain R neighbor nodes, wherein the target node is a node corresponding to the user in the social graph network;
weighting the feature vectors of the R neighbor nodes to obtain a fourth feature vector;
and transversely splicing the feature vector of the target node and the fourth feature vector to obtain the second feature vector.
6. The method according to any one of claims 1-5, wherein the social statistic information includes P items of information, and the deriving a third feature vector according to the social statistic information includes:
acquiring a sub-vector corresponding to each item of information in the P items of information to obtain P sub-vectors;
and splicing the P sub-vectors to obtain the third feature vector.
7. A user tag generation apparatus, comprising a processor, a transceiver and at least one circuit, the processor and the transceiver being connected via the at least one circuit;
the transceiver is used for acquiring a behavior text sequence, social network information and social statistics information of a user and sending the behavior text sequence, the social network information and the social statistics information to the processor, wherein the behavior text sequence is a sequence formed by texts which are generated according to a time sequence and used for representing user behaviors;
the processor is used for randomly initializing the behavior text sequence by using a bidirectional long-short term memory network (BilSTM) to obtain a first feature vector;
the processor is further used for processing the social network information according to an inductive learning algorithm to obtain a second feature vector;
the processor is further configured to obtain a third feature vector according to the social statistics information;
the processor is further configured to transversely splice the first feature vector, the second feature vector, and the third feature vector to obtain a user tag of the user.
8. The apparatus of claim 7, wherein the sequence of behavioral texts comprises N texts, the N texts being for characterizing behaviors of the user at N moments in time, the N being an integer greater than 1, and wherein before the sequence of behavioral texts is randomly initialized using a bidirectional long-short term memory network (BilTM), the processor is further configured to denoise the N texts according to a pre-constructed industry dictionary to obtain T texts, T ≦ N, the T being an integer greater than or equal to 1;
in terms of randomly initializing the behavior text sequence by using a bidirectional long-short term memory network (BilSTM) to obtain a first feature vector, the processor is specifically configured to:
randomly initializing each text in the T texts by using the BilSTM to obtain T feature vectors;
determining the weight of each text in the T texts according to the T feature vectors;
and carrying out weighting processing on the T characteristic vectors according to the weight of each text to obtain the first characteristic vector.
9. The apparatus of claim 8, wherein, in randomly initializing each of the T texts using the BiLSTM to obtain T feature vectors, the processor is specifically configured to:
extracting a keyword corresponding to each text in the T texts;
randomly initializing keywords corresponding to each text by using the forward LSTM of the BilSTM to obtain a forward feature vector corresponding to each text, and randomly initializing keywords corresponding to each text by using the backward LSTM of the BilSTM to obtain a backward feature vector corresponding to each text;
and transversely splicing the forward characteristic vector and the backward characteristic vector corresponding to each text in the T texts to obtain T characteristic vectors.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program which is executed by a processor to implement the method according to any one of claims 1-6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911000656.2A CN110781407A (en) | 2019-10-21 | 2019-10-21 | User label generation method and device and computer readable storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911000656.2A CN110781407A (en) | 2019-10-21 | 2019-10-21 | User label generation method and device and computer readable storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110781407A true CN110781407A (en) | 2020-02-11 |
Family
ID=69386111
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911000656.2A Pending CN110781407A (en) | 2019-10-21 | 2019-10-21 | User label generation method and device and computer readable storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110781407A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111488807A (en) * | 2020-03-29 | 2020-08-04 | 复旦大学 | Video description generation system based on graph convolution network |
CN112347316A (en) * | 2020-10-21 | 2021-02-09 | 上海淇玥信息技术有限公司 | GraphSAGE-based bad preference behavior detection method and device and electronic equipment |
CN112559904A (en) * | 2020-12-16 | 2021-03-26 | 中国计量大学 | Conversational social recommendation method based on door mechanism and multi-modal graph network |
CN112765481A (en) * | 2020-06-29 | 2021-05-07 | 腾讯科技(深圳)有限公司 | Data processing method and device, computer and readable storage medium |
CN113407845A (en) * | 2021-07-14 | 2021-09-17 | 上海明略人工智能(集团)有限公司 | Method and device for information recommendation, electronic equipment and storage medium |
CN113486235A (en) * | 2021-05-31 | 2021-10-08 | 杭州摸象大数据科技有限公司 | Method and system for identifying user interest |
CN113553433A (en) * | 2021-09-17 | 2021-10-26 | 平安科技(深圳)有限公司 | Product classification method, device, medium and terminal equipment based on artificial intelligence |
CN113591886A (en) * | 2020-04-30 | 2021-11-02 | 北京百度网讯科技有限公司 | Method, apparatus, device and computer readable storage medium for information classification |
CN114493535A (en) * | 2022-01-28 | 2022-05-13 | 北京乐开科技有限责任公司 | Asset model-based data center system for data driving |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110103682A1 (en) * | 2009-10-29 | 2011-05-05 | Xerox Corporation | Multi-modality classification for one-class classification in social networks |
CN108763319A (en) * | 2018-04-28 | 2018-11-06 | 中国科学院自动化研究所 | Merge the social robot detection method and system of user behavior and text message |
US20180336457A1 (en) * | 2017-05-17 | 2018-11-22 | Facebook, Inc. | Semi-Supervised Learning via Deep Label Propagation |
KR20190023218A (en) * | 2017-08-28 | 2019-03-08 | 서울시립대학교 산학협력단 | Apparatus and method for providing body tag recognizing model, and apparatus for applying body tag recognizing model |
CN110097125A (en) * | 2019-05-07 | 2019-08-06 | 郑州轻工业学院 | A kind of across a network account correlating method indicated based on insertion |
CN110197389A (en) * | 2019-03-04 | 2019-09-03 | 腾讯科技(深圳)有限公司 | A kind of user identification method and device |
CN110209823A (en) * | 2019-06-12 | 2019-09-06 | 齐鲁工业大学 | A kind of multi-tag file classification method and system |
-
2019
- 2019-10-21 CN CN201911000656.2A patent/CN110781407A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110103682A1 (en) * | 2009-10-29 | 2011-05-05 | Xerox Corporation | Multi-modality classification for one-class classification in social networks |
US20180336457A1 (en) * | 2017-05-17 | 2018-11-22 | Facebook, Inc. | Semi-Supervised Learning via Deep Label Propagation |
KR20190023218A (en) * | 2017-08-28 | 2019-03-08 | 서울시립대학교 산학협력단 | Apparatus and method for providing body tag recognizing model, and apparatus for applying body tag recognizing model |
CN108763319A (en) * | 2018-04-28 | 2018-11-06 | 中国科学院自动化研究所 | Merge the social robot detection method and system of user behavior and text message |
CN110197389A (en) * | 2019-03-04 | 2019-09-03 | 腾讯科技(深圳)有限公司 | A kind of user identification method and device |
CN110097125A (en) * | 2019-05-07 | 2019-08-06 | 郑州轻工业学院 | A kind of across a network account correlating method indicated based on insertion |
CN110209823A (en) * | 2019-06-12 | 2019-09-06 | 齐鲁工业大学 | A kind of multi-tag file classification method and system |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111488807B (en) * | 2020-03-29 | 2023-10-10 | 复旦大学 | Video description generation system based on graph rolling network |
CN111488807A (en) * | 2020-03-29 | 2020-08-04 | 复旦大学 | Video description generation system based on graph convolution network |
CN113591886A (en) * | 2020-04-30 | 2021-11-02 | 北京百度网讯科技有限公司 | Method, apparatus, device and computer readable storage medium for information classification |
CN113591886B (en) * | 2020-04-30 | 2023-11-07 | 北京百度网讯科技有限公司 | Method, apparatus, device and computer readable storage medium for information classification |
CN112765481A (en) * | 2020-06-29 | 2021-05-07 | 腾讯科技(深圳)有限公司 | Data processing method and device, computer and readable storage medium |
CN112765481B (en) * | 2020-06-29 | 2023-09-19 | 腾讯科技(深圳)有限公司 | Data processing method, device, computer and readable storage medium |
CN112347316A (en) * | 2020-10-21 | 2021-02-09 | 上海淇玥信息技术有限公司 | GraphSAGE-based bad preference behavior detection method and device and electronic equipment |
CN112559904B (en) * | 2020-12-16 | 2021-10-22 | 中国计量大学 | Conversational social recommendation method based on door mechanism and multi-modal graph network |
CN112559904A (en) * | 2020-12-16 | 2021-03-26 | 中国计量大学 | Conversational social recommendation method based on door mechanism and multi-modal graph network |
CN113486235A (en) * | 2021-05-31 | 2021-10-08 | 杭州摸象大数据科技有限公司 | Method and system for identifying user interest |
CN113407845A (en) * | 2021-07-14 | 2021-09-17 | 上海明略人工智能(集团)有限公司 | Method and device for information recommendation, electronic equipment and storage medium |
CN113553433A (en) * | 2021-09-17 | 2021-10-26 | 平安科技(深圳)有限公司 | Product classification method, device, medium and terminal equipment based on artificial intelligence |
CN113553433B (en) * | 2021-09-17 | 2022-01-07 | 平安科技(深圳)有限公司 | Product classification method, device, medium and terminal equipment based on artificial intelligence |
CN114493535A (en) * | 2022-01-28 | 2022-05-13 | 北京乐开科技有限责任公司 | Asset model-based data center system for data driving |
CN114493535B (en) * | 2022-01-28 | 2022-09-06 | 北京乐开科技有限责任公司 | Asset model-based data center system for data driving |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110781407A (en) | User label generation method and device and computer readable storage medium | |
US20210256403A1 (en) | Recommendation method and apparatus | |
CN111931062A (en) | Training method and related device of information recommendation model | |
CN111382555B (en) | Data processing method, medium, device and computing equipment | |
CN111125495A (en) | Information recommendation method, equipment and storage medium | |
CN107918778B (en) | Information matching method and related device | |
CN111625645B (en) | Training method and device for text generation model and electronic equipment | |
CN108959323B (en) | Video classification method and device | |
CN111125386B (en) | Media resource processing method and device, storage medium and electronic device | |
CN113254711B (en) | Interactive image display method and device, computer equipment and storage medium | |
CN112528164B (en) | User collaborative filtering recall method and device | |
CN116601626A (en) | Personal knowledge graph construction method and device and related equipment | |
CN113255798A (en) | Classification model training method, device, equipment and medium | |
CN113392179A (en) | Text labeling method and device, electronic equipment and storage medium | |
CN109819002B (en) | Data pushing method and device, storage medium and electronic device | |
CN112149604A (en) | Training method of video feature extraction model, video recommendation method and device | |
CN113779380A (en) | Cross-domain recommendation method, device and equipment, and content recommendation method, device and equipment | |
CN114398973B (en) | Media content tag identification method, device, equipment and storage medium | |
CN116541592A (en) | Vector generation method, information recommendation method, device, equipment and medium | |
CN110851629A (en) | Image retrieval method | |
CN112948626B (en) | Video processing method and device, electronic equipment and computer readable storage medium | |
CN115391638A (en) | Recommendation model training method and device based on social network | |
CN112734519B (en) | Commodity recommendation method based on convolution self-encoder network | |
CN115238191A (en) | Object recommendation method and device | |
CN114493674A (en) | Advertisement click rate prediction model and method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40022449 Country of ref document: HK |
|
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |