CN114065014A - Information matching method, device, equipment and storage medium - Google Patents
Information matching method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN114065014A CN114065014A CN202010761650.3A CN202010761650A CN114065014A CN 114065014 A CN114065014 A CN 114065014A CN 202010761650 A CN202010761650 A CN 202010761650A CN 114065014 A CN114065014 A CN 114065014A
- Authority
- CN
- China
- Prior art keywords
- user
- data
- recommended
- vector
- semantic
- 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 52
- 239000013598 vector Substances 0.000 claims abstract description 297
- 238000000605 extraction Methods 0.000 claims abstract description 73
- 230000006870 function Effects 0.000 claims description 18
- 239000011159 matrix material Substances 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 16
- 238000013528 artificial neural network Methods 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 4
- 230000008034 disappearance Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000012549 training Methods 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
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- 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
Abstract
The embodiment of the disclosure relates to an information matching method, an information matching device, information matching equipment and a storage medium. The method comprises the following steps: acquiring the characteristics of a user to acquire recommended data and the characteristics of data to be recommended; respectively inputting the characteristics of the user and the characteristics of the to-be-recommended data into a user vector extraction branch and a data vector extraction branch of a user semantic matching model, and acquiring semantic representations corresponding to the characteristics of the user and the characteristics of the to-be-recommended data; and generating a matching result of the user and the to-be-recommended data based on the semantic representation corresponding to the characteristics of the user and the semantic representation corresponding to the characteristics of the to-be-recommended data. The method solves the problem that semantic representation quality is not high when the number of network layers is shallow, and improves semantic representation precision by learning semantic representations corresponding to the characteristics of the user and the characteristics of the to-be-recommended data through a new network structure.
Description
Technical Field
The embodiment of the disclosure relates to the technical field of computers, and in particular relates to an information matching method, an information matching device, information matching equipment and a storage medium.
Background
With the continuous development of internet technology, the amount of information on a network is increased explosively, but for a user, only a small part of information is available, most of the information is redundant, the difficulty of the user for inquiring the available information is increased by the redundant information, and in order to improve the efficiency of the user for inquiring the information, a data recommendation system is generally adopted for recommending data for the user at present.
In the prior art, a data recommendation system needs to transform user characteristics and data characteristics respectively to obtain vector representations. The double-tower model has excellent performance and is widely applied to data recommendation services of various large Internet companies. However, when learning a higher-order semantic representation, the learning effect of the two-tower based recommendation model is less than ideal. The existing recommendation model based on the double tower is generally shallow in the network layer number, so that the learned semantic representation is inaccurate.
Disclosure of Invention
The embodiment of the disclosure provides an information matching method, an information matching device, information matching equipment and a storage medium, and aims to solve the problem that semantic representation is inaccurate when learning is performed when the number of network layers is shallow in the related art. The technical scheme of the disclosure is as follows:
according to a first aspect of the embodiments of the present disclosure, there is provided an information matching method, the method including:
acquiring the characteristics of a user to acquire recommended data and the characteristics of data to be recommended;
inputting the characteristics of the user and the characteristics of the data to be recommended into a user vector extraction branch and a data vector extraction branch of a user semantic matching model respectively, wherein the user vector extraction branch and the data vector extraction branch are used for converting the input characteristics into initial characteristic vectors respectively, generating intermediate characteristic vectors of each stage based on the initial characteristic vectors through at least two stages of residual error network blocks which are arranged respectively, and the intermediate characteristic vectors of each stage are used for generating semantic representation of corresponding characteristics;
and generating a matching result of the user and the to-be-recommended data based on the semantic representation corresponding to the characteristics of the user and the semantic representation corresponding to the characteristics of the to-be-recommended data.
Optionally, the user semantic matching model includes a user vector extraction branch and a data vector extraction branch, where the user vector extraction branch and the data vector extraction branch each include an input layer, at least two stages of residual error network blocks, and a vector convergence layer, and the two branches converge at a matching output layer;
after the features of the user and the features of the data to be recommended are respectively input to a user vector extraction branch and a data vector extraction branch of a user semantic matching model, the method further comprises the following steps:
converting the characteristics of the user and the characteristics of the data to be recommended into initial user characteristic vectors and initial data characteristic vectors respectively through input layers contained in the branches;
respectively and sequentially processing the initial user characteristic vector and the initial data characteristic vector through at least two levels of residual error network blocks contained in each branch to obtain a middle user characteristic vector and a middle data characteristic vector corresponding to each level of residual error network block;
and respectively processing the intermediate user characteristic vectors and the intermediate data characteristic vectors corresponding to the residual network blocks at each level through a vector convergence layer contained in each branch to obtain semantic representations corresponding to the characteristics of the user and the characteristics of the data to be recommended.
Optionally, the step of processing, through a vector convergence layer included in each branch, the intermediate user feature vector and the intermediate data feature vector corresponding to each level of residual network block respectively to obtain semantic representations corresponding to the features of the user and the features of the data to be recommended includes:
and calculating semantic representations corresponding to the characteristics of the user and the characteristics of the data to be recommended according to the intermediate characteristic vector output by the residual network block of at least one set level and the intermediate characteristic vector output by the residual network block of the last layer.
Optionally, the representation of the residual network block is as follows:
xl+1＝xl+F(xl,wl)，
where l is the hierarchy of residual network blocks, xlIs the output of the l-th stage residual network block, F (x)l,wl) Is formed by at least one layer of neural network as a residual function, wlIs a weight matrix.
Optionally, the activation function used by each residual network block is as follows:
f(x)＝x·sigmoid(x)。
optionally, the step of calculating semantic representations corresponding to the features of the user and the features of the data to be recommended according to the intermediate feature vector output by the at least one set-level residual network block and the intermediate feature vector output by the last-layer residual network block includes:
connecting the intermediate characteristic vector output by at least one residual error network block close to the input layer and the intermediate characteristic vector output by the last layer of residual error network block in parallel;
multiplying the parallel result by the parallel matrix, and adding the multiplied result and the offset vector to obtain semantic representations corresponding to the characteristics of the user and the characteristics of the to-be-recommended data, wherein the calculation mode is as follows:
xL＝wL[x1,x2,…,xL-1]+bL，
wherein x is1,x2… is the intermediate feature vector, x, of the residual network block output near the input layerL-1Is the intermediate feature vector, w, output by the last layer of residual network blockLIs a parallel matrix, bLIs a bias vector.
Optionally, the information matching method further includes:
inputting the characteristics of the user and the characteristics of the data to be recommended into a user semantic matching model;
semantic matching is carried out on the characteristics of the user and the characteristics of the to-be-recommended data through the user semantic matching model, and matching scores of the user and the to-be-recommended data are obtained;
the higher the matching score is, the higher the matching degree of the user and the data to be recommended is;
the characteristics of the user comprise user attribute information and user browsing information, and the characteristics of the data to be recommended comprise data attribute information.
According to a second aspect of the embodiments of the present disclosure, there is provided an information matching apparatus, the apparatus including:
the characteristic acquisition module is used for acquiring the characteristics of a user to acquire recommended data and the characteristics of the data to be recommended;
the characteristic input module is used for respectively inputting the characteristics of the user and the characteristics of the data to be recommended to a user vector extraction branch and a data vector extraction branch of a user semantic matching model, wherein the user vector extraction branch and the data vector extraction branch are respectively used for converting the input characteristics into initial characteristic vectors, intermediate characteristic vectors of each stage are generated on the basis of the initial characteristic vectors through at least two stages of residual error network blocks which are respectively arranged, and the intermediate characteristic vectors of each stage are used for generating semantic representation of corresponding characteristics;
and the matching result generation module is used for generating a matching result of the user and the to-be-recommended data based on the semantic representation corresponding to the characteristics of the user and the semantic representation corresponding to the characteristics of the to-be-recommended data.
Optionally, the user semantic matching model includes a user vector extraction branch and a data vector extraction branch, where the user vector extraction branch and the data vector extraction branch each include an input layer, at least two stages of residual error network blocks, and a vector convergence layer, and the two branches converge at a matching output layer;
the information matching device further comprises:
the initial feature vector acquisition module is used for respectively inputting the features of the user and the features of the to-be-recommended data into a user vector extraction branch and a data vector extraction branch of a user semantic matching model, and respectively converting the features of the user and the features of the to-be-recommended data into an initial user feature vector and an initial data feature vector through an input layer contained in each branch;
the intermediate characteristic vector acquisition module is used for respectively and sequentially processing the initial user characteristic vector and the initial data characteristic vector through at least two stages of residual error network blocks contained in each branch to obtain intermediate user characteristic vectors and intermediate data characteristic vectors corresponding to the residual error network blocks at each stage;
and the semantic representation acquisition module is used for respectively processing the intermediate user characteristic vectors and the intermediate data characteristic vectors corresponding to the residual network blocks at each level through the vector convergence layer contained in each branch to obtain semantic representations corresponding to the characteristics of the user and the characteristics of the data to be recommended.
Optionally, the semantic representation obtaining module includes:
and the semantic representation acquisition unit is used for calculating semantic representations corresponding to the characteristics of the user and the characteristics of the data to be recommended according to the intermediate characteristic vector output by the residual error network block of at least one set level and the intermediate characteristic vector output by the residual error network block of the last layer.
Optionally, the representation of the residual network block is as follows:
xl+1＝xl+F(xl,wl)，
where l is the hierarchy of residual network blocks, xlIs the output of the l-th stage residual network block, F (x)l,wl) Is formed by at least one layer of neural network as a residual function, wlIs a weight matrix.
Optionally, the activation function used by each residual network block is as follows:
f(x)＝x·sigmoid(x)。
optionally, the semantic representation acquiring unit is specifically configured to:
connecting the intermediate characteristic vector output by at least one residual error network block close to the input layer and the intermediate characteristic vector output by the last layer of residual error network block in parallel;
multiplying the parallel result by the parallel matrix, and adding the multiplied result and the offset vector to obtain semantic representations corresponding to the characteristics of the user and the characteristics of the to-be-recommended data, wherein the calculation mode is as follows:
xL＝wL[x1,x2,…,xL-1]+bL，
wherein x is1,x2… is the intermediate feature vector, x, of the residual network block output near the input layerL-1Is the last layer of residual netIntermediate eigenvectors, w, of the envelope block outputLIs a parallel matrix, bLIs a bias vector.
Optionally, the information matching apparatus is further configured to:
inputting the characteristics of the user and the characteristics of the data to be recommended into a user semantic matching model;
semantic matching is carried out on the characteristics of the user and the characteristics of the to-be-recommended data through the user semantic matching model, and matching scores of the user and the to-be-recommended data are obtained;
the higher the matching score is, the higher the matching degree of the user and the data to be recommended is;
the characteristics of the user comprise user attribute information and user browsing information, and the characteristics of the data to be recommended comprise data attribute information.
According to a third aspect of the embodiments of the present disclosure, there is provided an electronic apparatus including: a processor; a memory for storing the processor-executable instructions; wherein the processor is configured to execute the instructions to implement the information matching method according to any one of the embodiments of the present disclosure.
According to a fourth aspect of the embodiments of the present disclosure, there is provided a storage medium, wherein instructions, when executed by a processor of a server, enable the server to perform the information matching method according to any of the embodiments disclosed herein.
According to a fifth aspect of the embodiments of the present disclosure, there is provided a computer program product for use in conjunction with an electronic device, the computer program product comprising a computer-readable storage medium and a computer program mechanism embedded therein, the computer program being loaded into and executed by a computer to implement the information matching method according to any of the embodiments of the present disclosure.
The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects:
the problem that the semantic representation quality is not high when the number of network layers is shallow in the prior art is solved, and the learning capability of the model on high-order semantic representation is improved by carrying out structural optimization on the double-tower model.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure and are not to be construed as limiting the disclosure.
Fig. 1 is a flow chart illustrating a method of information matching according to an example embodiment.
Fig. 2a is a flow chart illustrating a method of information matching according to an example embodiment.
FIG. 2b is a diagram illustrating a user semantic matching model architecture in accordance with an illustrative embodiment.
FIG. 2c is a diagram illustrating an activation function employed by a user semantic matching model according to an exemplary embodiment.
Fig. 3 is a flow chart illustrating a method of information matching according to an example embodiment.
Fig. 4 is a flow chart illustrating a method of information matching according to an example embodiment.
Fig. 5 is a block diagram illustrating an information matching apparatus according to an example embodiment.
Fig. 6 is a schematic structural diagram of an electronic device according to an exemplary embodiment.
Detailed Description
In order to make the technical solutions of the present disclosure better understood by those of ordinary skill in the art, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.
It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein are capable of operation in sequences other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.
Fig. 1 is a flowchart illustrating an information matching method according to an exemplary embodiment, where the information matching method is used in an electronic device and executed by a processor configured in the electronic device, as shown in fig. 1, and the method includes the following steps:
in step S11, the characteristics of the user who is to acquire the recommendation data and the characteristics of the data to be recommended are acquired.
The characteristics of the user are information related to the user or historical behaviors of the user, and for example, the characteristics of the user can be information of the age, sex, region, browsed webpage, purchased goods and the like of the user; the characteristic of the data to be recommended is information related to the data to be recommended, and for example, when the data to be recommended is a short video type, the characteristic of the data to be recommended may be information such as a video name, a video category, and a publisher.
In the embodiment of the disclosure, in order to obtain a matching degree between a user and data to be recommended, first, characteristics of the user who needs to obtain the recommended data and characteristics of the data to be recommended are obtained, where the characteristics of the user refer to information capable of representing characteristics of the current user, such as age, gender, and location of the user, and the characteristics of the data to be recommended refer to information capable of representing characteristics of the currently obtained data to be recommended, such as a video name, a video type, and a video publisher to be recommended. Illustratively, information such as the age, sex, region, browsed web page and purchased goods of the user is obtained, and information such as the name, category and publisher of the video to be recommended is obtained.
In step S12, the features of the user and the features of the data to be recommended are respectively input to a user vector extraction branch and a data vector extraction branch of the user semantic matching model, where the user vector extraction branch and the data vector extraction branch are respectively used to convert the input features into initial feature vectors, and generate intermediate feature vectors of each stage based on the initial feature vectors through at least two stages of residual network blocks, and the intermediate feature vectors of each stage are used to generate semantic representations of corresponding features.
The user semantic matching model is used for processing the features of the user and the features of the to-be-recommended data respectively to obtain a feature vector capable of representing the features of the user and a feature vector capable of representing the features of the to-be-recommended data, and then processing the feature vectors to obtain semantic representations corresponding to the features of the user and semantic representations corresponding to the features of the to-be-recommended data, so that the matching degree of the user and the to-be-recommended data is further calculated according to the two semantic representations, and basis is provided for recommending the data to the user.
In the embodiment of the present disclosure, the features of the user acquired in step S11 are input to the user vector extraction branch of the user semantic matching model, the features of the data to be recommended are input to the data vector extraction branch of the user semantic matching model, the input features are converted into initial feature vectors by the user vector extraction branch and the data vector extraction branch, then intermediate feature vectors corresponding to the residual network blocks of each stage are generated based on the initial feature vectors by at least two stages of residual network blocks that are set, and finally semantic representations of the corresponding features are generated based on the intermediate feature vectors of each stage.
In step S13, a matching result between the user and the data to be recommended is generated based on the semantic representation corresponding to the feature of the user and the semantic representation corresponding to the feature of the data to be recommended.
In the embodiment of the disclosure, after the semantic representation corresponding to the features of the user and the semantic representation corresponding to the features of the to-be-recommended data are obtained, the matching degree between the user and the to-be-recommended data is calculated according to the semantic features. Illustratively, the semantic representation corresponding to the features of the user and the semantic representation corresponding to the features of the to-be-recommended data are simultaneously input to a matching output layer of the user semantic configuration model, the semantic representation corresponding to the features of the user and the semantic representation corresponding to the features of the to-be-recommended data are matched by the matching output layer in a set mode, and a final matching score is obtained.
According to the technical scheme of the embodiment, the method comprises the steps of firstly obtaining the characteristics of a user needing to obtain recommended data and the characteristics of the to-be-recommended data, then respectively inputting the characteristics of the user and the characteristics of the to-be-recommended data into a user vector extraction branch and a data vector extraction branch of a user semantic matching model to obtain semantic representations corresponding to the characteristics of the user and the characteristics of the to-be-recommended data, finally generating a matching result of the user and the to-be-recommended data based on the semantic representations corresponding to the characteristics of the user and the characteristics of the to-be-recommended data, learning the semantic representations corresponding to the characteristics of the user and the characteristics of the to-be-recommended data through a new network structure, and improving the precision of the to-be-expressed data.
Fig. 2a is a flowchart illustrating an information matching method according to an exemplary embodiment, where the information matching method is used in an electronic device and executed by a processor configured in the electronic device, as shown in fig. 2a, and the method includes the following steps:
in step S21, the characteristics of the user who is to acquire the recommendation data and the characteristics of the data to be recommended are acquired.
In step S22, the features of the user and the features of the data to be recommended are input to the user vector extraction branch and the data vector extraction branch of the user semantic matching model, respectively.
In an implementation manner of the embodiment of the present disclosure, optionally, the user semantic matching model includes a user vector extraction branch and a data vector extraction branch, where the user vector extraction branch and the data vector extraction branch each include an input layer, at least two levels of residual network blocks, and a vector convergence layer, and the two branches converge at the matching output layer.
In this optional embodiment, a specific structure of the user semantic matching model is as shown in fig. 2b, and includes a user vector extraction branch located on the left side and a data vector extraction branch located on the right side, each branch includes an input layer, at least two levels of residual network blocks, and a vector convergence layer, and the two branches converge at a matching output layer. The system comprises an input layer, at least two stages of residual error network blocks, a vector convergence layer and a final matching output layer, wherein the input layer is used for converting input features into initial feature vectors, the at least two stages of residual error network blocks are used for converting the initial feature vectors into intermediate feature vectors corresponding to the residual error network blocks at all stages, the vector convergence layer is used for generating semantic representations corresponding to the input features according to the intermediate feature vectors at all stages, the final matching output layer is used for calculating matching scores of the input features of the two branches according to the semantic representations respectively output by the two branches, and the higher the matching score is, the higher the matching degree is.
In step S23, the features of the user and the features of the data to be recommended are converted into an initial user feature vector and an initial data feature vector, respectively, through the input layer included in each branch.
In the embodiment of the present disclosure, the features of the user are converted into the initial user feature vector through the input layer located on the left side in the user semantic matching model, specifically, each item in the features of the user is converted into at least one vector element, and finally, the elements jointly form the initial user feature vector, which is x in fig. 2 b; the features of the data to be recommended are converted into initial data feature vectors through an input layer on the right side in the user semantic matching model, specifically, each item in the features of the data to be recommended is converted into at least one vector element, and finally, the elements form the initial data feature vectors, namely, y in fig. 2 b.
In step S24, the initial user feature vector and the initial data feature vector are sequentially processed through at least two levels of residual network blocks included in each branch, respectively, to obtain an intermediate user feature vector and an intermediate data feature vector corresponding to each level of residual network block.
Aiming at the learning process of high-order semantic representation, in order to improve the quality of the learned semantic representation, an easily-conceived method is to increase the depth of a neural network, because each layer of the neural network corresponds to the extraction of feature information of different layers, when the network is deeper, the extracted information of different layers is more, the combination of the layer information among different layers is more, and the level of the feature is higher along with the deepening of the network depth, so the depth of the network is an important factor for realizing a good learning effect, but the network cannot be converged due to the fact that the network is simply increased in the number of the layers, and high-quality semantic representation cannot be obtained.
In the embodiment of the disclosure, each of the two branches of the user semantic matching model includes at least two stages of residual network blocks, and the initial user feature vector and the initial data feature vector input by the input layer can be processed in sequence to obtain an intermediate user feature vector and an intermediate data feature vector corresponding to each residual network block.
In an implementation manner of the embodiment of the present disclosure, optionally, the representation of the residual network block is as follows:
xl+1＝xl+F(xl,wl)，
where l is the hierarchy of residual network blocks, xlIs the output of the l-th stage residual network block, F (x)l,wl) Is formed by at least one layer of neural network as a residual function, wlIs a weight matrix.
In the above optional embodiment, a specific representation of the residual network blocks is provided, the output of each residual network block is formed by a feature vector and a residual function output by the last residual network block, the residual network block can improve the quality of semantic representation by increasing the depth of the network, and a jump connection mode is adopted to alleviate the problem of gradient disappearance caused by increasing the depth in the deep neural network.
In an implementation manner of the embodiment of the present disclosure, optionally, the activation function adopted by each residual network block is as follows:
f(x)＝x·sigmoid(x)。
each deep neural network relies on a linearly transformed activation function to make the whole network highly nonlinear, wherein the most commonly used activation function is a linear rectification function ReLU, which is almost a default option for deep learning due to its simplicity in definition, high implementation efficiency and excellent performance in training the deep neural network, specifically: ReLU ═ max (0, x).
Since ReLU is not derivable at point 0, the derivative is 0 when x < 0, and for inputs less than 0, no effective learning can be performed. In this alternative embodiment, a new activation function, f (x) ═ x sigmoid (x), is used, and as shown in fig. 2c, the excitation function makes x globally derivable in the definition domain, so that the user semantic matching model can effectively learn the high-order semantic representation.
In step S25, the intermediate user feature vectors and the intermediate data feature vectors corresponding to the residual network blocks at each level are processed through the vector convergence layer included in each branch, so as to obtain semantic representations corresponding to the features of the user and the features of the data to be recommended.
Wherein, the semantic representation refers to a feature vector capable of clearly representing the user or quasi-recommended data semantics, and the vector convergence layer of the left user vector extraction branch of the user semantic matching model shown in fig. 2b obtains the vector
That is, a vector obtained by a vector convergence layer in a right-side data vector extraction branch for semantic representation of a feature of a user
I.e. a semantic representation of the features for the data to be recommended.
In order to improve the learning quality of high-order semantic representation, a deeper neural network is generally adopted, and in order to further solve the problem of gradient disappearance caused by the deep neural network, a vector convergence layer in two branches is used for processing intermediate user characteristic vectors and intermediate data characteristic vectors output by multi-level residual error network blocks in each branch, so that semantic representation comprising the output characteristic vectors of the multi-level residual error network blocks is finally obtained, the influence of output information in a bottom-layer residual error network block on the final semantic representation is further deepened, and the multi-level semantic information can be captured. For example, the vector convergence layer may process the intermediate user feature vectors and the intermediate data feature vectors output by the residual network blocks at each level by parallel-connecting the intermediate feature vectors and multiplying the parallel-connecting matrix to obtain a final semantic representation, where each intermediate feature vector may be an intermediate feature vector output by a residual network block at a set level selected according to an actual situation.
In step S26, a matching result between the user and the data to be recommended is generated based on the semantic representation corresponding to the feature of the user and the semantic representation corresponding to the feature of the data to be recommended.
The technical scheme of the embodiment of the disclosure includes that firstly, the characteristics of a user and the characteristics of data to be recommended are respectively converted into initial user characteristic vectors and initial data characteristic vectors through each input layer in a user semantic matching model, then the initial user characteristic vectors and the initial data characteristic vectors are respectively processed in sequence through at least two stages of residual error network blocks contained in each branch to obtain intermediate user characteristic vectors and intermediate data characteristic vectors corresponding to each stage of residual error network blocks, finally, semantic representations corresponding to the characteristics of the user and the characteristics of the data to be recommended are determined according to the intermediate user characteristic vectors and the intermediate data characteristic vectors corresponding to each residual error network block through a vector convergence layer contained in each branch, and the problems that learning to semantic representation is not high when the network quality is shallow and the network cannot be converged due to the fact that the network is simply increased in the number of network layers in the prior art are solved, high-order semantic representations of users and data can be learned through a new network structure, and therefore the precision of corresponding semantic representations of the users and the data to be recommended is improved.
Fig. 3 is a flowchart illustrating an information matching method according to an exemplary embodiment, where this embodiment is a further refinement of the above technical solutions, and the technical solutions in this embodiment may be combined with various optional solutions in one or more embodiments described above. As shown in fig. 3, the information matching method includes the following steps:
in step S31, the characteristics of the user who is to acquire the recommendation data and the characteristics of the data to be recommended are acquired.
In step S32, the features of the user and the features of the data to be recommended are input to the user vector extraction branch and the data vector extraction branch of the user semantic matching model, respectively.
In step S33, the features of the user and the features of the data to be recommended are converted into an initial user feature vector and an initial data feature vector, respectively, through the input layer included in each branch.
In step S34, the initial user feature vector and the initial data feature vector are sequentially processed through at least two levels of residual network blocks included in each branch, respectively, to obtain an intermediate user feature vector and an intermediate data feature vector corresponding to each level of residual network block.
In step S35, semantic representations corresponding to the features of the user and the features of the data to be recommended are calculated from the intermediate feature vector output from the at least one set hierarchical residual network block and the intermediate feature vector output from the last-layer residual network block.
In the embodiment of the disclosure, in order to alleviate the problem of gradient disappearance caused by a deep neural network, the vector convergence layer obtains intermediate feature vectors output by a plurality of residual network blocks of a set hierarchy and intermediate feature vectors output by a last layer of residual network block, performs preset convergence processing on the obtained plurality of intermediate feature vectors to obtain final semantic representation, and improves the influence of output information of the bottom layer of residual network blocks on a final result.
In a specific implementation manner of the embodiment of the present disclosure, in the vector convergence layer located on the left side of the user semantic matching model, first, an intermediate user feature vector output by at least one residual network block close to the model input layer, for example, an intermediate user feature vector output by a level 1-3 residual network block, and an intermediate user feature vector output by a last layer of residual network block are obtained, then, the obtained plurality of intermediate user feature vectors are connected in parallel to obtain a feature vector including a plurality of intermediate user feature vectors, and the feature vector is processed, for example, multiplied by a parallel matrix, to finally obtain a semantic representation corresponding to the features of the user. And the vector convergence layer positioned on the right side of the user semantic matching model also obtains the intermediate data feature vector output by at least one stage of residual error network block close to the input layer of the model and the intermediate user feature vector output by the last layer of residual error network block in the same way, and processes the obtained plurality of intermediate data feature vectors to obtain semantic representation corresponding to the features of the data to be recommended.
In an implementation manner of the embodiment of the present disclosure, optionally, the calculating, according to the intermediate feature vector output by the at least one set level residual network block and the intermediate feature vector output by the last layer residual network block, a semantic representation corresponding to the features of the user and the features of the data to be recommended includes:
connecting the intermediate characteristic vector output by at least one residual error network block close to the input layer and the intermediate characteristic vector output by the last layer of residual error network block in parallel;
multiplying the parallel result by the parallel matrix, and adding the multiplied result and the offset vector to obtain semantic representations corresponding to the characteristics of the user and the characteristics of the to-be-recommended data, wherein the calculation mode is as follows:
xL＝wL[x1,x2,…,xL-1]+bL，
wherein x is1,x2… is the intermediate feature vector, x, of the residual network block output near the input layerL-1Is the intermediate feature vector, w, output by the last layer of residual network blockLIs a parallel matrix, bLIs a bias vector.
In the above optional embodiment, there is further provided a specific manner of determining semantic representations corresponding to the features of the user and the features of the data to be recommended according to the intermediate feature vector output by the at least one set level residual network block and the intermediate feature vector output by the last layer of residual network block, where the intermediate feature vector output by the at least one residual network block close to the input layer and the intermediate feature vector output by the last layer of residual network block are connected in parallel, for example, the intermediate feature vector output by the 1-3 layers of residual network blocks and the intermediate feature vector output by the last layer of residual network block are selectedThe intermediate eigenvectors corresponding to the difference network blocks are x respectively1＝[x11,x12,x13,x14]，x2＝[x21,x22,x23,x24]，x3＝[x31,x32,x33]，xL-1＝[x(L-1)1,x(L-1)2]Then, the intermediate eigenvectors are connected in parallel to obtain x ═ x11,x12,x13,x14,x21,x22,x23,x24,x31,x32,x33,x(L-1)1,x(L-1)2]Then, the eigenvectors obtained by parallel connection and the parallel matrix w are alignedLMultiplied by the offset vector bLAnd adding to obtain the final semantic representation. The residual network blocks of at least one set level are selected from the residual network blocks close to the input layer according to actual conditions, and the output of the selected residual network block can have larger influence on final semantic representation compared with the output of the unselected residual network blocks, so that the underlying network information can be better transmitted to a deep network.
In step S36, a matching result between the user and the data to be recommended is generated based on the semantic representation corresponding to the feature of the user and the semantic representation corresponding to the feature of the data to be recommended.
The technical scheme of the embodiment of the disclosure includes that firstly, the characteristics of a user and the characteristics of data to be recommended are respectively converted into initial user characteristic vectors and initial data characteristic vectors through each input layer in a user semantic matching model, then the initial user characteristic vectors and the initial data characteristic vectors are respectively processed in sequence through at least two residual error network blocks contained in each branch to obtain intermediate user characteristic vectors and intermediate data characteristic vectors corresponding to each residual error network block, finally, semantic representation corresponding to the characteristics of the user and the characteristics of data to be recommended is determined according to the intermediate characteristic vectors output by at least one residual error network block of a set level and the intermediate characteristic vectors output by the last layer of residual error network block, and the problems that the semantic representation is not high when the number of network layers is shallow in the prior art and the network cannot be converged due to the fact that the number of network layers is simply increased are solved, and the influence of the output information of the bottom layer residual error network block on the final result can be improved, so that the learned semantic representation precision is improved.
Fig. 4 is a flowchart illustrating an information matching method according to an exemplary embodiment, which is an application scenario for the foregoing technical solution. As shown in fig. 4, the information matching method includes the following steps:
in step S41, the features of the user and the features of the pseudo-recommendation data are input to the user semantic matching model.
In the embodiment of the disclosure, the features of a user and the features of data to be recommended are input into a user semantic matching model, the input layers of two branches in the user semantic matching model respectively convert the features of the user and the features of the data to be recommended into an initial user feature vector and an initial data feature vector, then at least two stages of residual error network blocks contained in each branch sequentially process the initial user feature vector and the initial data feature vector, and a vector convergence layer converges the processing results of the residual error network blocks to obtain semantic representations corresponding to the features of the user and the features of the data to be recommended.
In an implementation manner of the embodiment of the present disclosure, optionally, the features of the user include user attribute information and user browsing information, and the features of the data to be recommended include data attribute information.
In the above optional embodiments, specific contents of the features of the user and the features of the recommendation data are provided, where the features of the user include user attribute information and user browsing information, and the user attribute information is information related to the user, for example, information that can represent the basic situation of the user, such as the age, sex, and location of the user; the user browsing information can be specific information such as a webpage browsed by the user, a purchased or browsed commodity, a watched short video and the like; the characteristics of the data to be recommended include data attribute information, such as keywords of the data to be recommended, a publisher, a type of the data to be recommended, and the like.
In step S42, semantic matching is performed on the features of the user and the features of the to-be-recommended data by the user semantic matching model to obtain a matching score between the user and the to-be-recommended data;
the higher the matching score, the higher the degree of matching between the user and the data to be recommended.
In the embodiment of the present disclosure, after the each vector convergence layer of the user semantic matching model respectively outputs semantic representations corresponding to the features of the user and the features of the data to be recommended, the matching output layer matches the semantic representations of the user and the semantic representations of the data in a set manner to obtain a final matching score, which is h in fig. 2bθ(x,y)＝<u(θ,x),v(θ,y)>. For example, the matching output layer may calculate an inner product of the semantic representation of the user and the semantic representation of the data, and use the calculation result as a matching score, where the higher the matching score is, the higher the matching degree between the features of the user and the features of the data to be recommended is, that is, data recommendation may be performed on the user according to the matching score output by the user semantic matching model.
According to the technical scheme, the user characteristics and the quasi-recommendation data characteristics are input into the user semantic matching model, semantic matching is conducted on the user characteristics and the quasi-recommendation data characteristics through the user semantic matching model, matching scores of the user characteristics and the quasi-recommendation data characteristics are obtained, the learning quality of high-order semantic representation is improved through the user semantic matching model, the matching degree of the user and the data is calculated according to semantic representation learning results, and the matching degree of the data recommended for the user and the user is improved.
Fig. 5 is a block diagram illustrating an information matching apparatus according to an example embodiment. Referring to fig. 5, the apparatus includes a feature acquisition module 510, a feature input module 520, and a matching result generation module 530.
The feature obtaining module 510 is configured to obtain features of a user who wants to obtain recommended data and features of data to be recommended;
a feature input module 520, configured to input the features of the user and the features of the data to be recommended to a user vector extraction branch and a data vector extraction branch of a user semantic matching model, respectively, where the user vector extraction branch and the data vector extraction branch are respectively configured to convert the input features into initial feature vectors, and generate intermediate feature vectors of each stage based on the initial feature vectors through at least two stages of residual network blocks that are set respectively, and the intermediate feature vectors of each stage are used to generate semantic representations of corresponding features;
a matching result generating module 530, configured to generate a matching result between the user and the data to be recommended based on the semantic representation corresponding to the feature of the user and the semantic representation corresponding to the feature of the data to be recommended.
In an implementation manner of the embodiment of the present disclosure, optionally, the user semantic matching model includes a user vector extraction branch and a data vector extraction branch, where the user vector extraction branch and the data vector extraction branch each include an input layer, at least two levels of residual error network blocks, and a vector convergence layer, and the two branches converge on a matching output layer;
the information matching device further comprises:
the initial feature vector acquisition module is used for respectively inputting the features of the user and the features of the to-be-recommended data into a user vector extraction branch and a data vector extraction branch of a user semantic matching model, and respectively converting the features of the user and the features of the to-be-recommended data into an initial user feature vector and an initial data feature vector through an input layer contained in each branch;
the intermediate characteristic vector acquisition module is used for respectively and sequentially processing the initial user characteristic vector and the initial data characteristic vector through at least two stages of residual error network blocks contained in each branch to obtain intermediate user characteristic vectors and intermediate data characteristic vectors corresponding to the residual error network blocks at each stage;
and the semantic representation acquisition module is used for respectively processing the intermediate user characteristic vectors and the intermediate data characteristic vectors corresponding to the residual network blocks at each level through the vector convergence layer contained in each branch to obtain semantic representations corresponding to the characteristics of the user and the characteristics of the data to be recommended.
In an implementation manner of the embodiment of the present disclosure, optionally, the semantic representation obtaining module includes:
and the semantic representation acquisition unit is used for calculating semantic representations corresponding to the characteristics of the user and the characteristics of the data to be recommended according to the intermediate characteristic vector output by the residual error network block of at least one set level and the intermediate characteristic vector output by the residual error network block of the last layer.
In an implementation manner of the embodiment of the present disclosure, optionally, the residual network block is represented as follows:
xl+1＝xl+F(xl,wl)，
where l is the hierarchy of residual network blocks, xlIs the output of the l-th stage residual network block, F (x)l,wl) Is formed by at least one layer of neural network as a residual function, wlIs a weight matrix.
In an implementation manner of the embodiment of the present disclosure, optionally, the activation function adopted by each residual network block is as follows:
f(x)＝x·sigmoid(x)。
in an implementation manner of the embodiment of the present disclosure, optionally, the semantic representation acquiring unit is specifically configured to:
connecting the intermediate characteristic vector output by at least one residual error network block close to the input layer and the intermediate characteristic vector output by the last layer of residual error network block in parallel;
multiplying the parallel result by the parallel matrix, and adding the multiplied result and the offset vector to obtain semantic representations corresponding to the characteristics of the user and the characteristics of the to-be-recommended data, wherein the calculation mode is as follows:
xL＝wL[x1,x2,…,xL-1]+bL，
wherein x is1,x2… is the intermediate feature vector, x, of the residual network block output near the input layerL-1Is the intermediate feature vector, w, output by the last layer of residual network blockLIs a parallel matrix, bLIs a bias vector.
In an implementation manner of the embodiment of the present disclosure, optionally, the information matching apparatus is further configured to:
inputting the characteristics of the user and the characteristics of the data to be recommended into a user semantic matching model;
semantic matching is carried out on the characteristics of the user and the characteristics of the to-be-recommended data through the user semantic matching model, and matching scores of the user and the to-be-recommended data are obtained;
the higher the matching score is, the higher the matching degree of the user and the data to be recommended is;
the characteristics of the user comprise user attribute information and user browsing information, and the characteristics of the data to be recommended comprise data attribute information.
With regard to the information matching apparatus in the above-described embodiment, the specific manner in which each unit performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.
Fig. 6 is a schematic structural diagram illustrating an electronic device according to an exemplary embodiment, where the electronic device includes, as shown in fig. 6:
one or more of the processors 610 may be capable of,
in FIG. 6, a processor 610 is illustrated;
a memory 620;
the processor 610 and the memory 620 in the device may be connected by a bus or other means, and fig. 6 illustrates the connection by a bus as an example.
The memory 620, as a non-transitory computer-readable storage medium, may be used for storing software programs, computer-executable programs, and modules, such as program instructions/modules corresponding to an information matching method in the embodiments of the present disclosure (for example, the feature acquisition module 510, the feature input module 520, and the matching result generation module 530 shown in fig. 5). The processor 610 executes various functional applications and data processing of the computer device by running software programs, instructions and modules stored in the memory 620, namely, implements an information matching method of the above method embodiment, namely:
acquiring the characteristics of a user to acquire recommended data and the characteristics of data to be recommended;
inputting the characteristics of the user and the characteristics of the data to be recommended into a user vector extraction branch and a data vector extraction branch of a user semantic matching model respectively, wherein the user vector extraction branch and the data vector extraction branch are used for converting the input characteristics into initial characteristic vectors respectively, generating intermediate characteristic vectors of each stage based on the initial characteristic vectors through at least two stages of residual error network blocks which are arranged respectively, and the intermediate characteristic vectors of each stage are used for generating semantic representation of corresponding characteristics;
and generating a matching result of the user and the to-be-recommended data based on the semantic representation corresponding to the characteristics of the user and the semantic representation corresponding to the characteristics of the to-be-recommended data.
The memory 620 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to use of the computer device, and the like. Further, the memory 620 may include high speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory 620 optionally includes memory located remotely from the processor 610, which may be connected to the terminal device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
In an exemplary embodiment, a storage medium comprising instructions, such as the memory 620 comprising instructions, executable by the processor 610 of the electronic device to perform the above-described method is also provided. Alternatively, the storage medium may be a non-transitory computer readable storage medium, such as a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.
In an exemplary embodiment, a computer program product for use in conjunction with an electronic device is also provided, the computer program product comprising a computer readable storage medium and a computer program mechanism embedded therein, the program being loaded into and executed by a computer to implement the information matching method according to any of the embodiments of the present disclosure.
Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims (10)
1. An information matching method, characterized in that the method comprises:
acquiring the characteristics of a user to acquire recommended data and the characteristics of data to be recommended;
inputting the characteristics of the user and the characteristics of the data to be recommended into a user vector extraction branch and a data vector extraction branch of a user semantic matching model respectively, wherein the user vector extraction branch and the data vector extraction branch are used for converting the input characteristics into initial characteristic vectors respectively, generating intermediate characteristic vectors of each stage based on the initial characteristic vectors through at least two stages of residual error network blocks which are arranged respectively, and the intermediate characteristic vectors of each stage are used for generating semantic representation of corresponding characteristics;
and generating a matching result of the user and the to-be-recommended data based on the semantic representation corresponding to the characteristics of the user and the semantic representation corresponding to the characteristics of the to-be-recommended data.
2. The information matching method according to claim 1, wherein the user semantic matching model comprises a user vector extraction branch and a data vector extraction branch, the user vector extraction branch and the data vector extraction branch each comprise an input layer, at least two stages of residual network blocks, and a vector convergence layer, and the two branches converge at a matching output layer;
after the features of the user and the features of the data to be recommended are respectively input to a user vector extraction branch and a data vector extraction branch of a user semantic matching model, the method further comprises the following steps:
converting the characteristics of the user and the characteristics of the data to be recommended into initial user characteristic vectors and initial data characteristic vectors respectively through input layers contained in the branches;
respectively and sequentially processing the initial user characteristic vector and the initial data characteristic vector through at least two stages of residual error network blocks contained in each branch to obtain a middle user characteristic vector and a middle data characteristic vector corresponding to each stage of residual error network block;
and respectively processing the intermediate user characteristic vectors and the intermediate data characteristic vectors corresponding to the residual error network blocks at each level through a vector convergence layer contained in each branch to obtain semantic representations corresponding to the characteristics of the user and the characteristics of the data to be recommended.
3. The information matching method according to claim 2, wherein the semantic representation step of obtaining the features of the user and the features of the data to be recommended by processing the intermediate user feature vectors and the intermediate data feature vectors corresponding to the residual network blocks of each stage through the vector convergence layer included in each branch comprises:
and calculating semantic representations corresponding to the characteristics of the user and the characteristics of the data to be recommended according to the intermediate characteristic vector output by the residual network block of at least one set level and the intermediate characteristic vector output by the residual network block of the last layer.
4. The information matching method according to claim 1, wherein the residual network block is represented as follows:
xl+1＝xl+F(xl,wl)，
where l is the hierarchy of residual network blocks, xlIs the output of the l-th stage residual network block, F (x)l,wl) Is formed by at least one layer of neural network as a residual function, wlIs a weight matrix.
5. The information matching method according to claim 4, wherein the activation function used by each residual network block is as follows:
f(x)＝x·sigmoid(x)。
6. the information matching method according to claim 3, wherein the step of calculating semantic representations corresponding to the features of the user and the features of the data to be recommended according to the intermediate feature vector output by the at least one set level of residual network blocks and the intermediate feature vector output by the last layer of residual network blocks comprises:
connecting the intermediate characteristic vector output by at least one residual error network block close to the input layer and the intermediate characteristic vector output by the last layer of residual error network block in parallel;
multiplying the parallel result by the parallel matrix, and adding the multiplied result and the offset vector to obtain semantic representations corresponding to the characteristics of the user and the characteristics of the to-be-recommended data, wherein the calculation mode is as follows:
xL＝wL[x1,x2,…,xL-1]+bL，
wherein x is1,x2… is the intermediate feature vector, x, of the residual network block output near the input layerL-1Is the intermediate eigenvector, w, output by the last layer of residual network blockLIs a parallel matrix, bLIs a bias vector.
7. The information matching method according to claim 1, further comprising:
inputting the characteristics of the user and the characteristics of the data to be recommended into a user semantic matching model;
semantic matching is carried out on the characteristics of the user and the characteristics of the to-be-recommended data through the user semantic matching model, and matching scores of the user and the to-be-recommended data are obtained;
the higher the matching score is, the higher the matching degree of the user and the data to be recommended is;
the characteristics of the user comprise user attribute information and user browsing information, and the characteristics of the data to be recommended comprise data attribute information.
8. An information matching apparatus, characterized in that the information matching apparatus comprises:
the characteristic acquisition module is used for acquiring the characteristics of a user to acquire recommended data and the characteristics of the data to be recommended;
the characteristic input module is used for respectively inputting the characteristics of the user and the characteristics of the data to be recommended to a user vector extraction branch and a data vector extraction branch of a user semantic matching model, wherein the user vector extraction branch and the data vector extraction branch are respectively used for converting the input characteristics into initial characteristic vectors, intermediate characteristic vectors of each stage are generated on the basis of the initial characteristic vectors through at least two stages of residual error network blocks which are respectively arranged, and the intermediate characteristic vectors of each stage are used for generating semantic representation of corresponding characteristics;
and the matching result generation module is used for generating a matching result of the user and the to-be-recommended data based on the semantic representation corresponding to the characteristics of the user and the semantic representation corresponding to the characteristics of the to-be-recommended data.
9. An electronic device, comprising:
a processor;
a memory for storing the processor-executable commands;
wherein the processor is configured to execute the command to implement the information matching method of any one of claims 1 to 7.
10. A storage medium in which commands, when executed by a processor of a server, enable the server to perform the information matching method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010761650.3A CN114065014A (en) | 2020-07-31 | 2020-07-31 | Information matching method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010761650.3A CN114065014A (en) | 2020-07-31 | 2020-07-31 | Information matching method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114065014A true CN114065014A (en) | 2022-02-18 |
Family
ID=80227763
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010761650.3A Pending CN114065014A (en) | 2020-07-31 | 2020-07-31 | Information matching method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114065014A (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108537373A (en) * | 2018-03-27 | 2018-09-14 | 黄晓鸣 | Travel information recommends method and apparatus |
| CN108694225A (en) * | 2017-03-31 | 2018-10-23 | 阿里巴巴集团控股有限公司 | A kind of image search method, the generation method of feature vector, device and electronic equipment |
| CN109710744A (en) * | 2018-12-28 | 2019-05-03 | 合肥讯飞数码科技有限公司 | A kind of data matching method, device, equipment and storage medium |
| US20190156210A1 (en) * | 2017-11-17 | 2019-05-23 | Facebook, Inc. | Machine-Learning Models Based on Non-local Neural Networks |
| CN110232152A (en) * | 2019-05-27 | 2019-09-13 | 腾讯科技（深圳）有限公司 | Content recommendation method, device, server and storage medium |
| CN110234018A (en) * | 2019-07-09 | 2019-09-13 | 腾讯科技（深圳）有限公司 | Multimedia content description generation method, training method, device, equipment and medium |
| CN110598109A (en) * | 2019-09-16 | 2019-12-20 | 上海喜马拉雅科技有限公司 | Information recommendation method, device, equipment and storage medium |
| CN110781776A (en) * | 2019-10-10 | 2020-02-11 | 湖北工业大学 | Road extraction method based on prediction and residual refinement network |
| CN110782462A (en) * | 2019-10-30 | 2020-02-11 | 浙江科技学院 | Semantic segmentation method based on double-flow feature fusion |
| CN111259197A (en) * | 2020-01-13 | 2020-06-09 | 清华大学 | Video description generation method based on pre-coding semantic features |
-
2020
- 2020-07-31 CN CN202010761650.3A patent/CN114065014A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108694225A (en) * | 2017-03-31 | 2018-10-23 | 阿里巴巴集团控股有限公司 | A kind of image search method, the generation method of feature vector, device and electronic equipment |
| US20190156210A1 (en) * | 2017-11-17 | 2019-05-23 | Facebook, Inc. | Machine-Learning Models Based on Non-local Neural Networks |
| CN108537373A (en) * | 2018-03-27 | 2018-09-14 | 黄晓鸣 | Travel information recommends method and apparatus |
| CN109710744A (en) * | 2018-12-28 | 2019-05-03 | 合肥讯飞数码科技有限公司 | A kind of data matching method, device, equipment and storage medium |
| CN110232152A (en) * | 2019-05-27 | 2019-09-13 | 腾讯科技（深圳）有限公司 | Content recommendation method, device, server and storage medium |
| CN110234018A (en) * | 2019-07-09 | 2019-09-13 | 腾讯科技（深圳）有限公司 | Multimedia content description generation method, training method, device, equipment and medium |
| CN110598109A (en) * | 2019-09-16 | 2019-12-20 | 上海喜马拉雅科技有限公司 | Information recommendation method, device, equipment and storage medium |
| CN110781776A (en) * | 2019-10-10 | 2020-02-11 | 湖北工业大学 | Road extraction method based on prediction and residual refinement network |
| CN110782462A (en) * | 2019-10-30 | 2020-02-11 | 浙江科技学院 | Semantic segmentation method based on double-flow feature fusion |
| CN111259197A (en) * | 2020-01-13 | 2020-06-09 | 清华大学 | Video description generation method based on pre-coding semantic features |
Non-Patent Citations (2)
| Title |
|---|
| BRIJRAJ SINGH ET. AL: "Shunt connection: An intelligent skipping of contiguous blocks for optimizing MobileNet-V2", 《NEURAL NETWORKS》, 31 October 2019 (2019-10-31), pages 192 - 203 * |
| 操江峰: "基于深度学习的图像与视频风格化研究与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》, 15 October 2017 (2017-10-15), pages 138 - 232 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10713317B2 (en) | Conversational agent for search | |
| US11790904B2 (en) | Voice application platform | |
| Hu et al. | Leveraging meta-path based context for top-n recommendation with a neural co-attention model | |
| US10621185B2 (en) | Method and apparatus for recalling search result based on neural network | |
| Kejriwal | Knowledge Graphs: Constructing, Completing, and Effectively Applying Knowledge Graphs in Tourism | |
| CN108628926A (en) | Topic association and marking for fine and close image | |
| EP3707622A1 (en) | Generation of text from structured data | |
| CN113762322A (en) | Video classification method, device and equipment based on multi-modal representation and storage medium | |
| WO2023065859A1 (en) | Item recommendation method and apparatus, and storage medium | |
| CN109902087B (en) | Data processing method and device for questions and answers and server | |
| CN110162766B (en) | Word vector updating method and device | |
| CN111309878B (en) | Search type question-answering method, model training method, server and storage medium | |
| CN108960574A (en) | Quality determination method, device, server and the storage medium of question and answer | |
| Wang et al. | Trust-aware collaborative filtering with a denoising autoencoder | |
| KR20210106398A (en) | Conversation-based recommending method, conversation-based recommending apparatus, and device | |
| Suzuki et al. | Stacked denoising autoencoder-based deep collaborative filtering using the change of similarity | |
| CN111709223B (en) | Sentence vector generation method and device based on bert and electronic equipment | |
| CN115564017A (en) | Model data processing method, electronic device and computer storage medium | |
| CN116431914A (en) | Cross-domain recommendation method and system based on personalized preference transfer model | |
| CN109992659B (en) | Method and device for text sorting | |
| CN117216219A (en) | Thinking chain reasoning method, device, equipment and storage medium | |
| Aggarwal et al. | Improving search through A3C reinforcement learning based conversational agent | |
| KR20210147771A (en) | Method and apparatus for increasing knowledge based on uncertainty in neural networks | |
| Qin et al. | OTDE: optimal transport distribution enhancement for few-shot video recognition | |
| CN114065014A (en) | Information matching method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |