CN113379500A - Sequencing model training method and device, and article sequencing method and device - Google Patents
Sequencing model training method and device, and article sequencing method and device Download PDFInfo
- Publication number
- CN113379500A CN113379500A CN202110684099.1A CN202110684099A CN113379500A CN 113379500 A CN113379500 A CN 113379500A CN 202110684099 A CN202110684099 A CN 202110684099A CN 113379500 A CN113379500 A CN 113379500A
- Authority
- CN
- China
- Prior art keywords
- candidate
- sample
- feature vector
- distribution table
- output matrix
- 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 42
- 238000012549 training Methods 0.000 title claims abstract description 38
- 238000012163 sequencing technique Methods 0.000 title claims abstract description 22
- 239000013598 vector Substances 0.000 claims abstract description 126
- 239000011159 matrix material Substances 0.000 claims abstract description 61
- 238000000605 extraction Methods 0.000 claims abstract description 18
- 230000006870 function Effects 0.000 claims abstract description 17
- 238000010801 machine learning Methods 0.000 claims abstract description 14
- 238000012545 processing Methods 0.000 claims description 46
- 238000011176 pooling Methods 0.000 claims description 32
- 230000004927 fusion Effects 0.000 claims description 4
- 238000010586 diagram Methods 0.000 description 10
- 238000003062 neural network model Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
Images
Classifications
-
- 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
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Software Systems (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Accounting & Taxation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Finance (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Algebra (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Apparatus Associated With Microorganisms And Enzymes (AREA)
- Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
Abstract
The disclosure provides a sequencing model training method and device and an article sequencing method and device. The training method of the ranking model comprises the following steps: inputting the related information of all sample articles in the sample set into a feature extraction model to respectively obtain a first feature vector of each sample article; fusing the first feature vectors of all sample articles to obtain fused feature vectors; fusing the fused feature vector with the first feature vector of each sample article respectively to obtain a second feature vector of each sample article; inputting the second feature vectors of all sample articles into a machine learning model to obtain a two-dimensional probability distribution table; obtaining an output matrix according to the two-dimensional probability distribution table; calculating a cross entropy loss function of each row and each column of the output matrix by using the label matrix of the sample set; and training the machine learning model by using the cross entropy loss function of each row and each column of the output matrix to obtain a sequencing model.
Description
Technical Field
The present disclosure relates to the field of information processing, and in particular, to a ranking model training method and apparatus, and an article ranking method and apparatus.
Background
In order to guarantee the experience of the online user, the network platform recommends a plurality of candidate items for the user after receiving the user request. Since different arrangement orders of the plurality of candidate items may affect the click rate of the user, the arrangement order of the plurality of candidate items needs to be optimized.
Disclosure of Invention
The inventor finds that, through research, due to the fact that the arrangement modes of a plurality of candidate articles are changeable, all sequence arrangement modes cannot be traversed and an optimal sequence cannot be selected within limited system response time, and therefore the experience of online users is affected.
Accordingly, the present disclosure provides an article sorting scheme, which can conveniently and quickly select an optimal sorting order, thereby improving the experience of the online user.
According to a first aspect of the embodiments of the present disclosure, there is provided a ranking model training method, including: inputting the related information of all sample articles in the sample set into a feature extraction model to respectively obtain a first feature vector of each sample article; fusing the first feature vectors of all the sample articles to obtain fused feature vectors; fusing the fused feature vector with the first feature vector of each sample article respectively to obtain a second feature vector of each sample article; inputting the second feature vectors of all the sample articles into a machine learning model to obtain a two-dimensional probability distribution table, wherein the ith column of the two-dimensional probability distribution table comprises the probability of the ith sample article at each position in a recommended sequence, the jth row comprises the probability of each sample article at the jth position in the recommended sequence, i is more than or equal to 1, j is more than or equal to N, and N is the total number of the sample articles in the sample set; obtaining an output matrix according to the two-dimensional probability distribution table; calculating a cross entropy loss function for each row and each column of the output matrix using a label matrix of the sample set; and training the machine learning model by using the cross entropy loss function of each row and each column of the output matrix to obtain a sequencing model.
In some embodiments, deriving an output matrix from the two-dimensional probability distribution table comprises: in each column of the two-dimensional probability distribution table, the maximum probability value is set to 1, and probability values other than the maximum probability value are set to 0 to obtain the output matrix.
In some embodiments, fusing the first feature vectors of all sample items comprises: and pooling the first feature vectors of all the sample articles to obtain a fusion feature vector.
In some embodiments, the Pooling treatment comprises a maximum pooled Max Pooling treatment or an Average pooled Average Pooling treatment.
According to a second aspect of the embodiments of the present disclosure, there is provided a ranking model training apparatus including: the first processing module is configured to input relevant information of all sample articles in the sample set into the feature extraction model so as to respectively obtain a first feature vector of each sample article; a second processing module configured to fuse the first feature vectors of all the sample items to obtain a fused feature vector; the third processing module is configured to fuse the fused feature vector with the first feature vector of each sample article respectively to obtain a second feature vector of each sample article; a fourth processing module configured to input the second feature vectors of all the sample items into a machine learning model to obtain a two-dimensional probability distribution table, wherein an ith column of the two-dimensional probability distribution table includes probabilities of the ith sample item at each position in a recommended sequence, a jth row includes probabilities of each sample item at a jth position in the recommended sequence, i is greater than or equal to 1, j is greater than or equal to N, and N is a total number of sample items in the sample set; obtaining an output matrix according to the two-dimensional probability distribution table; calculating a cross entropy loss function for each row and each column of the output matrix using a label matrix of the sample set; and training the machine learning model by using the cross entropy loss function of each row and each column of the output matrix to obtain a sequencing model.
According to a third aspect of the embodiments of the present disclosure, there is provided a ranking model training apparatus including: a memory configured to store instructions; a processor coupled to the memory, the processor configured to perform a method of implementing the ranking model training method according to any of the embodiments described above based on instructions stored in the memory.
According to a fourth aspect of the embodiments of the present disclosure, there is provided an article sorting method, including: inputting relevant information of all candidate articles in the candidate set into a feature extraction model to respectively obtain a first feature vector of each candidate article; fusing the first feature vectors of all the candidate articles to obtain fused feature vectors; fusing the fused feature vector with the first feature vector of each candidate article respectively to obtain a second feature vector of each candidate article; inputting the second feature vectors of all the candidate items into a ranking model to obtain a two-dimensional probability distribution table, wherein the ranking model is obtained by training with the training method described in any one of the embodiments, the ith column of the two-dimensional probability distribution table includes the probability of the ith candidate item at each position in a recommended sequence, the jth row includes the probability of each candidate item at the jth position in the recommended sequence, i is greater than or equal to 1, j is greater than or equal to M, and M is the total number of candidate items in the candidate set; obtaining an output matrix according to the two-dimensional probability distribution table; determining a position of each candidate item in the candidate set in the recommended sequence according to the output matrix so as to rank all candidate items in the candidate set.
In some embodiments, deriving an output matrix from the two-dimensional probability distribution table comprises: in each column of the two-dimensional probability distribution table, the maximum probability value is set to 1, and probability values other than the maximum probability value are set to 0 to obtain the output matrix.
In some embodiments, determining the position of each candidate item in the candidate set in the recommended sequence according to the output matrix comprises: and if the value of the ith candidate item in the candidate set in the jth row in the output matrix is 1, determining that the ith candidate item is located at the jth position in the recommended sequence.
In some embodiments, fusing the first feature vectors of all candidate items comprises: and pooling the first feature vectors of all the candidate articles to obtain a fusion feature vector.
In some embodiments, the Pooling treatment comprises a maximum pooled Max Pooling treatment or an Average pooled Average Pooling treatment.
According to a fifth aspect of embodiments of the present disclosure, there is provided an article sorting apparatus comprising: the fifth processing module is configured to input the relevant information of all candidate items in the candidate set into the feature extraction model to respectively obtain a first feature vector of each candidate item; a sixth processing module configured to fuse the first feature vectors of all the candidate items to obtain a fused feature vector; a seventh processing module, configured to fuse the fused feature vector with the first feature vector of each candidate item, respectively, to obtain a second feature vector of each candidate item; an eighth processing module, configured to input the second feature vectors of all the candidate items into a ranking model to obtain a two-dimensional probability distribution table, where the ranking model is obtained by training using the training method according to any of the above embodiments, an ith column of the two-dimensional probability distribution table includes probabilities of an ith candidate item at each position in a recommended sequence, a jth row includes probabilities of each candidate item at a jth position in the recommended sequence, i is greater than or equal to 1, j is less than or equal to M, and M is a total number of candidate items in the candidate set; obtaining an output matrix according to the two-dimensional probability distribution table; determining a position of each candidate item in the candidate set in the recommended sequence according to the output matrix so as to rank all candidate items in the candidate set.
According to a sixth aspect of embodiments of the present disclosure, there is provided an article sequencing apparatus comprising: a memory configured to store instructions; a processor coupled to the memory, the processor configured to perform a method of sequencing items implementing any of the embodiments described above based on instructions stored by the memory.
According to a seventh aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, in which computer instructions are stored, and when executed by a processor, the computer-readable storage medium implements the method according to any of the embodiments described above.
Other features of the present disclosure and advantages thereof will become apparent from the following detailed description of exemplary embodiments thereof, which proceeds with reference to the accompanying drawings.
Drawings
In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, it is obvious that the drawings in the following description are only some embodiments of the present disclosure, and for those skilled in the art, other drawings can be obtained according to the drawings without inventive exercise.
FIG. 1 is a schematic flow chart diagram illustrating a ranking model training method according to an embodiment of the present disclosure;
FIG. 2 is a schematic structural diagram of a ranking model training apparatus according to an embodiment of the present disclosure;
FIG. 3 is a schematic structural diagram of a training apparatus for ranking models according to another embodiment of the present disclosure;
FIG. 4 is a schematic flow chart diagram of an item sorting method according to an embodiment of the present disclosure;
FIG. 5 is a schematic diagram of an article sequencing device according to an embodiment of the present disclosure;
fig. 6 is a schematic structural diagram of an article sequencing device according to another embodiment of the present disclosure.
Fig. 7 is a flow chart illustrating an article sorting method according to yet another embodiment of the present disclosure.
Detailed Description
The technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the drawings in the embodiments of the present disclosure, and it is obvious that the described embodiments are only a part of the embodiments of the present disclosure, and not all of the embodiments. The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application, or uses. All other embodiments, which can be derived by a person skilled in the art from the embodiments disclosed herein without making any creative effort, shall fall within the protection scope of the present disclosure.
The relative arrangement of the components and steps, the numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.
Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description.
Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate.
In all examples shown and discussed herein, any particular value should be construed as merely illustrative, and not limiting. Thus, other examples of the exemplary embodiments may have different values.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.
Fig. 1 is a schematic flow chart of a ranking model training method according to an embodiment of the present disclosure. In some embodiments, the following ranking model training method is performed by a ranking model training apparatus.
In step 101, the related information of all sample items in the sample set is input into the feature extraction model to obtain a first feature vector of each sample item respectively.
In some embodiments, the feature extraction model is a neural network model for extracting the first feature vector of each sample item separately.
In step 102, the first feature vectors of all sample items are fused to obtain a fused feature vector.
In some embodiments, the first feature vectors of all sample items are pooled to obtain a fused feature vector. For example, Pooling includes maximum Pooling Max Pooling processing or Average Pooling processing.
In step 103, the fused feature vector is respectively fused with the first feature vector of each sample item to obtain a second feature vector of each sample item.
It should be noted that, since the fused feature vector is associated with the sample items in the entire sample set, the fused feature vector is fused with the first feature of each sample item, so as to help determine the position of each sample item in the entire sample set.
In step 104, the second feature vectors of all sample items are input into the machine learning model to obtain a two-dimensional probability distribution table. The ith column of the two-dimensional probability distribution table comprises the probability of the ith sample item at each position in the recommended sequence, the jth row comprises the probability of each sample item at the jth position in the recommended sequence, i is more than or equal to 1, j is more than or equal to N, and N is the total number of sample items in the sample set.
For example, the sample set includes 4 sample items, and the resulting two-dimensional probability distribution table is shown in table 1.
1 | 2 | 3 | 4 | |
1 | 0.9 | 0.1 | 0 | 0 |
2 | 0.05 | 0.8 | 0.15 | 0 |
3 | 0.05 | 0.1 | 0.7 | 0.15 |
4 | 0 | 0 | 0.15 | 0.85 |
TABLE 1
In step 105, an output matrix is obtained from the two-dimensional probability distribution table.
In some embodiments, in each column of the two-dimensional probability distribution table, the maximum probability value is set to 1, and probability values other than the maximum probability value are set to 0 to obtain an output matrix.
For example, the resulting output matrix is shown in table 2, according to table 1 above.
TABLE 2
At step 106, a cross entropy loss function is computed for each row and each column of the output matrix using the label matrix of the sample set.
For example, the cross entropy loss function is a softmax cross entropy loss function.
In step 107, the machine learning model is trained using the cross entropy loss function for each row and each column of the output matrix to obtain a ranking model.
Fig. 2 is a schematic structural diagram of a ranking model training apparatus according to an embodiment of the present disclosure. As shown in fig. 2, the ranking model training apparatus includes a first processing module 21, a second processing module 22, a third processing module 23, and a fourth processing module.
The first processing module 21 is configured to input the related information of all sample items in the sample set into the feature extraction model to obtain a first feature vector of each sample item respectively.
In some embodiments, the feature extraction model is a neural network model for extracting the first feature vector of each sample item separately.
The second processing module 22 is configured to fuse the first feature vectors of all sample items to obtain a fused feature vector.
In some embodiments, the second processing module 22 pools the first feature vectors of all sample items to obtain a fused feature vector. For example, Pooling includes maximum Pooling Max Pooling processing or Average Pooling processing.
The third processing module 23 is configured to fuse the fused feature vector with the first feature vector of each sample item, respectively, to obtain a second feature vector of each sample item.
The fourth processing module 24 is configured to input the second feature vectors of all sample items into the machine learning model to obtain a two-dimensional probability distribution table. The ith column of the two-dimensional probability distribution table comprises the probability of the ith sample article at each position in the recommended sequence, the jth row comprises the probability of each sample article at the jth position in the recommended sequence, i is more than or equal to 1, j is more than or equal to N, N is the total number of sample articles in the sample set, and an output matrix is obtained according to the two-dimensional probability distribution table.
For example, in each column of the two-dimensional probability distribution table, the maximum probability value is set to 1, and probability values other than the maximum probability value are set to 0 to obtain an output matrix.
The fourth processing module 24 calculates a cross entropy loss function of each row and each column of the output matrix by using the label matrix of the sample set, and trains the machine learning model by using the cross entropy loss function of each row and each column of the output matrix to obtain a ranking model.
Fig. 3 is a schematic structural diagram of a ranking model training apparatus according to another embodiment of the present disclosure. As shown in fig. 3, the ranking model training apparatus includes a memory 31 and a processor 32.
The memory 31 is used for storing instructions, the processor 32 is coupled to the memory 31, and the processor 32 is configured to execute the method according to any embodiment in fig. 1 based on the instructions stored in the memory.
As shown in fig. 3, the training apparatus for ranking model further includes a communication interface 33 for information interaction with other devices. Meanwhile, the training device of the ranking model further comprises a bus 34, and the processor 32, the communication interface 33 and the memory 31 are communicated with each other through the bus 34.
The memory 31 may comprise a high-speed RAM memory, and may also include a non-volatile memory (e.g., at least one disk memory). The memory 31 may also be a memory array. The storage 31 may also be partitioned and the blocks may be combined into virtual volumes according to certain rules.
Further, the processor 32 may be a central processing unit CPU, or may be an application specific integrated circuit ASIC, or one or more integrated circuits configured to implement embodiments of the present disclosure.
The present disclosure also relates to a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the instructions, when executed by a processor, implement the method according to any one of the embodiments in fig. 1.
Fig. 4 is a flow chart of an item sorting method according to an embodiment of the present disclosure. In some embodiments, the following method of sorting items is performed by an item sorting apparatus.
In step 401, the related information of all candidate items in the candidate set is input into the feature extraction model to obtain a first feature vector of each candidate item respectively.
In some embodiments, the feature extraction model is a neural network model for extracting the first feature vector of each candidate item respectively.
In step 402, the first feature vectors of all candidate items are fused to obtain a fused feature vector.
In some embodiments, the first feature vectors of all candidate items are pooled to obtain a fused feature vector. For example, Pooling includes maximum Pooling Max Pooling processing or Average Pooling processing.
In step 403, the fused feature vector is respectively fused with the first feature vector of each candidate item to obtain a second feature vector of each candidate item.
It should be noted here that, since the fused feature vector is associated with the candidate items in the entire candidate set, the fused feature vector is respectively fused with the first feature of each candidate item, so as to help determine the position of each candidate item in the entire candidate set.
In step 404, the second feature vectors of all candidate items are input into the ranking model to obtain a two-dimensional probability distribution table. The ranking model is trained by the training method of any one of the embodiments in fig. 1,
the ith column of the two-dimensional probability distribution table comprises the probability of the ith candidate item at each position in the recommended sequence, the jth row comprises the probability of each candidate item at the jth position in the recommended sequence, i is greater than or equal to 1, j is greater than or equal to M, and M is the total number of candidate items in the candidate set. For example, the two-dimensional probability distribution table is shown in table 1 above.
In step 405, an output matrix is obtained from the two-dimensional probability distribution table.
In some embodiments, in each column of the two-dimensional probability distribution table, the maximum probability value is set to 1, and probability values other than the maximum probability value are set to 0 to obtain an output matrix. As shown in table 2 above.
At step 406, the position of each candidate item in the candidate set in the recommended sequence is determined according to the output matrix so as to rank all candidate items in the candidate set.
In some embodiments, if the value of the ith candidate item in the candidate set in the jth row of the output matrix is 1, then the ith candidate item is determined to be located at the jth position in the recommended sequence.
For example, as shown in table 2 above, candidate item 1 has a value of 1 in row 1, indicating that candidate item 1 is located at position 1 in the recommended sequence. The value of candidate item 2 in row 2 is 1, indicating that candidate item 2 is located at position 2 in the recommended sequence. The value of candidate item 3 in row 3 is 1, indicating that candidate item 3 is located at position 3 in the recommended sequence. The value of candidate item 4 in row 4 is 1, indicating that candidate item 4 is located at position 4 in the recommended sequence. From this, the rank order of the 4 candidate items can be determined as candidate item 1, candidate item 2, candidate item 3, and candidate item 4.
Fig. 5 is a schematic structural diagram of an article sorting device according to an embodiment of the present disclosure. As shown in fig. 5, the article sorting apparatus includes a fifth processing module 51, a sixth processing module 52, a seventh processing module 53, and an eighth processing module 54.
The fifth processing module 51 is configured to input the relevant information of all candidate items in the candidate set into the feature extraction model to obtain the first feature vector of each candidate item respectively.
In some embodiments, the feature extraction model is a neural network model for extracting the first feature vector of each candidate item respectively.
The sixth processing module 52 is configured to fuse the first feature vectors of all candidate items to obtain a fused feature vector.
In some embodiments, the sixth processing module 52 pools the first feature vectors of all candidate items to obtain a fused feature vector. For example, Pooling includes maximum Pooling Max Pooling processing or Average Pooling processing.
The seventh processing module 53 is configured to fuse the fused feature vector with the first feature vector of each candidate item, respectively, to obtain a second feature vector of each candidate item.
The eighth processing module 54 is configured to input the second feature vectors of all candidate items into the ranking model to obtain the two-dimensional probability distribution table. The ranking model is trained by the training method of any one of the embodiments in fig. 1.
The ith column of the two-dimensional probability distribution table comprises the probability of the ith candidate item at each position in the recommended sequence, the jth row comprises the probability of each candidate item at the jth position in the recommended sequence, i is greater than or equal to 1, j is greater than or equal to M, and M is the total number of candidate items in the candidate set.
The eighth processing module 54 obtains an output matrix according to the two-dimensional probability distribution table.
In some embodiments, in each column of the two-dimensional probability distribution table, the maximum probability value is set to 1, and probability values other than the maximum probability value are set to 0 to obtain an output matrix. As shown in table 2 above.
The eighth processing module 54 determines the position of each candidate item in the candidate set in the recommended sequence according to the output matrix so as to rank all candidate items in the candidate set.
In some embodiments, if the value of the ith candidate item in the candidate set in the jth row of the output matrix is 1, then the ith candidate item is determined to be located at the jth position in the recommended sequence.
Fig. 6 is a schematic structural diagram of an article sequencing device according to another embodiment of the present disclosure. As shown in fig. 6, the item sequencing device includes a memory 61, a processor 62, a communication interface 63, and a bus 64. Fig. 6 differs from fig. 3 in that, in the embodiment shown in fig. 6, the processor 62 is configured to perform the method referred to in any of the embodiments of fig. 4 based on instructions stored in the memory.
The present disclosure also relates to a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the instructions, when executed by a processor, implement the method according to any one of the embodiments in fig. 4.
The following description is provided by way of a specific example of an article sequencing scheme to which the present disclosure relates.
As shown in fig. 7, the candidate set includes 4 candidate items. The relevant information of the 4 candidate items is input into the feature extraction model to obtain the characteristics 11 of the candidate item 1, the characteristics 21 of the candidate item 2, the characteristics 31 of the candidate item 3 and the features 41 of the candidate item 4, respectively.
Next, the feature vectors 11, 21, 31, and 41 of the 4 candidate items are fused to obtain a fused feature vector. And respectively fusing the fused feature vectors with the features 11 of the candidate item 1 to obtain features 12, respectively fusing the fused feature vectors with the features 21 of the candidate item 2 to obtain features 22, respectively fusing the fused feature vectors with the features 31 of the candidate item 3 to obtain features 32, respectively fusing the fused feature vectors with the features 41 of the candidate item 4 to obtain features 42.
The feature vectors 12, 22, 32, and 42 for the 4 candidate items are then input into the ranking model to obtain a two-dimensional probability distribution table. And obtaining an output matrix according to the two-dimensional probability distribution table, and determining the position of each candidate item in the candidate set in the recommended sequence according to the output matrix so as to sort all candidate items in the candidate set. And finally, presenting the sequencing result to the user.
In some embodiments, the functional unit modules described above can be implemented as a general purpose Processor, a Programmable Logic Controller (PLC), a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable Logic device, discrete Gate or transistor Logic, discrete hardware components, or any suitable combination thereof for performing the functions described in this disclosure.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program instructing relevant hardware, where the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The description of the present disclosure has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the disclosure in the form disclosed. Many modifications and variations will be apparent to practitioners skilled in this art. The embodiment was chosen and described in order to best explain the principles of the disclosure and the practical application, and to enable others of ordinary skill in the art to understand the disclosure for various embodiments with various modifications as are suited to the particular use contemplated.
Claims (14)
1. A ranking model training method comprising:
inputting the related information of all sample articles in the sample set into a feature extraction model to respectively obtain a first feature vector of each sample article;
fusing the first feature vectors of all the sample articles to obtain fused feature vectors;
fusing the fused feature vector with the first feature vector of each sample article respectively to obtain a second feature vector of each sample article;
inputting the second feature vectors of all the sample articles into a machine learning model to obtain a two-dimensional probability distribution table, wherein the ith column of the two-dimensional probability distribution table comprises the probability of the ith sample article at each position in a recommended sequence, the jth row comprises the probability of each sample article at the jth position in the recommended sequence, i is more than or equal to 1, j is more than or equal to N, and N is the total number of the sample articles in the sample set;
obtaining an output matrix according to the two-dimensional probability distribution table;
calculating a cross entropy loss function for each row and each column of the output matrix using a label matrix of the sample set;
and training the machine learning model by using the cross entropy loss function of each row and each column of the output matrix to obtain a sequencing model.
2. The method of claim 1, wherein deriving an output matrix from the two-dimensional probability distribution table comprises:
in each column of the two-dimensional probability distribution table, the maximum probability value is set to 1, and probability values other than the maximum probability value are set to 0 to obtain the output matrix.
3. The method of claim 1 or 2, wherein fusing the first feature vectors of all sample items comprises:
and pooling the first feature vectors of all the sample articles to obtain a fusion feature vector.
4. The method of claim 3, wherein,
the Pooling treatment includes a Max Pooling treatment or an Average Pooling treatment.
5. A ranking model training apparatus comprising:
the first processing module is configured to input relevant information of all sample articles in the sample set into the feature extraction model so as to respectively obtain a first feature vector of each sample article;
a second processing module configured to fuse the first feature vectors of all the sample items to obtain a fused feature vector;
the third processing module is configured to fuse the fused feature vector with the first feature vector of each sample article respectively to obtain a second feature vector of each sample article;
a fourth processing module configured to input the second feature vectors of all the sample items into a machine learning model to obtain a two-dimensional probability distribution table, wherein an ith column of the two-dimensional probability distribution table includes probabilities of the ith sample item at each position in a recommended sequence, a jth row includes probabilities of each sample item at a jth position in the recommended sequence, i is greater than or equal to 1, j is greater than or equal to N, and N is a total number of sample items in the sample set; obtaining an output matrix according to the two-dimensional probability distribution table; calculating a cross entropy loss function for each row and each column of the output matrix using a label matrix of the sample set; and training the machine learning model by using the cross entropy loss function of each row and each column of the output matrix to obtain a sequencing model.
6. A ranking model training apparatus comprising:
a memory configured to store instructions;
a processor coupled to the memory, the processor configured to perform implementing the method of any of claims 1-4 based on instructions stored by the memory.
7. A method of sequencing items, comprising:
inputting relevant information of all candidate articles in the candidate set into a feature extraction model to respectively obtain a first feature vector of each candidate article;
fusing the first feature vectors of all the candidate articles to obtain fused feature vectors;
fusing the fused feature vector with the first feature vector of each candidate article respectively to obtain a second feature vector of each candidate article;
inputting the second feature vectors of all the candidate items into a ranking model to obtain a two-dimensional probability distribution table, wherein the ranking model is obtained by training according to the training method of any one of claims 1 to 4, the ith column of the two-dimensional probability distribution table comprises the probability of the ith candidate item at each position in a recommended sequence, the jth row comprises the probability of each candidate item at the jth position in the recommended sequence, i is greater than or equal to 1, j is less than or equal to M, and M is the total number of candidate items in the candidate set;
obtaining an output matrix according to the two-dimensional probability distribution table;
determining a position of each candidate item in the candidate set in the recommended sequence according to the output matrix so as to rank all candidate items in the candidate set.
8. The method of claim 7, wherein deriving an output matrix from the two-dimensional probability distribution table comprises:
in each column of the two-dimensional probability distribution table, the maximum probability value is set to 1, and probability values other than the maximum probability value are set to 0 to obtain the output matrix.
9. The method of claim 8, wherein determining the position of each candidate item in the candidate set in the recommended sequence according to the output matrix comprises:
and if the value of the ith candidate item in the candidate set in the jth row in the output matrix is 1, determining that the ith candidate item is located at the jth position in the recommended sequence.
10. The method according to any one of claims 7-9, wherein fusing the first feature vectors of all candidate items comprises:
and pooling the first feature vectors of all the candidate articles to obtain a fusion feature vector.
11. The method of claim 10, wherein,
the Pooling treatment includes a Max Pooling treatment or an Average Pooling treatment.
12. An article sequencing device, comprising:
the fifth processing module is configured to input the relevant information of all candidate items in the candidate set into the feature extraction model to respectively obtain a first feature vector of each candidate item;
a sixth processing module configured to fuse the first feature vectors of all the candidate items to obtain a fused feature vector;
a seventh processing module, configured to fuse the fused feature vector with the first feature vector of each candidate item, respectively, to obtain a second feature vector of each candidate item;
an eighth processing module, configured to input the second feature vectors of all the candidate items into a ranking model to obtain a two-dimensional probability distribution table, wherein the ranking model is obtained by training according to any one of claims 1 to 4, an ith column of the two-dimensional probability distribution table includes probabilities of an ith candidate item at each position in a recommended sequence, a jth row includes probabilities of each candidate item at a jth position in the recommended sequence, 1 ≦ i, j ≦ M, and M is a total number of candidate items in the candidate set; obtaining an output matrix according to the two-dimensional probability distribution table; determining a position of each candidate item in the candidate set in the recommended sequence according to the output matrix so as to rank all candidate items in the candidate set.
13. An article sequencing device, comprising:
a memory configured to store instructions;
a processor coupled to the memory, the processor configured to perform implementing the method of any of claims 6-9 based on instructions stored by the memory.
14. A computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions which, when executed by a processor, implement the method of any one of claims 1-4, 7-11.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110684099.1A CN113379500A (en) | 2021-06-21 | 2021-06-21 | Sequencing model training method and device, and article sequencing method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110684099.1A CN113379500A (en) | 2021-06-21 | 2021-06-21 | Sequencing model training method and device, and article sequencing method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113379500A true CN113379500A (en) | 2021-09-10 |
Family
ID=77577913
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110684099.1A Pending CN113379500A (en) | 2021-06-21 | 2021-06-21 | Sequencing model training method and device, and article sequencing method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113379500A (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120233159A1 (en) * | 2011-03-10 | 2012-09-13 | International Business Machines Corporation | Hierarchical ranking of facial attributes |
CN106897419A (en) * | 2017-02-23 | 2017-06-27 | 同济大学 | The study recommendation method that sorted to level of fusion social information |
CN107291822A (en) * | 2017-05-24 | 2017-10-24 | 北京邮电大学 | The problem of based on deep learning disaggregated model training method, sorting technique and device |
CN108804689A (en) * | 2018-06-14 | 2018-11-13 | 合肥工业大学 | The label recommendation method of the fusion hidden connection relation of user towards answer platform |
WO2020211566A1 (en) * | 2019-04-18 | 2020-10-22 | 腾讯科技(深圳)有限公司 | Method and device for making recommendation to user, computing apparatus, and storage medium |
CN111831898A (en) * | 2019-09-17 | 2020-10-27 | 北京嘀嘀无限科技发展有限公司 | Sorting method and device, electronic equipment and readable storage medium |
WO2020221298A1 (en) * | 2019-04-30 | 2020-11-05 | 北京金山云网络技术有限公司 | Text detection model training method and apparatus, text region determination method and apparatus, and text content determination method and apparatus |
CN112084435A (en) * | 2020-08-07 | 2020-12-15 | 北京三快在线科技有限公司 | Search ranking model training method and device and search ranking method and device |
-
2021
- 2021-06-21 CN CN202110684099.1A patent/CN113379500A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120233159A1 (en) * | 2011-03-10 | 2012-09-13 | International Business Machines Corporation | Hierarchical ranking of facial attributes |
CN106897419A (en) * | 2017-02-23 | 2017-06-27 | 同济大学 | The study recommendation method that sorted to level of fusion social information |
CN107291822A (en) * | 2017-05-24 | 2017-10-24 | 北京邮电大学 | The problem of based on deep learning disaggregated model training method, sorting technique and device |
CN108804689A (en) * | 2018-06-14 | 2018-11-13 | 合肥工业大学 | The label recommendation method of the fusion hidden connection relation of user towards answer platform |
WO2020211566A1 (en) * | 2019-04-18 | 2020-10-22 | 腾讯科技(深圳)有限公司 | Method and device for making recommendation to user, computing apparatus, and storage medium |
WO2020221298A1 (en) * | 2019-04-30 | 2020-11-05 | 北京金山云网络技术有限公司 | Text detection model training method and apparatus, text region determination method and apparatus, and text content determination method and apparatus |
CN111831898A (en) * | 2019-09-17 | 2020-10-27 | 北京嘀嘀无限科技发展有限公司 | Sorting method and device, electronic equipment and readable storage medium |
CN112084435A (en) * | 2020-08-07 | 2020-12-15 | 北京三快在线科技有限公司 | Search ranking model training method and device and search ranking method and device |
Non-Patent Citations (1)
Title |
---|
李志宇: "在线商品评论效用排序模型研究", 《情报分析与研究》, no. 4, 30 April 2013 (2013-04-30), pages 62 - 65 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109871532B (en) | Text theme extraction method and device and storage medium | |
CN106651542B (en) | Article recommendation method and device | |
CN110458627B (en) | Commodity sequence personalized recommendation method for dynamic preference of user | |
CN106815192A (en) | Model training method and device and sentence emotion identification method and device | |
CN109829775A (en) | A kind of item recommendation method, device, equipment and readable storage medium storing program for executing | |
CN111444334A (en) | Data processing method, text recognition device and computer equipment | |
CN109389168A (en) | Project recommendation model training method, item recommendation method and device | |
CN109189921B (en) | Comment evaluation model training method and device | |
CN108108347B (en) | Dialogue mode analysis system and method | |
KR102134472B1 (en) | A method for searching optimal structure of convolution neural network using genetic algorithms | |
CN103761254A (en) | Method for matching and recommending service themes in various fields | |
CN109299462A (en) | Short text similarity calculating method based on multidimensional convolution feature | |
CN107918778B (en) | Information matching method and related device | |
CN107203558B (en) | Object recommendation method and device, and recommendation information processing method and device | |
CN108920665A (en) | Recommendation score method and device based on network structure and comment text | |
CN111080360A (en) | Behavior prediction method, model training method, device, server and storage medium | |
CN107016589A (en) | The determination method and device of recommended products | |
CN110969172A (en) | Text classification method and related equipment | |
CN112417848A (en) | Corpus generation method and device and computer equipment | |
EP2988229A1 (en) | Information retrieval method and device | |
CN105260458A (en) | Video recommendation method for display apparatus and display apparatus | |
CN106600347A (en) | Method for constructing sequence forecasting model based on multi-visual-angle data and cycle network | |
CN110796504B (en) | Article recommendation method and device | |
CN110134861B (en) | Recommendation method based on item categories and user preferences | |
CN113379500A (en) | Sequencing model training method and device, and article sequencing method and device |
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 |