CN111506822B - Data coding and information recommending method, device and equipment - Google Patents
Data coding and information recommending method, device and equipment Download PDFInfo
- Publication number
- CN111506822B CN111506822B CN202010471060.7A CN202010471060A CN111506822B CN 111506822 B CN111506822 B CN 111506822B CN 202010471060 A CN202010471060 A CN 202010471060A CN 111506822 B CN111506822 B CN 111506822B
- Authority
- CN
- China
- Prior art keywords
- data
- vector
- term behavior
- encoding
- coding
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 76
- 239000013598 vector Substances 0.000 claims abstract description 263
- 230000007774 longterm Effects 0.000 claims abstract description 103
- 238000013528 artificial neural network Methods 0.000 claims abstract description 59
- 230000006399 behavior Effects 0.000 claims description 256
- 238000012545 processing Methods 0.000 claims description 36
- 238000010606 normalization Methods 0.000 claims description 29
- 230000007246 mechanism Effects 0.000 claims description 13
- 238000013527 convolutional neural network Methods 0.000 claims description 11
- 230000000694 effects Effects 0.000 claims description 11
- 238000010276 construction Methods 0.000 claims description 4
- 238000012549 training Methods 0.000 description 14
- 230000004927 fusion Effects 0.000 description 13
- 230000008569 process Effects 0.000 description 13
- 238000004590 computer program Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 10
- 230000006870 function Effects 0.000 description 7
- 230000009286 beneficial effect Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 239000000284 extract Substances 0.000 description 2
- 238000000926 separation method Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application provides a data coding and information recommending method, device and equipment. The method comprises the following steps: the state data used for determining the attention weight is input into a first neural network to obtain a first coding vector. And inputting the long-term behavior data of the user, the first interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior into a second neural network to obtain a second coding vector. And inputting the short-term behavior data of the user, the second interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior into a third neural network to obtain a third coding vector. And determining an attention weight vector according to the first coding vector. And encoding the second encoding vector and the third encoding vector based on the attention weight vector.
Description
Technical Field
The present application relates to computer technology, and in particular, to a method, apparatus, and device for data encoding and information recommendation.
Background
When a user accesses a business system, the business system typically makes information recommendations to the user.
When the system recommends information to the user, the long-term behavior data and the short-term behavior data of the user are generally referred to so as to recommend information matched with the expectations of the user.
Disclosure of Invention
In view of the above, the present application discloses a data encoding method, comprising:
inputting state data for determining attention weight into a first neural network to obtain a first coding vector;
inputting the long-term behavior data of the user, the first interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior into a second neural network to obtain a second coding vector;
inputting the short-term behavior data of the user, the occurrence time of the short-term behavior and the second interval duration data between the occurrence time of the last short-term behavior of the short-term behavior into a third neural network to obtain a third coding vector;
determining an attention weight vector according to the first coding vector;
and encoding the second encoding vector and the third encoding vector based on the attention weight vector.
In an embodiment shown, the method further comprises:
and further encoding the intermediate vector obtained by encoding the second encoded vector and the third encoded vector based on the attention weight vector with the state data.
In one embodiment, the encoding the second encoding vector and the third encoding vector based on the attention weight vector includes:
multiplying the attention weight vector by the second code vector to obtain a first result;
multiplying the result obtained by subtracting the attention weight vector from 1 by the third coding vector to obtain a second result;
and adding the first result and the second result.
In an embodiment shown, determining the attention weight vector according to the first coding vector includes:
normalizing each dimension in the first code vector;
and constructing the attention weight vector based on the normalized data of each dimension.
In one embodiment, the constructing the attention weight vector based on the normalized dimension data includes:
taking each dimension data after normalization processing as a molecule respectively, taking the sum of each dimension data after normalization processing as a denominator, and obtaining weight values respectively corresponding to each dimension data;
And constructing an attention weight vector based on the weight values corresponding to the data of each dimension.
In an embodiment, the intermediate vector obtained by encoding the second encoded vector and the third encoded vector based on the attention weight vector further encodes the state data, and the intermediate vector includes any one or a combination of the following:
splicing the intermediate vector with the first coding vector;
adding the intermediate vector to the first encoded vector;
multiplying the intermediate vector by the first code vector.
In the illustrated embodiment, the first neural network, the second neural network, and the third neural network are networks or attention mechanism networks constructed based on any one or a combination of several of the following:
an RNN network; a transducer network; LSTM networks; CNN networks.
In an embodiment, the long-term behavior of the user includes a behavior in which an interval duration between a behavior occurrence time and a current encoding time reaches a preset duration;
the user short-term behavior comprises the behavior that the interval time between the behavior occurrence time and the current coding time does not reach the preset time;
The status data includes any one or more of the following:
system state data corresponding to the current code; business activity state data corresponding to the code; user state data of the user.
The application also discloses an information recommendation method, which comprises the following steps:
acquiring long-term behavior data, short-term behavior data, first interval duration data, second interval duration data and state data for determining attention weights of a target user;
the first interval duration data includes interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior; the second interval duration data includes interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior;
coding the long-term behavior data, the short-term behavior data, the first interval duration data, the second interval duration data and the state data by adopting any one of the data coding methods disclosed in the claims 1-8 to obtain a coding result;
based on the coding result, determining recommendation information corresponding to the current recommendation, and outputting the recommendation information.
The application also discloses a data coding device, which comprises:
the first coding module is used for inputting state data for determining attention weight into a first neural network to obtain a first coding vector;
the second coding module inputs the long-term behavior data of the user, the first interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior into a second neural network to obtain a second coding vector;
the third coding module inputs the short-term behavior data of the user, the second interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior into a third neural network to obtain a third coding vector;
the attention weight determining module is used for determining an attention weight vector according to the first coding vector;
and a fourth encoding module configured to encode the second encoding vector and the third encoding vector based on the attention weight vector.
In one embodiment shown, the apparatus further comprises:
and a further encoding module configured to encode an intermediate vector obtained by encoding the second encoding vector and the third encoding vector based on the attention weight vector, and further encode the intermediate vector and the state data.
In an embodiment shown, the fourth coding module includes:
multiplying the attention weight vector by the second code vector to obtain a first result;
multiplying the result obtained by subtracting the attention weight vector from 1 by the third coding vector to obtain a second result;
and adding the first result and the second result.
In an embodiment shown, the attention weight determination module includes:
the normalization processing module is used for performing normalization processing on each dimension in the first coding vector;
the construction module is used for constructing the attention weight vector based on the normalized data in each dimension.
In an embodiment, the normalization processing module includes:
taking each dimension data after normalization processing as a molecule respectively, taking the sum of each dimension data after normalization processing as a denominator, and obtaining weight values respectively corresponding to each dimension data;
and constructing an attention weight vector based on the weight values corresponding to the data of each dimension.
In an embodiment shown, the above further coding module comprises any one or a combination of the following:
splicing the intermediate vector with the first coding vector;
Adding the intermediate vector to the first encoded vector;
multiplying the intermediate vector by the first code vector.
In the illustrated embodiment, the first neural network, the second neural network, and the third neural network are networks or attention mechanism networks constructed based on any one or a combination of several of the following:
an RNN network; a transducer network; LSTM networks; CNN networks.
In an embodiment, the long-term behavior of the user includes a behavior in which an interval duration between a behavior occurrence time and a current encoding time reaches a preset duration;
the user short-term behavior comprises the behavior that the interval time between the behavior occurrence time and the current coding time does not reach the preset time;
the status data includes any one or more of the following:
system state data corresponding to the current code; business activity state data corresponding to the code; user state data of the user.
The application also discloses an information recommendation device, which comprises:
the acquisition module acquires long-term behavior data, short-term behavior data, first interval duration data, second interval duration data and state data for determining attention weight of a target user;
The first interval duration data includes interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior; the second interval duration data includes interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior;
the coding module adopts any one of the data coding methods disclosed in the claims 1-8 to code the long-term behavior data, the short-term behavior data, the first interval duration data, the second interval duration data and the state data to obtain a coding result;
and the recommendation module is used for determining recommendation information corresponding to the current recommendation based on the coding result and outputting the recommendation information.
The application also discloses a data coding device, which comprises:
a processor;
a memory for storing processor-executable instructions;
wherein the processor executes the executable instructions to implement the data encoding method disclosed in any of the embodiments.
The application also discloses an information recommendation device, which comprises:
A processor;
a memory for storing processor-executable instructions;
the processor executes the executable instructions to implement the information recommendation method disclosed in any of the embodiments.
According to the technical scheme, in the process of coding the long-term behavior of the user, on one hand, the long-term behavior data of the user and the first interval duration data are fused and coded; fusion encoding is carried out on the short-term behavior data of the user and the second interval duration data, so that more comprehensive characteristic information is extracted;
on the other hand, according to the first coding vector, an attention weight vector is determined, and based on the attention weight vector, the first coding vector and the second coding vector are subjected to fusion coding, so that the feature is extracted by fusing the long-term behavior and the short-term behavior, and more beneficial features are extracted.
Of course, the accuracy of the recommendation information determined by the encoded data obtained by encoding by the encoding method is higher.
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 application as claimed.
Drawings
In order to more clearly illustrate one or more embodiments of the present application or the technical solutions in the related art, the drawings that are required for the description of the embodiments or the related art will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments described in one or more embodiments of the present application, and other drawings may be obtained according to the drawings without inventive effort to those of ordinary skill in the art.
FIG. 1 is a method flow chart of a data encoding method of the present application;
FIG. 2 is a flow chart of a method for recommending information according to the present application;
FIG. 3 is a flow chart of a method for recommending information according to the present application;
FIG. 4 is a block diagram of an information recommendation network according to the present application;
fig. 5 is a block diagram showing a data encoding apparatus of the present application;
FIG. 6 is a block diagram of an information recommendation device according to the present application;
fig. 7 is a hardware configuration diagram of a data encoding apparatus according to the present application;
fig. 8 is a hardware configuration diagram of an information recommendation apparatus according to the present application.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the application. Rather, they are merely examples of apparatus and methods consistent with aspects of the application as detailed in the accompanying claims.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any or all possible combinations of one or more of the associated listed items. It will also be appreciated that the term "if," as used herein, may be interpreted as "at … …" or "at … …" or "responsive to a determination," depending on the context.
When a user accesses a business system, the business system typically makes information recommendations to the user.
When the system recommends information to the user, the long-term behavior data and the short-term behavior data of the user are generally referred to so as to recommend information matched with the expectations of the user.
The following briefly describes how the business system makes information recommendations with reference to the user's long-term behavior data and short-term behavior data.
In the related art, a service system needs to encode a long-term behavior of a user and a short-term behavior of the user.
At this time, the service system generally encodes the long-term behavior of the user and the short-term behavior of the user, respectively, to obtain a long-term behavior encoding vector and a short-term behavior encoding vector.
After the coding is finished, the service system inputs the long-term behavior coding vector and the short-term behavior coding vector into a pre-trained multi-classifier to calculate, and information to be recommended is obtained. The multi-classifier is obtained by training based on sample data marked with a plurality of recommended information.
After the information to be recommended is obtained, the service system can output the information to be recommended to the user.
According to the scheme, in the related technology, in the process of coding the long-short-term behaviors of the user, only the characteristics related to the long-short-term behaviors are extracted respectively, so that on one hand, the extracted characteristics are not comprehensive; on the other hand, the extracted features are not screened, and the emphasis is not highlighted. Of course, the accuracy of the recommendation information determined by the encoded data obtained by encoding by the encoding method is also low.
Based on the above, the present application proposes a data encoding method. In the process of encoding the long-short-period behaviors of the user, on one hand, the method extracts the related characteristics of the interval duration data of the adjacent long-short-period behaviors; on the other hand, determining the attention weight based on the state data for determining the attention weight has an important point of fusing long-term behavior with short-term behavior to extract features, thereby extracting more beneficial features. Of course, the accuracy of the information recommendation network recommendation information for data coding by adopting the coding method is higher.
The following description is made with reference to specific embodiments.
Referring to fig. 1, fig. 1 is a flowchart of a data encoding method according to the present application. As shown in fig. 1, the method may include:
s102, inputting state data for determining attention weights into a first neural network to obtain a first coding vector;
s104, inputting the long-term behavior data of the user, the first interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior into a second neural network to obtain a second coding vector;
s106, inputting the short-term behavior data of the user, the occurrence time of the short-term behavior and the second interval duration data between the occurrence time of the last short-term behavior of the short-term behavior into a third neural network to obtain a third coding vector;
s108, determining an attention weight vector according to the first coding vector;
s110, encoding the second encoding vector and the third encoding vector based on the attention weight vector.
The data encoding method described above may be installed in any system in the form of a software device. For example, the system may be an information recommendation system.
It will be appreciated that the above system may be implemented in any terminal device. Such as PC terminals, mobile terminals, PAD terminals, etc. In implementing the above method, it is generally necessary to provide a computing power to the apparatus on which the method is mounted.
The following describes a system in which the method is carried by the execution subject.
The state data may specifically be state data for determining attention weights. In practical applications, the status data may include any one or more of the following:
system state data corresponding to the current code; business activity state data corresponding to the code; user state data of the user.
In practical applications, the system state data corresponding to the present encoding may specifically be state information of a system executing the present encoding, for example, a system time corresponding to the present encoding, a system version, and the like. The business activity state data corresponding to the present code may be business activity (for example, a proxy bank card activity) in which the system is being promoted during the present code. The user status data of the user may include the age, interest, sex, credit record, etc. of the user.
The system may first fuse the state data when encoding the state data. Specific ways of fusing the state data include, but are not limited to, adding, multiplying and splicing.
For example, the system may splice and fuse the system state data corresponding to the current code, the service activity state data corresponding to the current code, and the user state data of the user, to obtain fused state data.
After the fused data is obtained, the system can input the fused state data into a first neural network to encode, and a first encoding vector is obtained.
Wherein, the first neural network may be a neural network constructed based on any one or a combination of several networks (the specific structure of the network may refer to the related art and is not described in detail herein):
an RNN network; a transducer network; LSTM networks; CNN networks.
In order to extract more relevant features from the state data, in an embodiment, the first neural network may be an attention mechanism network constructed based on any one or a combination of several networks (the specific structure of the attention mechanism network may refer to the relevant technology and is not described in detail herein):
An RNN network; a transducer network; LSTM networks; CNN networks.
In one embodiment, the data after fusion may be encoded without using a neural network. For example, an encoding rule (for example, normalization processing or normalization processing) may be preset in the above system, and during encoding, the above system may map the obtained fused data according to the encoding rule to obtain a first encoding vector.
The long-term behavior data may be a behavior sequence corresponding to a behavior in which a time interval between a behavior occurrence time and a current encoding time reaches a preset time, where the behavior sequence is counted in the system.
The preset time length can be set according to actual service requirements. For example, the preset time period may be 10 hours. At this time, the behavior in which the interval duration between the behavior occurrence time and the current encoding time counted by the system reaches 10 hours may be regarded as a long-term behavior.
The first interval duration data is specifically interval duration data between the occurrence of the long-term behavior and the last long-term behavior of the occurrence of the long-term behavior.
For example, the first interval duration may be an interval duration between counting the current long-term behavior time and counting the last long-term behavior time in the system.
When the long-term behavior data of the user and the first interval duration data are encoded, the system may first fuse the long-term behavior data of the user and the first interval duration data. Specific ways of performing fusion include, but are not limited to, adding, multiplying, and splicing.
For example, the system may perform splice fusion on the long-term behavior data of the user and the first interval duration data to obtain fused behavior data.
After the fused behavior data is obtained, the system can input the fused behavior data into a second neural network to encode, and a second encoding vector is obtained.
Wherein, the second neural network may be a network constructed based on any one or a combination of several networks (the specific structure of the network may refer to the related art and is not described in detail here):
an RNN network; a transducer network; LSTM networks; CNN networks.
In order to extract more relevant features from the state data, in an embodiment, the second neural network may be an attention mechanism network constructed based on any one or a combination of several networks (the specific structure of the attention mechanism network may refer to the relevant technology and is not described in detail here):
An RNN network; a transducer network; LSTM networks; CNN networks.
In one embodiment, the fused behavior data may be encoded without using a neural network. For example, the encoding rule (for example, normalization processing or normalization processing, etc.) may be preset in the above system. During encoding, the system can map the fused behavior data according to the encoding rule to obtain a second encoding vector.
The short-term behavior data may be a behavior sequence corresponding to a behavior in which a time interval between a behavior occurrence time and a current encoding time does not reach the preset time interval, which is counted in the system.
For example, the preset time period may be 10 hours. At this time, the behavior in which the interval duration between the behavior occurrence time and the current encoding time counted by the system does not reach 10 hours may be regarded as a short-term behavior.
The second interval duration data may specifically be interval duration data between the occurrence of the short-term behavior and the last short-term behavior in which the short-term behavior occurs.
For example, the second interval duration may be an interval duration between counting the current short-term behavior time and counting the last short-term behavior time in the system.
The system may first fuse the user short-term behavior data with the second interval duration data when encoding the user short-term behavior data with the second interval duration data. Specific ways of performing fusion include, but are not limited to, adding, multiplying, and splicing.
For example, the system may perform stitching and fusion on the short-term behavior data of the user and the second interval duration data to obtain fused behavior data.
After the fused behavior data is obtained, the system can input the fused behavior data into a third neural network for coding to obtain a third coding vector.
Wherein, the third neural network may be a network constructed based on any one or a combination of several networks (the specific structure of the network may refer to the related art and is not described in detail herein):
an RNN network; a transducer network; LSTM networks; CNN networks.
In order to extract more relevant features from the state data, in an embodiment, the third neural network may be an attention mechanism network constructed based on any one or a combination of several networks (the specific structure of the attention mechanism network may refer to the relevant technology and is not described in detail herein):
An RNN network; a transducer network; LSTM networks; CNN networks.
In one embodiment, the fused behavior data may be encoded without using a neural network. For example, the encoding rule (for example, normalization processing or normalization processing, etc.) may be preset in the above system. During encoding, the system can map the fused behavior data according to the encoding rule to obtain a third encoding vector.
In order to simplify the network operation, in one embodiment, the dimensions of the first encoding vector, the second encoding vector, and the third encoding vector may be the same.
The attention weight vector may specifically be a weight vector determined based on the first code vector.
In one embodiment, the system may normalize each of the data in the first encoded vector when determining the weight vector based on the first encoded vector.
In practical applications, the system may use a sigmod function to perform the normalization operation. The function used for performing the normalization processing operation may be set according to the actual service requirement, and is not particularly limited herein.
After normalizing each dimension of data in the first encoded vector, the system may construct an attention weight vector based on the normalized each dimension of data.
In practical applications, the system may construct the attention weight vector according to the dimension information of the first encoding vector where each dimension is located.
For example, assuming that the first encoding vector is a feature of A, B, C in three dimensions, A, B, C three-dimensional data obtained by normalizing the feature of each dimension is in the 1 st, 2 nd and 3 rd dimensions of the first encoding vector. At this time, the constructed attention weight vector may be expressed as { A, B, C }.
In an embodiment, when constructing the attention weight vector based on the normalized dimension data, the system may calculate the weight value of each dimension feature based on the SOFTMAX function, and construct the attention weight vector based on the weight value corresponding to each dimension data.
For example, when constructing the attention weight vector based on the normalized dimension data, the system may use the normalized dimension data as a numerator and the sum of the normalized dimension data as a denominator to obtain the weight value corresponding to each dimension data.
After obtaining the weight value corresponding to each data, the system can construct an attention weight vector according to the dimension information of the first coding vector of each data.
After the attention weight vector is obtained, the system may encode the second encoded vector and the third encoded vector.
In one embodiment, when the second encoding vector and the third encoding vector are encoded based on the attention weight vector, the system may multiply the attention weight vector by the second encoding vector to obtain a first result, and multiply 1 by the attention weight vector to obtain a second result.
After the first result and the second result are obtained, the system may add the first result to the second result.
Here, the order of calculating the first result and the second result is not particularly limited. On the other hand, when the first result and the second result are fused, vector calculation methods such as multiplying the first result or the second result and splicing may be employed.
It should be understood that, when the second encoding vector and the third encoding vector are encoded based on the attention weight vector, the system may also multiply the second encoding vector by a result obtained by subtracting the attention weight vector from 1 to obtain a first result, and multiply the attention weight vector by the third encoding vector to obtain a second result, which is not particularly limited herein.
According to the technical scheme, in the process of coding the long-term behavior of the user, on one hand, the long-term behavior data of the user and the first interval duration data are fused and coded; fusion encoding is carried out on the short-term behavior data of the user and the second interval duration data, so that more comprehensive characteristic information is extracted;
on the other hand, according to the first coding vector, an attention weight vector is determined, and based on the attention weight vector, the first coding vector and the second coding vector are subjected to fusion coding, so that the feature is extracted by fusing the long-term behavior and the short-term behavior, and more beneficial features are extracted.
Of course, the accuracy of the recommendation information determined by the encoded data obtained by encoding by the encoding method is higher.
In one embodiment, in order to further extract the more comprehensive features, after encoding the second encoding vector and the third encoding vector based on the attention weight vector, the system may further encode an intermediate vector obtained by encoding the second encoding vector and the third encoding vector based on the attention weight vector, and further encode the intermediate vector with the state data.
In practical applications, the system may use any one or a combination of the following methods when encoding the intermediate vector obtained by encoding the second encoding vector and the third encoding vector based on the attention weight vector, and further encoding the state data:
splicing the intermediate vector with the first coding vector;
adding the intermediate vector to the first encoded vector;
multiplying the intermediate vector by the first code vector.
The method of fusing the intermediate vector and the first encoded vector includes, but is not limited to, the method described above, and may include, for example, performing dot product operation on the intermediate vector and the first encoded vector, which is not described in detail herein.
In the present embodiment, since the state data for determining the attention weight and the above intermediate vectors are fused at the time of data encoding, more advantageous features can be extracted. Of course, the accuracy of the information recommendation network recommendation information for data coding by adopting the coding method is higher.
The application further provides an information recommendation method. According to the method, the information recommendation network for recommending information adopts the coding method shown in any embodiment, long-term behavior data, short-term behavior data, first interval duration data, second interval duration data and state data for determining attention weight of a target user are coded, so that characteristics beneficial to more recommended information can be extracted, and the accuracy of the recommended information is improved.
Referring to fig. 2, fig. 2 is a flowchart of a method for recommending information according to the present application, and as shown in fig. 2, the method includes:
s202, long-term behavior data, short-term behavior data, first interval duration data, second interval duration data and state data for determining attention weight of a target user are acquired;
The first interval duration data includes interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior; the second interval duration data includes interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior;
s204, coding the long-term behavior data, the short-term behavior data, the first interval duration data, the second interval duration data and the state data by adopting the data coding method disclosed in any embodiment to obtain a coding result;
s206, determining recommendation information corresponding to the current recommendation based on the coding result, and outputting the recommendation information.
In executing the step S204, the long-term behavior data, the short-term behavior data, the first interval duration data, the second interval duration data, and the state data for determining the attention weight of the target user may be input into a trained coding network to be calculated, so as to obtain a coding result. When executing S206, the coding result may be input into a trained multi-classifier to calculate, and recommendation information corresponding to the current recommendation may be obtained.
It should be noted that the coding network and the multi-classifier may be two independent networks or combined into one information recommendation network, which is not limited herein. The output of the encoding network may be the input of the multi-classifier.
The present embodiment is described below in connection with actual scenes.
Referring to fig. 3, fig. 3 is a flowchart of a method for recommending information according to the present application. As shown in fig. 3, the method may include:
s302, long-term behavior data, short-term behavior data, first interval duration data, second interval duration data and state data for determining attention weight of a target user are input into an information recommendation network to obtain information to be recommended;
the first interval duration data includes interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior; the second interval duration data includes interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior.
The method for encoding the long-period behavior data of the target user and the interval duration data of the long-period behavior of the target user in the information recommendation network adopts the data encoding method disclosed in any embodiment.
S304, outputting the information to be recommended.
Suppose that a user is accessing a banking system. The banking service system is provided with an information recommendation network, and can recommend related information to a user based on the network.
Referring to fig. 4, fig. 4 is a block diagram of an information recommendation network according to the present application.
As shown in fig. 4, the above-described information recommendation network includes a first neural network for encoding status data for determining attention weights.
The information recommendation network further comprises a second neural network for encoding long-term behavior of the user, and first interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior.
The information recommendation network further comprises a third neural network for encoding short-term behavior data of the user, and second interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior.
Wherein the first neural network, the second neural network, and the third neural network may be attention mechanism networks constructed based on RNN networks.
The information recommendation network further comprises a fusion device for constructing feature vectors based on an attention mechanism;
The information recommendation network further comprises a multi-classifier for recommending information to be recommended.
The training process of the information recommendation network is described below.
It should be noted here that, for convenience of illustrating the embodiment, only the process of training the information recommendation network will be described below.
It will be appreciated that the training of the information recommendation network, and in fact, the number of sub-networks included in the information recommendation network, is not described in detail herein.
During training, network parameters of the information recommendation network may be initialized first, and a loss function (e.g., cross entropy) used for training, and a number of training iterations may be determined.
In an embodiment, in order to train the network efficiently, when initializing the network parameters of the information recommendation network, the network parameters of the information recommendation network may be initialized based on the network parameters of the first neural network, the second neural network, and the third neural network after the pre-training. It should be noted that the method of pre-training is not limited herein with reference to the related art.
Of course, when initializing the network parameters of the information recommendation network, the network parameters may be randomly specified.
After the basic parameters are determined, a plurality of training samples marked with system recommendation information can be obtained.
In practical applications, the system may count the access behaviors of the user, and based on the counted access behaviors of the user, divide the access behaviors of the user into long-term behaviors of the user, short-term behaviors of the user, and actual access behaviors of the user (at this time, the actual access behaviors of the user may be regarded as the system recommendation information).
Based on the statistical access behavior, a plurality of training samples can be constructed.
The structure of the training sample is not limited herein.
After the training samples are obtained, the network parameters can be counter-propagated based on a gradient descent method until the information recommendation model converges.
Thus, training of the information recommendation network is completed.
The following description describes a process of information recommendation using the above information recommendation network.
After the system receives the access behaviors of the target users, the card handling activity data which are promoted by the system, timestamp data at the current moment and user state data of the target users can be spliced to form a 100-dimensional environment vector. The dimensions of the vector may be set according to actual services, and are not particularly limited herein.
Then, the system may input the environmental vector into the first neural network to perform calculation, so as to obtain a first coding vector of 100 dimensions.
The system may further take user behavior data other than 10 hours from the current time as user long-term behavior data, and splice the user long-term behavior data and the occurrence time of the user long-term behavior with first interval duration data between the occurrence time of the last long-term behavior of the user long-term behavior to form a first behavior vector of 100 dimensions.
The system may then input the first motion vector to the second neural network for computation to obtain a 100-dimensional second encoded vector.
The system may further take user behavior data within 10 hours from the current time as user short-term behavior data, and splice the user short-term behavior data and the occurrence time of the user short-term behavior with second interval duration data between the occurrence time of the last short-term behavior of the short-term behavior to form a second behavior vector of 100 dimensions.
The system may then input the second behavior vector to the third neural network for calculation to obtain a third encoded vector of 100 dimensions.
After the first code vector, the second code vector, and the third code vector are obtained, the system may input the first code vector, the second code vector, and the third code vector to the fusion device to perform vector fusion.
In the fusion device, the system may normalize each dimension data in the first code vector, and construct a 100-dimensional attention weight vector based on each dimension data after normalization.
After obtaining the attention vector, the system may multiply the attention weight vector with the second code vector to obtain a first result;
multiplying the result obtained by subtracting the attention weight vector from 1 by the third coding vector to obtain a second result;
and adding the first result and the second result to obtain a 100-dimensional intermediate vector.
After obtaining the intermediate vector, the system may splice the intermediate vector with the first encoded vector to obtain a 200-dimensional feature vector (i.e., input to the multi-classifier).
After the feature vector is obtained, the system can input the feature vector into the multi-classifier to calculate, so as to obtain information to be recommended.
After obtaining the information to be recommended, the system can output the information to be recommended. For example, the system may display the information to be recommended through an interface that interacts with the target user.
So far, the system completes the process of recommending information by using the information recommending network.
According to the technical scheme, the data coding method shown in any embodiment is adopted in the process of coding the long-short-term behaviors of the user, so that more beneficial characteristics of the recommended information are extracted, and the information recommendation accuracy is improved.
Corresponding to any embodiment, the application also provides a data encoding device. Referring to fig. 5, fig. 5 is a block diagram illustrating a data encoding apparatus according to the present application. As shown in fig. 5, the apparatus 500 includes:
the first encoding module 510 inputs the state data for determining the attention weight into the first neural network to obtain a first encoded vector;
the second encoding module 520 inputs the long-term behavior data of the user, and the first interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior into a second neural network to obtain a second encoding vector;
A third encoding module 530, configured to input the user short-term behavior data, and second interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior, into a third neural network to obtain a third encoding vector;
the attention weight determining module 540 determines an attention weight vector according to the first encoded vector;
the fourth encoding module 550 encodes the second encoding vector and the third encoding vector based on the attention weight vector.
In the illustrated embodiment, the apparatus 500 further includes:
the further encoding module 560 further encodes the intermediate vector obtained by encoding the second encoding vector and the third encoding vector based on the attention weight vector, with the state data.
In the illustrated embodiment, the fourth encoding module 550 includes:
multiplying the attention weight vector by the second code vector to obtain a first result;
multiplying the result obtained by subtracting the attention weight vector from 1 by the third coding vector to obtain a second result;
and adding the first result and the second result.
In the illustrated embodiment, the attention weight determination module 540 includes:
the normalization processing module is used for performing normalization processing on each dimension in the first coding vector;
the construction module is used for constructing the attention weight vector based on the normalized data in each dimension.
In an embodiment, the normalization processing module includes:
taking each dimension data after normalization processing as a molecule respectively, taking the sum of each dimension data after normalization processing as a denominator, and obtaining weight values respectively corresponding to each dimension data;
and constructing an attention weight vector based on the weight values corresponding to the data of each dimension.
In the illustrated embodiment, the further encoding module 560 includes any one or a combination of the following:
splicing the intermediate vector with the first coding vector;
adding the intermediate vector to the first encoded vector;
multiplying the intermediate vector by the first code vector.
In the illustrated embodiment, the first neural network, the second neural network, and the third neural network are networks or attention mechanism networks constructed based on any one or a combination of several of the following:
An RNN network; a transducer network; LSTM networks; CNN networks.
In an embodiment, the long-term behavior of the user includes a behavior in which an interval duration between a behavior occurrence time and a current encoding time reaches a preset duration;
the user short-term behavior comprises the behavior that the interval time between the behavior occurrence time and the current coding time does not reach the preset time;
the status data includes any one or more of the following:
system state data corresponding to the current code; business activity state data corresponding to the code; user state data of the user.
The application further provides an information recommendation device. Referring to fig. 6, fig. 6 is a block diagram of an information recommendation device according to the present application. As shown in fig. 6, the apparatus 600 includes:
an acquisition module 610 that acquires long-term behavior data, short-term behavior data, first interval duration data, second interval duration data, and state data for determining attention weights of a target user;
the first interval duration data includes interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior; the second interval duration data includes interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior;
The encoding module 620 encodes the long-term behavior data, the short-term behavior data, the first interval duration data, the second interval duration data, and the status data by using any one of the data encoding methods disclosed in claims 1-8, to obtain an encoding result;
the recommendation module 630 determines recommendation information corresponding to the current recommendation based on the encoding result, and outputs the recommendation information.
The embodiment of the data encoding device shown in the application can be applied to a data encoding device. The apparatus embodiments may be implemented by software, or may be implemented by hardware or a combination of hardware and software. Taking software implementation as an example, the device in a logic sense is formed by reading corresponding computer program instructions in a nonvolatile memory into a memory by a processor of an electronic device where the device is located for operation. In terms of hardware, as shown in fig. 7, a hardware structure diagram of a data encoding apparatus according to the present application is shown, and in addition to the processor, the memory, the network interface, and the nonvolatile memory shown in fig. 7, the electronic apparatus in which the device is located in the embodiment generally may include other hardware according to the actual function of the electronic apparatus, which is not described herein again.
Referring to fig. 7, a data encoding apparatus includes: a processor.
A memory for storing processor-executable instructions.
Wherein the processor executes the executable instructions to implement the data encoding method disclosed in any of the above embodiments.
The present application proposes a computer-readable storage medium storing a computer program for implementing the data encoding method disclosed in any one of the above embodiments.
The embodiment of the information recommending device shown in the application can be applied to information recommending equipment. The apparatus embodiments may be implemented by software, or may be implemented by hardware or a combination of hardware and software. Taking software implementation as an example, the device in a logic sense is formed by reading corresponding computer program instructions in a nonvolatile memory into a memory by a processor of an electronic device where the device is located for operation. In terms of hardware, as shown in fig. 8, a hardware structure diagram of an information recommendation device according to the present application is shown, and in addition to the processor, the memory, the network interface, and the nonvolatile memory shown in fig. 8, the electronic device in which the device is located in the embodiment generally may include other hardware according to the actual function of the electronic device, which is not described herein.
Referring to fig. 8, an information recommendation apparatus includes: a processor.
A memory for storing processor-executable instructions.
The processor executes the executable instructions to implement the information recommendation method disclosed in any of the embodiments.
The present application proposes a computer-readable storage medium storing a computer program for implementing the information recommendation method disclosed in any one of the above embodiments.
One skilled in the relevant art will recognize that one or more embodiments of the application may be provided as a method, system, or computer program product. Accordingly, one or more embodiments of the application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, one or more embodiments of the application may take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
The embodiments of the present application are described in a progressive manner, and the same and similar parts of the embodiments are all referred to each other, and each embodiment is mainly described in the differences from the other embodiments. In particular, for data processing apparatus embodiments, the description is relatively simple, as it is substantially similar to method embodiments, with reference to the description of method embodiments in part.
The foregoing describes certain embodiments of the present application. Other embodiments are within the scope of the following claims. In some cases, the acts or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
Embodiments of the subject matter and functional operations described in this disclosure may be implemented in the following: digital electronic circuitry, tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this application and structural equivalents thereof, or a combination of one or more of them. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory program carrier for execution by, or to control the operation of, data processing apparatus. Alternatively or additionally, the program instructions may be encoded on a manually-generated propagated signal, e.g., a machine-generated electrical, optical, or electromagnetic signal, that is generated to encode and transmit information to suitable receiver apparatus for execution by data processing apparatus. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.
The processes and logic flows described in this application can be performed by one or more programmable computers executing one or more computer programs to perform corresponding functions by operating on input data and generating output. The processes and logic flows described above may also be performed by, and apparatus may also be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).
Computers suitable for executing computer programs include, for example, general purpose and/or special purpose microprocessors, or any other type of central processing unit. Typically, the central processing unit will receive instructions and data from a read only memory and/or a random access memory. The essential elements of a computer include a central processing unit for carrying out or executing instructions and one or more memory devices for storing instructions and data. Typically, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks, etc. However, a computer does not have to have such a device. Furthermore, the computer may be embedded in another device, such as a mobile phone, a Personal Digital Assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device such as a Universal Serial Bus (USB) flash drive, to name a few.
Computer readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices including, for example, semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices), magnetic disks (e.g., internal hard disk or removable disks), magneto-optical disks, and CD-ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.
While the application contains many specific implementation details, these should not be construed as limiting the scope of any disclosure or the scope of the claims, but rather as primarily describing features of particular embodiments of the particular disclosure. Certain features that are described in this application in the context of separate embodiments can also be implemented in combination in a single embodiment. On the other hand, the various features described in the individual embodiments may also be implemented separately in the various embodiments or in any suitable subcombination. Furthermore, although features may be acting in certain combinations and even initially claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.
Similarly, although operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In some cases, multitasking and parallel processing may be advantageous. Moreover, the separation of various system modules and components in the embodiments described above should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.
Thus, particular embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve desirable results. Furthermore, the processes depicted in the accompanying drawings are not necessarily required to be in the particular order shown, or sequential order, to achieve desirable results. In some implementations, multitasking and parallel processing may be advantageous.
The foregoing description of the preferred embodiment(s) of the application is merely illustrative of the presently preferred embodiment(s) of the application, and is not intended to limit the embodiment(s) of the application to the particular form disclosed, but on the contrary, the intention is to cover all modifications, equivalents, alternatives, and alternatives falling within the spirit and scope of the embodiment(s) of the application.
Claims (14)
1. A data encoding method, comprising:
inputting state data for determining attention weight into a first neural network to obtain a first coding vector;
inputting the long-term behavior data of the user, the first interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior into a second neural network to obtain a second coding vector;
inputting the short-term behavior data of the user, the occurrence time of the short-term behavior and the second interval duration data between the occurrence time of the last short-term behavior of the short-term behavior into a third neural network to obtain a third coding vector;
normalizing each dimension in the first coding vector; constructing an attention weight vector based on the normalized data of each dimension; the value of the attention weight vector is between 0 and 1;
multiplying the attention weight vector by the second encoding vector to obtain a first result;
multiplying the result obtained by subtracting the attention weight vector from 1 by the third coding vector to obtain a second result;
adding the first result to the second result;
And an intermediate vector obtained by encoding the second encoding vector and the third encoding vector based on the attention weight vector is further encoded with the state data.
2. The method of claim 1, the constructing an attention weight vector based on the normalized dimension data, comprising:
taking each dimension data after normalization processing as a molecule respectively, taking the sum of each dimension data after normalization processing as a denominator, and obtaining weight values respectively corresponding to each dimension data;
and constructing an attention weight vector based on the weight values corresponding to the data of each dimension.
3. The method of claim 1, wherein the intermediate vector obtained by encoding the second encoding vector and the third encoding vector based on the attention weight vector is further encoded with the state data, including any one or a combination of the following:
splicing the intermediate vector with the first coding vector;
adding the intermediate vector to the first encoded vector;
multiplying the intermediate vector with the first encoded vector.
4. The method of claim 3, the first, second, and third neural networks being neural networks based on a neural network or attention mechanism constructed from any one or a combination of networks:
An RNN network; a transducer network; LSTM networks; CNN networks.
5. The method according to claim 4, wherein the long-term behavior of the user includes a behavior in which an interval duration between a behavior occurrence time and a current encoding time reaches a preset duration;
the short-term behavior of the user comprises the behavior that the interval duration between the behavior occurrence time and the current coding time does not reach the preset duration;
the status data includes any one or more of the following:
system state data corresponding to the current code; business activity state data corresponding to the code; user status data of the user.
6. An information recommendation method, comprising:
acquiring long-term behavior data, short-term behavior data, first interval duration data, second interval duration data and state data for determining attention weights of a target user;
the first interval duration data comprises interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior; the second interval duration data comprises interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior;
Coding the long-term behavior data, the short-term behavior data, the first interval duration data, the second interval duration data and the state data by adopting any one of the data coding methods disclosed in claims 1-5 to obtain a coding result;
and determining recommendation information corresponding to the current recommendation based on the coding result, and outputting the recommendation information.
7. A data encoding apparatus comprising:
the first coding module is used for inputting state data for determining attention weight into a first neural network to obtain a first coding vector;
the second coding module inputs the long-term behavior data of the user, the first interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior into a second neural network to obtain a second coding vector;
the third coding module inputs the short-term behavior data of the user, the second interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior into a third neural network to obtain a third coding vector;
the attention weight determining module comprises a normalization processing module and a construction module; wherein,
The normalization processing module is used for performing normalization processing on each dimension in the first coding vector;
the construction module is used for constructing an attention weight vector based on the normalized data of each dimension; the value of the attention weight vector is between 0 and 1;
a fourth encoding module that multiplies the attention weight vector by the second encoding vector to obtain a first result; multiplying the result obtained by subtracting the attention weight vector from 1 by the third coding vector to obtain a second result; adding the first result to the second result;
and a further encoding module for further encoding the intermediate vector obtained by encoding the second encoding vector and the third encoding vector based on the attention weight vector with the state data.
8. The apparatus of claim 7, the normalization processing module comprising:
taking each dimension data after normalization processing as a molecule respectively, taking the sum of each dimension data after normalization processing as a denominator, and obtaining weight values respectively corresponding to each dimension data;
and constructing an attention weight vector based on the weight values corresponding to the data of each dimension.
9. The apparatus of claim 7, the further encoding module comprising any one or a combination of:
splicing the intermediate vector with the first coding vector;
adding the intermediate vector to the first encoded vector;
multiplying the intermediate vector with the first encoded vector.
10. The apparatus of claim 9, the first, second, and third neural networks being networks or attention mechanism networks constructed based on a combination of any one or more of the following:
an RNN network; a transducer network; LSTM networks; CNN networks.
11. The apparatus of claim 10, wherein the long-term behavior of the user includes a behavior in which an interval duration between a behavior occurrence time and a current encoding time reaches a preset duration;
the short-term behavior of the user comprises the behavior that the interval duration between the behavior occurrence time and the current coding time does not reach the preset duration;
the status data includes any one or more of the following:
system state data corresponding to the current code; business activity state data corresponding to the code; user status data of the user.
12. An information recommendation apparatus, comprising:
the acquisition module acquires long-term behavior data, short-term behavior data, first interval duration data, second interval duration data and state data for determining attention weight of a target user;
the first interval duration data comprises interval duration data between the occurrence time of the long-term behavior and the occurrence time of the last long-term behavior of the long-term behavior; the second interval duration data comprises interval duration data between the occurrence time of the short-term behavior and the occurrence time of the last short-term behavior of the short-term behavior;
the coding module is used for coding the long-term behavior data, the short-term behavior data, the first interval duration data, the second interval duration data and the state data by adopting any data coding method disclosed in the claims 1-5 to obtain a coding result;
and the recommendation module is used for determining recommendation information corresponding to the current recommendation based on the coding result and outputting the recommendation information.
13. A data encoding apparatus, the apparatus comprising:
a processor;
a memory for storing processor-executable instructions;
Wherein the processor implements the data encoding method of any of claims 1-5 by executing the executable instructions.
14. An information recommendation device, the device comprising:
a processor;
a memory for storing processor-executable instructions;
wherein the processor implements the information recommendation method of claim 6 by executing the executable instructions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010471060.7A CN111506822B (en) | 2020-05-28 | 2020-05-28 | Data coding and information recommending method, device and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010471060.7A CN111506822B (en) | 2020-05-28 | 2020-05-28 | Data coding and information recommending method, device and equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111506822A CN111506822A (en) | 2020-08-07 |
| CN111506822B true CN111506822B (en) | 2023-08-18 |
Family
ID=71864506
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010471060.7A Active CN111506822B (en) | 2020-05-28 | 2020-05-28 | Data coding and information recommending method, device and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111506822B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110929164A (en) * | 2019-12-09 | 2020-03-27 | 北京交通大学 | Interest point recommendation method based on user dynamic preference and attention mechanism |
| CN110941764A (en) * | 2019-12-03 | 2020-03-31 | 腾讯科技（深圳）有限公司 | Object recommendation method and device, computer equipment and storage medium |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10296793B2 (en) * | 2016-04-06 | 2019-05-21 | Nec Corporation | Deep 3D attention long short-term memory for video-based action recognition |
| US10885905B2 (en) * | 2018-07-13 | 2021-01-05 | Samsung Electronics Co., Ltd. | Predicting user actions on ubiquitous devices |
-
2020
- 2020-05-28 CN CN202010471060.7A patent/CN111506822B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110941764A (en) * | 2019-12-03 | 2020-03-31 | 腾讯科技（深圳）有限公司 | Object recommendation method and device, computer equipment and storage medium |
| CN110929164A (en) * | 2019-12-09 | 2020-03-27 | 北京交通大学 | Interest point recommendation method based on user dynamic preference and attention mechanism |
Non-Patent Citations (1)
| Title |
|---|
| 王飞飞 ; 张生太 ; .移动社交网络微信用户信息发布行为统计特征分析.数据分析与知识发现.2018,(04),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111506822A (en) | 2020-08-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110366734B (en) | Optimizing neural network architecture | |
| CN109492772B (en) | Method and device for generating information | |
| WO2022174805A1 (en) | Model training method and apparatus, image processing method and apparatus, electronic device and storage medium | |
| CN109614414B (en) | User information determining method and device | |
| CN113538070B (en) | User life value cycle detection method and device and computer equipment | |
| CN113240510B (en) | Abnormal user prediction method, device, equipment and storage medium | |
| CN112801719A (en) | User behavior prediction method, user behavior prediction device, storage medium, and apparatus | |
| CN107403311B (en) | Account use identification method and device | |
| CN110706055A (en) | Commodity information pushing method and device, storage medium and computer equipment | |
| CN112149699A (en) | Method and device for generating model and method and device for recognizing image | |
| CN112883227A (en) | Video abstract generation method and device based on multi-scale time sequence characteristics | |
| CN113656699B (en) | User feature vector determining method, related equipment and medium | |
| CN112580368B (en) | Method, device, equipment and storage medium for identifying intention sequence of conversation text | |
| JP2024508502A (en) | Methods and devices for pushing information | |
| CN111506822B (en) | Data coding and information recommending method, device and equipment | |
| CN113344067A (en) | Method, device and equipment for generating customer portrait | |
| CN107665202B (en) | Method and device for constructing interest model and electronic equipment | |
| CN110390015B (en) | Data information processing method, device and system | |
| CN114358023B (en) | Intelligent question-answer recall method, intelligent question-answer recall device, computer equipment and storage medium | |
| CN113935788B (en) | Model evaluation method, device, equipment and computer readable storage medium | |
| CN115017362A (en) | Data processing method, electronic device and storage medium | |
| CN112017634B (en) | Data processing method, device, equipment and storage medium | |
| CN110209878B (en) | Video processing method and device, computer readable medium and electronic equipment | |
| CN113989562A (en) | Model training and image classification method and device | |
| CN112784032A (en) | Conversation corpus recommendation evaluation method and device, storage medium and electronic equipment |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |