CN113688327A - Data prediction method, device and equipment for fusion neural graph collaborative filtering network - Google Patents
Data prediction method, device and equipment for fusion neural graph collaborative filtering network Download PDFInfo
- Publication number
- CN113688327A CN113688327A CN202111013255.8A CN202111013255A CN113688327A CN 113688327 A CN113688327 A CN 113688327A CN 202111013255 A CN202111013255 A CN 202111013255A CN 113688327 A CN113688327 A CN 113688327A
- Authority
- CN
- China
- Prior art keywords
- network
- predicted
- input
- trained
- prediction
- 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.)
- Granted
Links
- 230000001537 neural effect Effects 0.000 title claims abstract description 82
- 238000001914 filtration Methods 0.000 title claims abstract description 77
- 238000000034 method Methods 0.000 title claims abstract description 50
- 230000004927 fusion Effects 0.000 title claims abstract description 17
- 238000003860 storage Methods 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims description 60
- 239000013598 vector Substances 0.000 claims description 31
- 230000006870 function Effects 0.000 claims description 26
- 238000004590 computer program Methods 0.000 claims description 15
- 238000010586 diagram Methods 0.000 claims description 15
- 238000012216 screening Methods 0.000 claims description 13
- 238000005516 engineering process Methods 0.000 abstract description 11
- 238000013473 artificial intelligence Methods 0.000 abstract description 7
- 230000003993 interaction Effects 0.000 description 8
- 238000012545 processing Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 5
- 230000006399 behavior Effects 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 4
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 238000013500 data storage Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 230000007480 spreading Effects 0.000 description 2
- 238000003892 spreading Methods 0.000 description 2
- 230000001960 triggered effect Effects 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- 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
- 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
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Entrepreneurship & Innovation (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Game Theory and Decision Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a data prediction method, a device, equipment and a storage medium for a fusion neural graph collaborative filtering network, and relates to an artificial intelligence technology.
Description
Technical Field
The invention relates to the technical field of artificial intelligence intelligent decision making, in particular to a data prediction method, a device, equipment and a storage medium for a fusion neural diagram collaborative filtering network.
Background
The matching sub-network in the original deep MCP model (a recommendation algorithm developed by the Alababa company) is mainly used for learning the cooperative information between a user and an item (namely, a user and an article), in particular to the matching sub-network in the original deep MCP model which is used for modeling a user-advertisement relationship and learning whether the user is matched with an advertisement, more particularly to a double-tower DSSM model (namely, a double-tower deep semantic matching model) adopted by the matching sub-network, the core idea of the double-tower DSSM model is to respectively learn to obtain two Embedding vectors which are respectively a user-side Embedding vector and an advertisement-side Embedding vector, a matching score (which can be understood as a matching score) is then computed using the two Embedding vectors, the existing matching sub-network only learns the direct association relationship between the user and the item, and cannot learn the higher-order association relationship between the user and the item, so that the accuracy of an output result is low when prediction is performed based on the deep MCP model.
Disclosure of Invention
The embodiment of the invention provides a data prediction method, a device, equipment and a storage medium for fusing a neural diagram collaborative filtering network, and aims to solve the problems that in the prior art, an original DeepMCP model is adopted in information recommendation, only the direct association relationship between a user and a product is learned, the higher-order association relationship between the user and the product cannot be learned, and the accuracy of an output result is low when prediction is carried out based on the DeepMCP model.
In a first aspect, an embodiment of the present invention provides a data prediction method for fusing a neural graph collaborative filtering network, including:
if a model training starting instruction is detected, acquiring a stored input feature set;
training the DeepMCP network to be trained through the input feature set to obtain the DeepMCP network fusing the neural map collaborative filtering network;
if the prediction instruction is detected, receiving input features to be predicted and task types to be predicted which are uploaded by a user side;
if the task type to be predicted is determined to be the click rate prediction task, inputting the input feature to be predicted into a prediction sub-network in the DeepMCP network as a first input feature to be predicted for operation to obtain a click rate prediction result; and
and if the task type to be predicted is determined to be a course and user interest correlation degree estimation task, forming a second input feature to be predicted by partial features obtained according to a preset screening strategy in the input features to be predicted, inputting the second input feature to be predicted into a matching sub-network in the deep MCP network, and calculating to obtain a course and user interest correlation degree prediction result.
In a second aspect, an embodiment of the present invention provides a data prediction apparatus fusing a neural graph collaborative filtering network, including:
the input feature set acquisition unit is used for acquiring a stored input feature set if a model training starting instruction is detected;
the model training unit is used for training the DeepMCP network to be trained through the input feature set to obtain the DeepMCP network fused with the neural map collaborative filtering network;
the uploaded data receiving unit is used for receiving the input characteristics to be predicted and the task types to be predicted uploaded by the user side if the prediction instruction is detected;
the first prediction unit is used for inputting the input features to be predicted as first input features to be predicted into a prediction sub-network in the deep MCP network for operation to obtain click rate prediction results if the type of the task to be predicted is determined to be a click rate prediction task; and
and the second prediction unit is used for forming a second input feature to be predicted by partial features obtained according to a preset screening strategy in the input features to be predicted and inputting the second input feature to be predicted into a matching sub-network in the deep MCP network for operation to obtain a prediction result of the relevance between the course and the user interest if the type of the task to be predicted is determined to be a course and user interest relevance estimation task.
In a third aspect, an embodiment of the present invention further provides a computer device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor, when executing the computer program, implements the data prediction method for the fused neural graph collaborative filtering network according to the first aspect.
In a fourth aspect, the embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and the computer program, when executed by a processor, causes the processor to execute the data prediction method of the fused neural graph collaborative filtering network according to the first aspect.
The embodiment of the invention provides a data prediction method, a device, equipment and a storage medium for fusing a neural diagram collaborative filtering network, wherein an input feature set is trained on a deep MCP network to be trained to obtain the deep MCP network fusing the neural diagram collaborative filtering network, then corresponding operation is carried out according to the input feature to be predicted uploaded by a user side and the task type to be predicted to obtain a prediction result, the user and a course are represented as bipartite graphs through the deep MCP network fusing the neural diagram collaborative filtering network, and the user and the course are embedded and processed through algorithm transmission on the graphs, so that the high-order user and product collaborative relationship can be fully learned by utilizing the interaction information of the user and the course, a matching network is strengthened, the high-order user and product collaborative relationship is transmitted to the prediction network, and the prediction result is more accurate.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a schematic view of an application scenario of a data prediction method for a fusion neural graph collaborative filtering network according to an embodiment of the present invention;
fig. 2 is a schematic flowchart of a data prediction method of a fusion neural graph collaborative filtering network according to an embodiment of the present invention;
fig. 3 is a schematic block diagram of a data prediction apparatus of a fusion neural graph collaborative filtering network according to an embodiment of the present invention;
FIG. 4 is a schematic block diagram of a computer device provided by an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It will be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It is also to be understood that the terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the specification of the present invention 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 be further understood that the term "and/or" as used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.
Referring to fig. 1 and fig. 2, fig. 1 is a schematic view of an application scenario of a data prediction method of a fusion neural graph collaborative filtering network according to an embodiment of the present invention; fig. 2 is a schematic flowchart of a data prediction method for a fusion neural graph collaborative filtering network according to an embodiment of the present invention, where the data prediction method for a fusion neural graph collaborative filtering network is applied to a server, and the method is executed by application software installed in the server.
As shown in fig. 2, the method includes steps S101 to S105.
S101, if a model training starting instruction is detected, a stored input feature set is obtained.
In this embodiment, a server is used as an execution subject to describe the technical solution. When the server detects a locally triggered model training start instruction, a locally stored input feature set needs to be acquired. The input feature set comprises a plurality of input features, and each input feature is divided into four groups, namely user features (such as user id, age and the like), query features (such as user historical behaviors), course features (such as course id) and time features.
For example, the user characteristics of the input characteristic a include id (such as a user account) of the user 1 and age of the user 1, the query characteristic of the input characteristic a includes query input by the user 1, the lesson characteristics include lesson id corresponding to the lesson clicked and watched by the user 1, and the time characteristic includes watching duration of the user 1. When the input features with the four groups of data are constructed locally, the input features can be combined into an input feature set to carry out model training on the DeepMCP network to be trained.
In the application, the difference between the deep mcp network to be trained and the existing deep mcp network is that a matching sub-network in the existing deep mcp network adopts a double-tower DSSM model, whereas in the application, the matching sub-network is realized by adopting an Ngcf network. That is, the sub-matching network in the deep mcp network to be trained in the present application is defined to adopt an Ngcf network (Ngcf network, i.e., a neural map collaborative filter network).
S102, training the DeepMCP network to be trained through the input feature set to obtain the DeepMCP network fusing the neural map collaborative filtering network.
In this embodiment, the to-be-trained DeepMCP network includes a to-be-trained prediction subnetwork, a to-be-trained matching subnetwork, and a to-be-trained association subnetwork, and the prediction subnetwork (i.e., prediction subnetwork), the matching subnetwork (i.e., matching subnetwork in which a neural graph collaborative filter network is used), and the association subnetwork (i.e., correlation subnetwork) are obtained after performing joint training based on the input feature set, respectively, so as to finally fuse the to-be-trained DeepMCP network with the neural graph collaborative filter network.
In one embodiment, step S102 includes:
acquiring a plurality of input features included in the input feature set; each input feature comprises a user feature, a query feature, a course feature and a time feature;
inputting each input feature as a first input feature into a to-be-trained prediction sub-network of the to-be-trained DeepMCP network for training to obtain a prediction sub-network;
inputting corresponding second input features consisting of the user features, the query features and the course features of all the input features into a matching sub-network to be trained of the deep MCP network to be trained to obtain a matching sub-network; wherein the matching sub-network is a neural map collaborative filtering network;
inputting the course sequence characteristics included in the course characteristics in the input characteristics as third input characteristics into a to-be-trained associated sub-network of the to-be-trained DeepMCP network for training to obtain an associated sub-network;
and fusing the prediction sub-network, the matching sub-network and the associated sub-network to obtain the DeepMCP network fused with the neural map collaborative filtering network.
In this embodiment, the matching sub-network is used to mine user-course connections, the association sub-network is used to mine course-course connections, and the prediction sub-network is used to mine feature-CTR connections (CTR represents click-through rate). The three sub-networks are trained together in a joint training mode, and meanwhile, the three sub-networks share the embedded network, so that updating of the matching subnet and the correlation subnet can also affect the embedded network, and the effect of the prediction subnet is affected. By the combined training method, the trained DeepMCP network fused with the neural map collaborative filtering network can predict results more accurately.
In an embodiment, step S102 is followed by:
and uploading the model parameters of the DeepMCP network to a block chain network.
In this embodiment, the corresponding digest information is obtained based on the model parameters of the depmcp network, and specifically, the digest information is obtained by performing hash processing on the model parameters of the depmcp network, for example, by using a sha256s algorithm. Uploading summary information to the blockchain can ensure the safety and the fair transparency of the user. The user equipment may download the summary information from the blockchain to verify if xx has been tampered with. The blockchain referred to in this example is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm, and the like.
A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like.
In an embodiment, the training of the to-be-trained prediction subnetwork of the to-be-trained DeepMCP network by using each input feature as the first input feature to obtain the prediction subnetwork includes:
acquiring a DNN-pCTR network to be trained as a prediction subnetwork to be trained;
respectively inputting four groups of characteristics included in the ith first input characteristic into an embedded network for operation and splicing to obtain an ith Embedding vector corresponding to the ith first input characteristic; the initial value of i is 1, the value range of i is [1, N ], and N represents the total number of input features included in the input feature set;
inputting the No. i Embedding vector into a to-be-trained prediction subnetwork for operation to obtain a No. i first prediction result corresponding to the No. i first input feature;
obtaining an ith loss function value corresponding to the ith first prediction result, obtaining an ith model parameter set of the to-be-trained prediction sub-network, and adjusting the ith model parameter set of the to-be-trained prediction sub-network according to the ith loss function value to obtain an ith adjusted model parameter set;
increasing i by 1 and updating the value of i;
if the value of i does not exceed N, returning to execute the step of inputting four groups of characteristics included in the ith first input characteristic into the embedded network respectively for operation and splicing to obtain the ith Embedding vector corresponding to the ith first input characteristic;
and if the value of i exceeds N, stopping model training, and acquiring the model parameter set after the Nth adjustment as the model parameter set of the DNN-pCTR network to be trained.
In this embodiment, the prediction subnet is a conventional DNN-pCTR network, that is, four groups of features included in the input features are embedded into the network to obtain a plurality of Embedding vectors (a boosting operation needs to be performed), then the Embedding vectors are spliced to form a vector m, then the vector m is fed to four groups of features included in the subsequent input features, and data of the last layer of network is pCTR (i.e., a click-through rate prediction result). The prediction sub-network is obtained by training the prediction sub-network to be trained of the DeepMCP network to be trained, and can be used as the prediction sub-network in the final DeepMCP network.
In an embodiment, the step of inputting the second input features, which are composed of the user features, the query features, and the course features of the input features, into the to-be-trained matching sub-network of the to-be-trained depmcp network for training to obtain the matching sub-network includes:
acquiring a neural diagram collaborative filtering network to be trained as a matching sub-network to be trained;
correspondingly constructing a second user-course graph of the jth by the jth second input characteristic; wherein the initial value of j is 1, the value range of j is [1, M ], and M represents the total number of input features included in the input feature set;
inputting the jth user-course bipartite graph into a neural graph collaborative filtering network to be trained for graph propagation to obtain a jth Embedding vector;
inputting the jh Embedding vector into a neural graph collaborative filtering network to be trained for prediction to obtain a jh score;
obtaining a j loss function value corresponding to the j score, obtaining a j model parameter set of a matching sub-network to be trained, and adjusting the j model parameter set of the matching sub-network to be trained according to the j loss function value to obtain a j adjusted model parameter set;
increasing j by 1 and updating the value of j;
if the value of j does not exceed M, returning to execute the step of constructing a jth user-course bipartite graph corresponding to the jth second input feature;
and if the value of j exceeds M, stopping model training, and acquiring the model parameter set after the M number adjustment as the model parameter set of the neural graph collaborative filtering network to be trained.
In this embodiment, the neural map collaborative filtering network to be trained, that is, the NGCF network to be trained, mainly includes an Embedding Layer, Embedding Propagation Layers, and a Prediction Layer. The user and the course are represented as bipartite graphs in the neural graph collaborative filtering network to be trained, and the user and the course are embeddedly transmitted on the graphs through an algorithm, so that the interaction information of the user and the course is utilized, the high-order connectivity between the user and the course is shown, and the collaborative filtering can be better carried out.
More specifically, the network architecture of the NGCF network includes an Embedding Layer, an Embedding Propagation Layer, and a Prediction Layer. Wherein, the Embeddings layer randomly initializes a trainable embedding for users and courses; the resulting embedding matrix is: e ═ Eu1,……,euN,ei1,……，eiM]. The Embedding Propagation layer, i.e. the multiple layers, is the same as the first layer, but is repeated several times, here taking one layer as an example. The propagation of the first layer is divided into two steps: message construction and message aggregation. When constructing the message, for the connected user-course pair (u, i), defining the message vector propagated by the course i to the user u as: m isu←i＝f(ei,eu,pui) Wherein p isuiIs the attenuation factor of the edge (u, i) at each propagation. When the messages are aggregated, the messages transmitted by all the neighbors of the user u are aggregated to update the vector representation of u:
where, (1) denotes one-layer propagation, and mu ← u denotes self-join.
The Prediction Layer gives the user u and the class i scores as:
finally, updating the network parameters by a loss function, wherein the loss function is as follows:
wherein, (u, i, j) epsilon O indicates that u and i have interaction, u and j have no interaction, sigma indicates a sigmod function, and finally
Regularization is l2 prevents overfitting. And training the NGCF network to be trained of the DeepMCP network to be trained to obtain a model parameter set of the neural map collaborative filtering network to be trained, wherein the model parameter set can be used as a matching sub-network in the final DeepMCP network.
In an embodiment, the training of the to-be-trained associated sub-network of the to-be-trained DeepMCP network performed by using the course sequence feature included in the course features in the input features as the third input feature obtains an associated sub-network, including:
and obtaining the course sequence characteristics corresponding to the input characteristics through a skip-gram model, inputting the course sequence characteristics serving as third input characteristics into a to-be-trained associated sub-network of the to-be-trained DeepMCP network for model training, and stopping the model training until the log-likelihood function is maximized to obtain the associated sub-network.
In the present embodiment, the correlation subnet is mainly responsible for mining the association between courses and courses, and users generally consider that there is a certain correlation between courses within a certain time window for a course click sequence of a user. The idea of skip-gram is used here, and for a course click sequence, the optimization goal is to maximize the log-likelihood function, and the associated sub-network is trained quickly in the above manner.
S103, receiving the input features to be predicted and the task types to be predicted uploaded by the user side if the prediction instruction is detected.
In this embodiment, when it is necessary to predict a relevant prediction result (for example, a click rate prediction result of a course) of a user based on a certain input feature, the input feature to be predicted and a task type to be predicted may be uploaded by a user side, and after receiving the input feature to be predicted and the task type to be predicted, the server further performs result prediction based on a trained DeepMCP network.
In an embodiment, step S103 is followed by:
acquiring user characteristics to be predicted, query characteristics to be predicted, course characteristics to be predicted and time characteristics to be predicted, wherein the user characteristics to be predicted, the query characteristics to be predicted, the course characteristics to be predicted and the time characteristics to be predicted are included in the input characteristics to be predicted;
taking the input feature to be predicted as a first input feature to be predicted;
and acquiring the user characteristic to be predicted, the query characteristic to be predicted and the course characteristic to be predicted in the input characteristics to be predicted according to the screening strategy, wherein the user characteristic to be predicted, the query characteristic to be predicted and the course characteristic to be predicted in the input characteristics to be predicted form a second input characteristic to be predicted.
In this embodiment, similar to the input features included in the input feature set, the input features to be predicted are also divided into four groups, that is, user features (such as user id, age, and the like), query features (such as user historical behavior), class features (such as class id), and time features, and more specifically, the input features to be predicted include the user features to be predicted, the query features to be predicted, the class features to be predicted, and the time features to be predicted. When the user characteristic to be predicted, the query characteristic to be predicted, the course characteristic to be predicted and the time characteristic to be predicted are obtained, the input characteristic to be predicted can be used as a first input characteristic to be predicted, the user characteristic to be predicted, the query characteristic to be predicted and the course characteristic to be predicted in the input characteristic to be predicted are obtained according to the screening strategy, and a second input characteristic to be predicted is formed by the user characteristic to be predicted, the query characteristic to be predicted and the course characteristic to be predicted in the input characteristic to be predicted. The sub-input features acquired in the mode are more representative and can be used as input parameters of a more accurate prediction process.
And S104, if the task type to be predicted is determined to be the click rate prediction task, inputting the input feature to be predicted as a first input feature to be predicted into a prediction sub-network in the deep MCP network for operation, and obtaining a click rate prediction result.
In this embodiment, when it is determined that the type of the task to be predicted is the click rate prediction task, a prediction subnetwork in the deepMCP network is fully utilized to perform prediction at this time, specifically, the input feature to be predicted is input to the prediction subnetwork in the deepMCP network as the first input feature to be predicted to perform operation, so as to obtain a click rate prediction result.
For example, the user characteristics of the input characteristics B include an id (such as a user account) of the user 2 and an age of the user 2, the query characteristics of the input characteristics B include a query input by the user 2, the course characteristics include a course id corresponding to a course clicked and watched by the user 2, the time characteristics include a watching duration of the user 2, and these characteristics form a first input characteristic to be predicted and are input to a prediction sub-network in the DeepMCP network for operation, so as to obtain a click rate prediction result.
And S105, if the task type to be predicted is determined to be a course and user interest correlation degree estimation task, forming a second input feature to be predicted by using part of the input features to be predicted according to a preset screening strategy, inputting the second input feature to be predicted into a matching sub-network in the deep MCP network, and calculating to obtain a course and user interest correlation degree prediction result.
In this embodiment, when it is determined that the type of the task to be predicted is a course and user interest correlation prediction task, which is different from the click rate prediction, a matching subnetwork in the deep mcp network is used, specifically, the user feature to be predicted, the query feature to be predicted, and the course feature to be predicted in the input features to be predicted are obtained according to the screening strategy, and a second input feature to be predicted, which is composed of the user feature to be predicted, the query feature to be predicted, and the course feature to be predicted in the input features to be predicted, is input to the matching subnetwork in the deep mcp network for operation, so that a prediction result of the course and user interest correlation is obtained. In this way, each sub-network in the DeepMCP network is fully utilized to realize prediction.
The embodiment of the application can acquire and process related data based on an artificial intelligence technology. Among them, Artificial Intelligence (AI) is a theory, method, technique and application system that simulates, extends and expands human Intelligence using a digital computer or a machine controlled by a digital computer, senses the environment, acquires knowledge and uses the knowledge to obtain the best result.
The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a robot technology, a biological recognition technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.
According to the method, the user and the course are represented as bipartite graphs by the DeepMCP network fusing the neural graph collaborative filtering network, the user and the course are embedded and processed by spreading the algorithm on the graphs, the high-order user and product collaborative relationship can be fully learned by using the interactive information of the user and the course, so that the matching network is strengthened, the high-order user and product collaborative relationship is transmitted for the prediction network, and the prediction result is more accurate.
The embodiment of the invention also provides a data prediction device of the fusion neural diagram collaborative filtering network, which is used for executing any embodiment of the data prediction method of the fusion neural diagram collaborative filtering network. Specifically, referring to fig. 3, fig. 3 is a schematic block diagram of a data prediction apparatus of a fusion neural graph collaborative filtering network according to an embodiment of the present invention. The data prediction apparatus 100 of the merged neural graph collaborative filtering network may be configured in a server.
The server may be an independent server, or may be a cloud server that provides basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a network service, cloud communication, a middleware service, a domain name service, a security service, a Content Delivery Network (CDN), and a big data and artificial intelligence platform.
As shown in fig. 3, the data prediction apparatus 100 for a merged neural graph collaborative filtering network includes: the system comprises an input feature set acquisition unit 101, a model training unit 102, an uploaded data receiving unit 103, a first prediction unit 104 and a second prediction unit 105.
The input feature set obtaining unit 101 is configured to obtain a stored input feature set if a model training start instruction is detected.
In this embodiment, a server is used as an execution subject to describe the technical solution. When the server detects a locally triggered model training start instruction, a locally stored input feature set needs to be acquired. The input feature set comprises a plurality of input features, and each input feature is divided into four groups, namely user features (such as user id, age and the like), query features (such as user historical behaviors), course features (such as course id) and time features.
For example, the user characteristics of the input characteristic a include id (such as a user account) of the user 1 and age of the user 1, the query characteristic of the input characteristic a includes query input by the user 1, the lesson characteristics include lesson id corresponding to the lesson clicked and watched by the user 1, and the time characteristic includes watching duration of the user 1. When the input features with the four groups of data are constructed locally, the input features can be combined into an input feature set to carry out model training on the DeepMCP network to be trained.
In the application, the difference between the deep mcp network to be trained and the existing deep mcp network is that a matching sub-network in the existing deep mcp network adopts a double-tower DSSM model, whereas in the application, the matching sub-network is realized by adopting an Ngcf network. That is, the sub-matching network in the deep mcp network to be trained in the present application is defined to adopt an Ngcf network (Ngcf network, i.e., a neural map collaborative filter network).
And the model training unit 102 is configured to train the deep mcp network to be trained through the input feature set to obtain a deep mcp network fused with the neural map collaborative filtering network.
In this embodiment, the to-be-trained DeepMCP network includes a to-be-trained prediction subnetwork, a to-be-trained matching subnetwork, and a to-be-trained association subnetwork, and the prediction subnetwork (i.e., prediction subnetwork), the matching subnetwork (i.e., matching subnetwork in which a neural graph collaborative filter network is used), and the association subnetwork (i.e., correlation subnetwork) are obtained after performing joint training based on the input feature set, respectively, so as to finally fuse the to-be-trained DeepMCP network with the neural graph collaborative filter network.
In one embodiment, the model training unit 102 comprises:
an input feature acquisition unit, configured to acquire a plurality of input features included in the input feature set; each input feature comprises a user feature, a query feature, a course feature and a time feature;
the first training unit is used for inputting all input features serving as first input features into a to-be-trained prediction sub-network of the to-be-trained DeepMCP network for training to obtain a prediction sub-network;
the second training unit is used for inputting corresponding second input features consisting of the user features, the query features and the course features of all the input features into a matching sub-network to be trained of the deep MCP network to be trained to obtain a matching sub-network; wherein the matching sub-network is a neural map collaborative filtering network;
the third training unit is used for inputting the course sequence characteristics included in the course characteristics in the input characteristics as third input characteristics into a to-be-trained associated sub-network of the to-be-trained deep MCP network for training to obtain an associated sub-network;
and the network fusion unit is used for obtaining the DeepMCP network of the fusion neural graph collaborative filtering network by fusing the prediction sub-network, the matching sub-network and the associated sub-network.
In this embodiment, the matching sub-network is used to mine user-course connections, the association sub-network is used to mine course-course connections, and the prediction sub-network is used to mine feature-CTR connections (CTR represents click-through rate). The three sub-networks are trained together in a joint training mode, and meanwhile, the three sub-networks share the embedded network, so that updating of the matching subnet and the correlation subnet can also affect the embedded network, and the effect of the prediction subnet is affected. By the combined training method, the trained DeepMCP network fused with the neural map collaborative filtering network can predict results more accurately.
In an embodiment, the data prediction apparatus 100 for fusing a neural graph collaborative filtering network further includes:
and the data uplink unit is used for uploading the model parameters of the DeepMCP network to the block chain network.
In this embodiment, the corresponding digest information is obtained based on the model parameters of the depmcp network, and specifically, the digest information is obtained by performing hash processing on the model parameters of the depmcp network, for example, by using a sha256s algorithm. Uploading summary information to the blockchain can ensure the safety and the fair transparency of the user. The user equipment may download the summary information from the blockchain to verify if xx has been tampered with. The blockchain referred to in this example is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm, and the like.
A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like.
In an embodiment, the first training unit is further configured to:
acquiring a DNN-pCTR network to be trained as a prediction subnetwork to be trained;
respectively inputting four groups of characteristics included in the ith first input characteristic into an embedded network for operation and splicing to obtain an ith Embedding vector corresponding to the ith first input characteristic; the initial value of i is 1, the value range of i is [1, N ], and N represents the total number of input features included in the input feature set;
inputting the No. i Embedding vector into a to-be-trained prediction subnetwork for operation to obtain a No. i first prediction result corresponding to the No. i first input feature;
obtaining an ith loss function value corresponding to the ith first prediction result, obtaining an ith model parameter set of the to-be-trained prediction sub-network, and adjusting the ith model parameter set of the to-be-trained prediction sub-network according to the ith loss function value to obtain an ith adjusted model parameter set;
increasing i by 1 and updating the value of i;
if the value of i does not exceed N, returning to execute the step of inputting four groups of characteristics included in the ith first input characteristic into the embedded network respectively for operation and splicing to obtain the ith Embedding vector corresponding to the ith first input characteristic;
and if the value of i exceeds N, stopping model training, and acquiring the model parameter set after the Nth adjustment as the model parameter set of the DNN-pCTR network to be trained.
In this embodiment, the prediction subnet is a conventional DNN-pCTR network, that is, four groups of features included in the input features are embedded into the network to obtain a plurality of Embedding vectors (a boosting operation needs to be performed), then the Embedding vectors are spliced to form a vector m, then the vector m is fed to four groups of features included in the subsequent input features, and data of the last layer of network is pCTR (i.e., a click-through rate prediction result). The prediction sub-network is obtained by training the prediction sub-network to be trained of the DeepMCP network to be trained, and can be used as the prediction sub-network in the final DeepMCP network.
In an embodiment, the second training unit is further configured to:
acquiring a neural diagram collaborative filtering network to be trained as a matching sub-network to be trained;
correspondingly constructing a second user-course graph of the jth by the jth second input characteristic; wherein the initial value of j is 1, the value range of j is [1, M ], and M represents the total number of input features included in the input feature set;
inputting the jth user-course bipartite graph into a neural graph collaborative filtering network to be trained for graph propagation to obtain a jth Embedding vector;
inputting the jh Embedding vector into a neural graph collaborative filtering network to be trained for prediction to obtain a jh score;
obtaining a j loss function value corresponding to the j score, obtaining a j model parameter set of a matching sub-network to be trained, and adjusting the j model parameter set of the matching sub-network to be trained according to the j loss function value to obtain a j adjusted model parameter set;
increasing j by 1 and updating the value of j;
if the value of j does not exceed M, returning to execute the step of constructing a jth user-course bipartite graph corresponding to the jth second input feature;
and if the value of j exceeds M, stopping model training, and acquiring the model parameter set after the M number adjustment as the model parameter set of the neural graph collaborative filtering network to be trained.
In this embodiment, the neural map collaborative filtering network to be trained, that is, the NGCF network to be trained, mainly includes an Embedding Layer, Embedding Propagation Layers, and a Prediction Layer. The user and the course are represented as bipartite graphs in the neural graph collaborative filtering network to be trained, and the user and the course are embeddedly transmitted on the graphs through an algorithm, so that the interaction information of the user and the course is utilized, the high-order connectivity between the user and the course is shown, and the collaborative filtering can be better carried out.
More specifically, the network architecture of the NGCF network includes an Embedding Layer, an Embedding Propagation Layer, and a Prediction Layer. Wherein, the Embeddings layer randomly initializes a trainable embedding for users and courses; the resulting embedding matrix is: e ═ Eu1,……,euN,ei1,……，eiM]. The Embedding Propagation layer, i.e. the multiple layers, is the same as the first layer, but is repeated several times, here taking one layer as an example. The propagation of the first layer is divided into two steps: message construction and message aggregation. Wherein a message is being sentFor the connected user-course pair (u, i) during construction, defining the message vector propagated by the course i to the user u as follows: m isu←i＝f(ei,eu,pui) Wherein p isuiIs the attenuation factor of the edge (u, i) at each propagation. When the messages are aggregated, the messages transmitted by all the neighbors of the user u are aggregated to update the vector representation of u:
where, (1) denotes one-layer propagation, and mu ← u denotes self-join.
The Prediction Layer gives the user u and the class i scores as:
finally, updating the network parameters by a loss function, wherein the loss function is as follows:
wherein, (u, i, j) epsilon O indicates that u and i have interaction, u and j have no interaction, sigma indicates a sigmod function, and finally
Regularization is l2 prevents overfitting. And training the NGCF network to be trained of the DeepMCP network to be trained to obtain a model parameter set of the neural map collaborative filtering network to be trained, wherein the model parameter set can be used as a matching sub-network in the final DeepMCP network.
In an embodiment, the third training unit is further configured to:
and obtaining the course sequence characteristics corresponding to the input characteristics through a skip-gram model, inputting the course sequence characteristics serving as third input characteristics into a to-be-trained associated sub-network of the to-be-trained DeepMCP network for model training, and stopping the model training until the log-likelihood function is maximized to obtain the associated sub-network.
In the present embodiment, the correlation subnet is mainly responsible for mining the association between courses and courses, and users generally consider that there is a certain correlation between courses within a certain time window for a course click sequence of a user. The idea of skip-gram is used here, and for a course click sequence, the optimization goal is to maximize the log-likelihood function, and the associated sub-network is trained quickly in the above manner.
The uploaded data receiving unit 103 is configured to receive the input feature to be predicted and the task type to be predicted, which are uploaded by the user side, if the prediction instruction is detected.
In this embodiment, when it is necessary to predict a relevant prediction result (for example, a click rate prediction result of a course) of a user based on a certain input feature, the input feature to be predicted and a task type to be predicted may be uploaded by a user side, and after receiving the input feature to be predicted and the task type to be predicted, the server further performs result prediction based on a trained DeepMCP network.
In an embodiment, the data prediction apparatus 100 for fusing a neural graph collaborative filtering network further includes:
the sub-feature obtaining unit is used for obtaining the user feature to be predicted, the query feature to be predicted, the course feature to be predicted and the time feature to be predicted which are included in the input feature to be predicted;
a first sub-feature obtaining unit, configured to use the input feature to be predicted as a first input feature to be predicted;
and the second sub-feature obtaining unit is used for obtaining the user feature to be predicted, the query feature to be predicted and the course feature to be predicted in the input features to be predicted according to the screening strategy, and the second input feature to be predicted is formed by the user feature to be predicted, the query feature to be predicted and the course feature to be predicted in the input features to be predicted.
In this embodiment, similar to the input features included in the input feature set, the input features to be predicted are also divided into four groups, that is, user features (such as user id, age, and the like), query features (such as user historical behavior), class features (such as class id), and time features, and more specifically, the input features to be predicted include the user features to be predicted, the query features to be predicted, the class features to be predicted, and the time features to be predicted. When the user characteristic to be predicted, the query characteristic to be predicted, the course characteristic to be predicted and the time characteristic to be predicted are obtained, the input characteristic to be predicted can be used as a first input characteristic to be predicted, the user characteristic to be predicted, the query characteristic to be predicted and the course characteristic to be predicted in the input characteristic to be predicted are obtained according to the screening strategy, and a second input characteristic to be predicted is formed by the user characteristic to be predicted, the query characteristic to be predicted and the course characteristic to be predicted in the input characteristic to be predicted. The sub-input features acquired in the mode are more representative and can be used as input parameters of a more accurate prediction process.
And the first prediction unit 104 is configured to, if it is determined that the type of the task to be predicted is the click rate prediction task, input the input feature to be predicted as a first input feature to be predicted into a prediction sub-network in the deepMCP network for operation, so as to obtain a click rate prediction result.
In this embodiment, when it is determined that the type of the task to be predicted is the click rate prediction task, a prediction subnetwork in the deepMCP network is fully utilized to perform prediction at this time, specifically, the input feature to be predicted is input to the prediction subnetwork in the deepMCP network as the first input feature to be predicted to perform operation, so as to obtain a click rate prediction result.
For example, the user characteristics of the input characteristics B include an id (such as a user account) of the user 2 and an age of the user 2, the query characteristics of the input characteristics B include a query input by the user 2, the course characteristics include a course id corresponding to a course clicked and watched by the user 2, the time characteristics include a watching duration of the user 2, and these characteristics form a first input characteristic to be predicted and are input to a prediction sub-network in the DeepMCP network for operation, so as to obtain a click rate prediction result.
And the second prediction unit 105 is configured to, if it is determined that the type of the task to be predicted is a course and user interest correlation degree estimation task, form a second input feature to be predicted by using a part of the input features to be predicted, which are obtained according to a preset screening strategy, and input the second input feature to be predicted into a matching sub-network in the DeepMCP network to perform operation, so as to obtain a prediction result of the course and user interest correlation degree.
In this embodiment, when it is determined that the type of the task to be predicted is a course and user interest correlation prediction task, which is different from the click rate prediction, a matching subnetwork in the deep mcp network is used, specifically, the user feature to be predicted, the query feature to be predicted, and the course feature to be predicted in the input features to be predicted are obtained according to the screening strategy, and a second input feature to be predicted, which is composed of the user feature to be predicted, the query feature to be predicted, and the course feature to be predicted in the input features to be predicted, is input to the matching subnetwork in the deep mcp network for operation, so that a prediction result of the course and user interest correlation is obtained. In this way, each sub-network in the DeepMCP network is fully utilized to realize prediction.
The device represents the user and the course as bipartite graphs by fusing the DeepMCP network of the neural graph collaborative filtering network, and carries out embedded processing on the user and the course by spreading the algorithm on the graphs, so that the collaborative relationship between the high-order user and the product can be fully learned by utilizing the interactive information of the user and the course, the matching network is strengthened, the collaborative relationship between the high-order user and the product is transmitted for the prediction network, and the prediction result is more accurate.
The data prediction apparatus of the fused neural graph collaborative filtering network may be implemented in the form of a computer program, which may be run on a computer device as shown in fig. 4.
Referring to fig. 4, fig. 4 is a schematic block diagram of a computer device according to an embodiment of the present invention. The computer device 500 is a server, and the server may be an independent server or a server cluster composed of a plurality of servers.
Referring to fig. 4, the computer device 500 includes a processor 502, memory, and a network interface 505 connected by a system bus 501, where the memory may include a storage medium 503 and an internal memory 504.
The storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032, when executed, causes the processor 502 to perform a data prediction method that fuses a neural graph collaborative filter network.
The processor 502 is used to provide computing and control capabilities that support the operation of the overall computer device 500.
The internal memory 504 provides an environment for the operation of the computer program 5032 in the storage medium 503, and when the computer program 5032 is executed by the processor 502, the processor 502 may be enabled to execute the data prediction method of the fused neural graph collaborative filtering network.
The network interface 505 is used for network communication, such as providing transmission of data information. Those skilled in the art will appreciate that the configuration shown in fig. 4 is a block diagram of only a portion of the configuration associated with aspects of the present invention and is not intended to limit the computing device 500 to which aspects of the present invention may be applied, and that a particular computing device 500 may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
The processor 502 is configured to run a computer program 5032 stored in the memory to implement the data prediction method of the fused neural graph collaborative filtering network disclosed in the embodiment of the present invention.
Those skilled in the art will appreciate that the embodiment of a computer device illustrated in fig. 4 does not constitute a limitation on the specific construction of the computer device, and that in other embodiments a computer device may include more or fewer components than those illustrated, or some components may be combined, or a different arrangement of components. For example, in some embodiments, the computer device may only include a memory and a processor, and in such embodiments, the structures and functions of the memory and the processor are consistent with those of the embodiment shown in fig. 4, and are not described herein again.
It should be understood that, in the embodiment of the present invention, the Processor 502 may be a Central Processing Unit (CPU), and the Processor 502 may also be other general purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Wherein a general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
In another embodiment of the invention, a computer-readable storage medium is provided. The computer-readable storage medium may be a nonvolatile computer-readable storage medium or a volatile computer-readable storage medium. The computer readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the data prediction method of the fusion neural graph collaborative filtering network disclosed by the embodiment of the invention.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described apparatuses, devices and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again. Those of ordinary skill in the art will appreciate that the elements and algorithm steps of the examples described in connection with the embodiments disclosed herein may be embodied in electronic hardware, computer software, or combinations of both, and that the components and steps of the examples have been described in a functional general in the foregoing description for the purpose of illustrating clearly the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
In the embodiments provided by the present invention, it should be understood that the disclosed apparatus, device and method can be implemented in other ways. For example, the above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only a logical division, and there may be other divisions when the actual implementation is performed, or units having the same function may be grouped into one unit, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may also be an electric, mechanical or other form of connection.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment of the present invention.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a storage medium. Based on such understanding, the technical solution of the present invention essentially or partially contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a magnetic disk, or an optical disk.
While the invention has been described with reference to specific embodiments, the invention is not limited thereto, and various equivalent modifications and substitutions can be easily made by those skilled in the art within the technical scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. A data prediction method fusing a neural graph collaborative filtering network is characterized by comprising the following steps:
if a model training starting instruction is detected, acquiring a stored input feature set;
training the DeepMCP network to be trained through the input feature set to obtain the DeepMCP network fusing the neural map collaborative filtering network;
if the prediction instruction is detected, receiving input features to be predicted and task types to be predicted which are uploaded by a user side;
if the task type to be predicted is determined to be the click rate prediction task, inputting the input feature to be predicted into a prediction sub-network in the DeepMCP network as a first input feature to be predicted for operation to obtain a click rate prediction result; and
and if the task type to be predicted is determined to be a course and user interest correlation degree estimation task, forming a second input feature to be predicted by partial features obtained according to a preset screening strategy in the input features to be predicted, inputting the second input feature to be predicted into a matching sub-network in the deep MCP network, and calculating to obtain a course and user interest correlation degree prediction result.
2. The method for predicting the data of the fused neural map collaborative filtering network according to claim 1, wherein the training of the deep mcpcb network to be trained through the input feature set is performed to obtain the deep mcpcb network of the fused neural map collaborative filtering network, and the method comprises the following steps:
acquiring a plurality of input features included in the input feature set; each input feature comprises a user feature, a query feature, a course feature and a time feature;
inputting each input feature as a first input feature into a to-be-trained prediction sub-network of the to-be-trained DeepMCP network for training to obtain a prediction sub-network;
inputting corresponding second input features consisting of the user features, the query features and the course features of all the input features into a matching sub-network to be trained of the deep MCP network to be trained to obtain a matching sub-network; wherein the matching sub-network is a neural map collaborative filtering network;
inputting the course sequence characteristics included in the course characteristics in the input characteristics as third input characteristics into a to-be-trained associated sub-network of the to-be-trained DeepMCP network for training to obtain an associated sub-network;
and fusing the prediction sub-network, the matching sub-network and the associated sub-network to obtain the DeepMCP network fused with the neural map collaborative filtering network.
3. The data prediction method of the fused neural graph collaborative filtering network as claimed in claim 2, wherein the prediction sub-network obtained by training the input features as the first input features input to the prediction sub-network to be trained of the deep mcp network to be trained comprises:
acquiring a DNN-pCTR network to be trained as a prediction subnetwork to be trained;
respectively inputting four groups of characteristics included in the ith first input characteristic into an embedded network for operation and splicing to obtain an ith Embedding vector corresponding to the ith first input characteristic; the initial value of i is 1, the value range of i is [1, N ], and N represents the total number of input features included in the input feature set;
inputting the No. i Embedding vector into a to-be-trained prediction subnetwork for operation to obtain a No. i first prediction result corresponding to the No. i first input feature;
obtaining an ith loss function value corresponding to the ith first prediction result, obtaining an ith model parameter set of the to-be-trained prediction sub-network, and adjusting the ith model parameter set of the to-be-trained prediction sub-network according to the ith loss function value to obtain an ith adjusted model parameter set;
increasing i by 1 and updating the value of i;
if the value of i does not exceed N, returning to execute the step of inputting four groups of characteristics included in the ith first input characteristic into the embedded network respectively for operation and splicing to obtain the ith Embedding vector corresponding to the ith first input characteristic;
and if the value of i exceeds N, stopping model training, and acquiring the model parameter set after the Nth adjustment as the model parameter set of the DNN-pCTR network to be trained.
4. The data prediction method of the fusion neural graph collaborative filtering network as claimed in claim 2, wherein the second input feature corresponding to the user feature, the query feature and the course feature of each input feature is input to a matching sub-network to be trained of a deep mcp network to be trained to obtain the matching sub-network, and the method comprises:
acquiring a neural diagram collaborative filtering network to be trained as a matching sub-network to be trained;
correspondingly constructing a second user-course graph of the jth by the jth second input characteristic; wherein the initial value of j is 1, the value range of j is [1, M ], and M represents the total number of input features included in the input feature set;
inputting the jth user-course bipartite graph into a neural graph collaborative filtering network to be trained for graph propagation to obtain a jth Embedding vector;
inputting the jh Embedding vector into a neural graph collaborative filtering network to be trained for prediction to obtain a jh score;
obtaining a j loss function value corresponding to the j score, obtaining a j model parameter set of a matching sub-network to be trained, and adjusting the j model parameter set of the matching sub-network to be trained according to the j loss function value to obtain a j adjusted model parameter set;
increasing j by 1 and updating the value of j;
if the value of j does not exceed M, returning to execute the step of constructing a jth user-course bipartite graph corresponding to the jth second input feature;
and if the value of j exceeds M, stopping model training, and acquiring the model parameter set after the M number adjustment as the model parameter set of the neural graph collaborative filtering network to be trained.
5. The method for predicting data of a fused neural graph collaborative filtering network according to claim 2, wherein the training of the to-be-trained associated sub-network of the deep mcp network is performed by inputting the course sequence features included in the course features in the input features as third input features, so as to obtain the associated sub-network, and the method comprises:
and obtaining the course sequence characteristics corresponding to the input characteristics through a skip-gram model, inputting the course sequence characteristics serving as third input characteristics into a to-be-trained associated sub-network of the to-be-trained DeepMCP network for model training, and stopping the model training until the log-likelihood function is maximized to obtain the associated sub-network.
6. The data prediction method of the merged neural graph collaborative filtering network according to claim 1, wherein after receiving the input feature to be predicted and the task type to be predicted uploaded by the user side if the prediction instruction is detected, the method further comprises:
acquiring user characteristics to be predicted, query characteristics to be predicted, course characteristics to be predicted and time characteristics to be predicted, wherein the user characteristics to be predicted, the query characteristics to be predicted, the course characteristics to be predicted and the time characteristics to be predicted are included in the input characteristics to be predicted;
taking the input feature to be predicted as a first input feature to be predicted;
and acquiring the user characteristic to be predicted, the query characteristic to be predicted and the course characteristic to be predicted in the input characteristics to be predicted according to the screening strategy, wherein the user characteristic to be predicted, the query characteristic to be predicted and the course characteristic to be predicted in the input characteristics to be predicted form a second input characteristic to be predicted.
7. The method for predicting data of a fused neural map collaborative filtering network according to claim 1, wherein after the deep mcp network to be trained is trained through the input feature set to obtain the deep mcp network of the fused neural map collaborative filtering network, the method further comprises:
and uploading the model parameters of the DeepMCP network to a block chain network.
8. A data prediction device fusing a neural graph collaborative filtering network is characterized by comprising:
the input feature set acquisition unit is used for acquiring a stored input feature set if a model training starting instruction is detected;
the model training unit is used for training the DeepMCP network to be trained through the input feature set to obtain the DeepMCP network fused with the neural map collaborative filtering network;
the uploaded data receiving unit is used for receiving the input characteristics to be predicted and the task types to be predicted uploaded by the user side if the prediction instruction is detected;
the first prediction unit is used for inputting the input features to be predicted as first input features to be predicted into a prediction sub-network in the deep MCP network for operation to obtain click rate prediction results if the type of the task to be predicted is determined to be a click rate prediction task; and
and the second prediction unit is used for forming a second input feature to be predicted by partial features obtained according to a preset screening strategy in the input features to be predicted and inputting the second input feature to be predicted into a matching sub-network in the deep MCP network for operation to obtain a prediction result of the relevance between the course and the user interest if the type of the task to be predicted is determined to be a course and user interest relevance estimation task.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the data prediction method of the fused neural graph collaborative filtering network according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program which, when executed by a processor, causes the processor to execute a data prediction method of a fused neural graph collaborative filtering network according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111013255.8A CN113688327B (en) | 2021-08-31 | 2021-08-31 | Data prediction method, device and equipment for combining neural map with filter network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111013255.8A CN113688327B (en) | 2021-08-31 | 2021-08-31 | Data prediction method, device and equipment for combining neural map with filter network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113688327A true CN113688327A (en) | 2021-11-23 |
| CN113688327B CN113688327B (en) | 2024-06-14 |
Family
ID=78584608
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111013255.8A Active CN113688327B (en) | 2021-08-31 | 2021-08-31 | Data prediction method, device and equipment for combining neural map with filter network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113688327B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020020088A1 (en) * | 2018-07-23 | 2020-01-30 | 第四范式（北京）技术有限公司 | Neural network model training method and system, and prediction method and system |
| CN111506811A (en) * | 2020-03-19 | 2020-08-07 | 上海理工大学 | Click rate prediction method based on deep residual error network |
| CN112084450A (en) * | 2020-09-09 | 2020-12-15 | 长沙理工大学 | Click rate prediction method and system based on convolutional attention network deep session sequence |
| CN112084407A (en) * | 2020-09-08 | 2020-12-15 | 辽宁工程技术大学 | Collaborative filtering recommendation method fusing graph neural network and attention mechanism |
| WO2021051515A1 (en) * | 2019-09-16 | 2021-03-25 | 平安科技（深圳）有限公司 | Vector shift-based recommendation method, apparatus, computer device, and non-volatile readable storage medium |
-
2021
- 2021-08-31 CN CN202111013255.8A patent/CN113688327B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020020088A1 (en) * | 2018-07-23 | 2020-01-30 | 第四范式（北京）技术有限公司 | Neural network model training method and system, and prediction method and system |
| WO2021051515A1 (en) * | 2019-09-16 | 2021-03-25 | 平安科技（深圳）有限公司 | Vector shift-based recommendation method, apparatus, computer device, and non-volatile readable storage medium |
| CN111506811A (en) * | 2020-03-19 | 2020-08-07 | 上海理工大学 | Click rate prediction method based on deep residual error network |
| CN112084407A (en) * | 2020-09-08 | 2020-12-15 | 辽宁工程技术大学 | Collaborative filtering recommendation method fusing graph neural network and attention mechanism |
| CN112084450A (en) * | 2020-09-09 | 2020-12-15 | 长沙理工大学 | Click rate prediction method and system based on convolutional attention network deep session sequence |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113688327B (en) | 2024-06-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11138521B2 (en) | System and method for defining and using different levels of ground truth | |
| CN108021983A (en) | Neural framework search | |
| CN111602148A (en) | Regularized neural network architecture search | |
| CN110659678B (en) | User behavior classification method, system and storage medium | |
| JP5624562B2 (en) | Method and system for calculating website visitor ratings | |
| CN110825975B (en) | Information recommendation method, device, equipment and computer readable storage medium | |
| CN110991789B (en) | Method and device for determining confidence interval, storage medium and electronic device | |
| CN107968842A (en) | News push method, apparatus and equipment based on distributed system | |
| CN112148758A (en) | Community diet health management method and system based on big data | |
| CN112085171A (en) | Recommendation method based on clustering multi-entity graph neural network | |
| CN111126578A (en) | Joint data processing method, device and system for model training | |
| CN111680798A (en) | Joint learning model system and method, apparatus, and computer-readable storage medium | |
| CN115249082A (en) | User interest prediction method, device, storage medium and electronic equipment | |
| CN111126607B (en) | Data processing method, device and system for model training | |
| CN117435999A (en) | Risk assessment method, apparatus, device and medium | |
| CN111079175B (en) | Data processing method, data processing device, computer readable storage medium and computer equipment | |
| CN117056595A (en) | Interactive project recommendation method and device and computer readable storage medium | |
| CN110347916A (en) | Cross-scenario item recommendation method, device, electronic equipment and storage medium | |
| CN113688327B (en) | Data prediction method, device and equipment for combining neural map with filter network | |
| AU2020240239A1 (en) | Conversion system, method and program | |
| CN115795156A (en) | Material recall and neural network training method, device, equipment and storage medium | |
| CN109918576B (en) | Microblog attention recommendation method based on joint probability matrix decomposition | |
| CN113988915A (en) | Method and device for positioning product passenger group, electronic equipment and storage medium | |
| CN113407312A (en) | Task cooperative processing method, device and system for model training | |
| CN112651811A (en) | Order processing method and system based on multi-source data |
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 |