CN115982467A - Multi-interest recommendation method and device for depolarized user and storage medium - Google Patents
Multi-interest recommendation method and device for depolarized user and storage medium Download PDFInfo
- Publication number
- CN115982467A CN115982467A CN202310001232.8A CN202310001232A CN115982467A CN 115982467 A CN115982467 A CN 115982467A CN 202310001232 A CN202310001232 A CN 202310001232A CN 115982467 A CN115982467 A CN 115982467A
- Authority
- CN
- China
- Prior art keywords
- user
- interest
- commodity
- data
- learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a method, a device and a storage medium for recommending multi-interest of a depolarized user, wherein the method comprises the following steps: collecting a data set of user commodity interaction records, and acquiring a training set according to the data set; storing data in a sparse matrix form, and constructing the data into graph structure data which can be processed by a graph convolution neural network; learning a node feature vector on the user-commodity relation graph; node embedding vector learning; optimizing the global relationship to obtain more potential semantic relationships; performing overcide characterization and interest prototype optimization; obtaining a multi-interest representation; depolarization assisted learning; combining the supervision loss function with the comparison loss function, and using an optimizer to perform back propagation to optimize network parameters; iteratively training the training set data until the model converges; and recommending prediction. According to the invention, through comparing the learning task with the self-adaptive sampling mode, the deviation generated by mass psychological interference when the user interest is captured is eliminated, the recommendation accuracy is improved, and the method can be widely applied to the technical field of machine learning.
Description
Technical Field
The invention relates to the technical field of machine learning, in particular to a method, a device and a storage medium for recommending multiple interests of a depolarized user.
Background
Under the background of big data, how to efficiently extract and mine useful information from massive data becomes a research topic which is concerned about, and more experts and scholars at home and abroad research the field. Processing and mining of such data is beneficial to both the user and the service provider. With the increasing development and popularization of artificial intelligence technology, more and more researchers begin to put a great deal of time and energy into the research of personalized recommendation systems, the recommendation task becomes very common and important, and how to better and more accurately recommend the personalized recommendation system is the target pursued by people.
In recent years, the application of the graph neural network in the field of recommendation systems enables the recommendation effect to be greatly improved, the data sparsity problem is relieved to a certain extent, the problem of excessive smoothness is easily caused when global information is acquired through the iterative multilayer graph neural network, node characterization tends to be consistent, and finally the acquisition of the global information is influenced. Meanwhile, many existing recommendation system algorithms are simple and general user and commodity characterization optimization, and cannot reflect multiple interests of users, but people are often influenced by multiple interests when performing certain behaviors, for example, people like a movie because of the interest of a director or a subject, different users in different real scenes can arouse different interests, and by researching the multiple interests of the users, the situation that the user is caused to generate a certain behavior due to a specific reason or a certain reason can be accurately positioned at a fine granularity, so that the recommendation accuracy and efficiency are improved. At present, a plurality of experts and scholars are put into multi-interest mining, but most of the existing multi-interest mining methods use a clustering algorithm, clustering can be performed only by means of designating interest numbers, and the multi-interest numbers of different users cannot be changed in a self-adaptive mode.
Disclosure of Invention
In order to solve at least one of the technical problems in the prior art to a certain extent, the present invention aims to provide a method, an apparatus and a storage medium for recommending multiple interests of a depolarized user.
The technical scheme adopted by the invention is as follows:
a method for recommending multi-interest of a depolarized user comprises the following steps:
collecting a data set of user commodity interaction records, and acquiring a training set according to the data set;
storing data in a sparse matrix form according to the interaction relation of a data set, and constructing graph structure data which can be processed by a graph convolution neural network;
learning a node feature vector on the user-commodity relation graph by using a graph convolution submodule;
node embedding vector learning: acquiring high-order semantic information by using a plurality of convolutional network layers, summing the feature output of each layer to obtain a commodity feature vector, and fusing user characteristics and multi-interest characteristics of a user to obtain a final embedded vector of the user;
optimizing the global relationship by using a hypergraph convolution neural network to obtain more potential semantic relationships;
and (3) performing super-edge characterization and interest prototype optimization: using multilayer hypergraph convolution network layer iteration to apply the optimized hyperedge global representation to the update of the interest prototype, and finally enabling the existing interest representation to be established on the basis of the existing interactive information and the inherent latent semantics thereof;
obtaining a multi-interest representation: calculating the similarity of the user characteristics and the interest prototype characteristics of the large interest pool, sequencing the similarity in a descending order, and selecting K interests which are most similar to the user in the interest pool as the multi-interest characteristics of the user through personalized selection of a similarity threshold value, namely selecting the most similar K interests until the similarity interval with the K +1 th interest is maximum;
depolarization assisted learning: obtaining deskew data through self-adaptive sampling, selecting data with the maximum bias to ensure self-adaptive deskew, namely forming positive and negative pair, constructing a contrast learning task, and calculating contrast loss in the contrast learning task to assist learning;
combining the supervision loss function with the comparison loss function, and performing back propagation by using an Adam optimizer to optimize network parameters; iteratively training the training set data until the model converges;
and (3) recommendation and prediction: and (4) carrying out scoring prediction on the embedded vectors of the user and the commodity finally learned by the model to obtain a recommended commodity sequence.
Further, the obtaining a training set according to the data set includes:
filtering invalid users according to the condition that the commodity interaction number of the users is less than 2, and reserving the valid users and corresponding commodity nodes; dividing the data set, randomly selecting an interaction for the verification set and the test set of each user respectively, and taking the remaining interaction items as a training set; judging the prediction result, and carrying out negative sampling on the verification set and the test set;
the processing of the de-bias data is to count the popularity redefined threshold value of each commodity, different from users, and calculate according to the proportion of the number of the interactive commodities of each user, namely if the value of the popularity is larger than the margin value, the popular commodity is identified, and if the value of the popularity is smaller than the pool value, the popular commodity is specified.
Further, the learning of the node feature vector on the user-commodity relation graph includes:
inputting the user and the commodity into a graph convolution neural network encoder together, and outputting the node characteristics in the respective relational graphs;
acquiring an interactive representation and a super edge from the updated user representation and commodity representation, inputting the interactive representation and the super edge into a hypergraph convolutional neural network, updating the interactive representation and the super edge, inputting the super edge into another hypergraph neural network together with the interest, and finally optimizing and updating the interest representation;
and adding and averaging the output of all layers and the initialized feature vector to obtain the feature vectors of the final user and the commodity.
Further, the multi-interest representation obtaining includes:
and obtaining the similarity between the user representation and the interests through cosine similarity, performing descending sorting according to the similarity, automatically selecting the most similar K interests until the similarity interval with the K +1 th interest is maximum, and obtaining the representation of the multiple interests of the user.
Further, the feature vector finally obtained by the model is used for obtaining the supervised Loss recommended and predicted by the model by using a Margin-Loss function.
Further, the depolarization aided learning includes:
and constructing a comparison learning task through the bias data and the depolarization data obtained by self-adaptive sampling, eliminating the deviation generated by mass psychological interference when capturing the user interest, introducing a supervision signal, further optimizing the characterization vectors of the user and the commodity, and obtaining the unsupervised loss of model comparison learning.
Furthermore, the target loss function of the model is the sum of the supervised loss and the unsupervised loss, the gradient descent method is used, the model parameters are continuously updated until the target loss reaches the minimum, the high-quality expression of the user and the commodity is learned, and the accurate recommendation prediction is realized.
The invention adopts another technical scheme that:
a de-polarized user multi-interest recommendation device, comprising:
at least one processor;
at least one memory for storing at least one program;
when executed by the at least one processor, cause the at least one processor to implement the method described above.
The invention adopts another technical scheme that:
a computer readable storage medium in which a processor executable program is stored, which when executed by a processor is for performing the method as described above.
The invention has the beneficial effects that: according to the invention, through comparing the learning task with the self-adaptive sampling mode, the deviation generated by the mass psychological interference when the user interest is captured is eliminated, the characterization vectors of the user and the commodity are further optimized, the preference prediction is more accurate by using the user final representation embedding vector and the commodity final representation embedding vector, and the interpretability and the robustness of the model are enhanced.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following description is made on the drawings of the embodiments of the present invention or the related technical solutions in the prior art, and it should be understood that the drawings in the following description are only for convenience and clarity of describing some embodiments in the technical solutions of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flowchart illustrating a multi-interest mining recommendation method based on a hypergraph neural network according to an embodiment of the present invention;
FIG. 2 is a block diagram of a recommendation model in an embodiment of the invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention. The step numbers in the following embodiments are provided only for convenience of illustration, the order between the steps is not limited at all, and the execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.
In the description of the present invention, it should be understood that the orientation or positional relationship referred to in the description of the orientation, such as the upper, lower, front, rear, left, right, etc., is based on the orientation or positional relationship shown in the drawings, and is only for convenience of description and simplification of description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus, should not be construed as limiting the present invention.
In the description of the present invention, the meaning of a plurality of means is one or more, the meaning of a plurality of means is two or more, and larger, smaller, larger, etc. are understood as excluding the number, and larger, smaller, inner, etc. are understood as including the number. If the first and second are described for the purpose of distinguishing technical features, they are not to be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated or implicitly indicating the precedence of the technical features indicated.
In the description of the present invention, unless otherwise specifically limited, terms such as set, installation, connection and the like should be understood in a broad sense, and those skilled in the art can reasonably determine the specific meanings of the above terms in the present invention by combining the specific contents of the technical solutions.
As shown in fig. 1 and fig. 2, the present embodiment provides a multi-interest mining recommendation method based on a hypergraph neural network, which includes firstly performing division and preprocessing of a training set, a verification set and a test set on an obtained data set, constructing a relationship graph of commodity interaction, sending the relationship graph into a graph convolutional network for encoding, then sending unbiased interaction data and biased interaction data into the hypergraph convolutional neural network respectively, constructing a comparative learning task, introducing a supervision signal, calculating to obtain unsupervised Loss, and finally performing joint training with supervised Loss (Margin-Loss) and unsupervised Loss, and performing final recommendation prediction. The method specifically comprises the following steps:
s1, constructing a data set, acquiring the data set with user personal information, interaction records and commodity category information under an Internet platform, and further filtering, dividing and coding.
User project interaction records are available in scenes such as a data collection data set, an e-commerce platform, a video website, a comment website and the like, and the data set with time information is selected when data collection is carried out considering that interest points of users are dynamically changed. The Gowalla, taobao and Yelp comment data sets which are disclosed now can be used as the candidate data sets of the invention to carry out filtering preprocessing and data set division. And converting the unique identification of the user commodity into a numerical value, and eliminating incomplete data in the mapping process.
After data collection, the method also comprises the following steps of preprocessing the data: selecting a user set and a commodity set with the user and project interaction number larger than 2, constructing an adjacent matrix form, storing the adjacent matrix form in a file in a CSR format, and acquiring depolarization data and biased data through adaptive sampling in each step training. Specifically, after data collection, users with the number of interaction less than 2 are removed, and high quality of the data is guaranteed. And then, dividing the interactive items of each user, respectively and randomly selecting one interactive record as a verification and test set, and taking the rest interactive records as a training set. And finally, in order to verify and test the model and recommend the prediction effect, taking the form of m: n of positive and negative samples as an evaluation mode, namely an all-ranking mode, for the verification set and the test set, taking all the existing interactions of each user as positive samples, and taking all the samples lacking the interactions as negative samples.
S2, learning node feature vectors, inputting the relational data stored in a graph structure mode into a user commodity interaction graph convolution network for feature learning, and outputting aggregated updated node features, wherein the graph convolution neural network mainly comprises two steps of message transmission and message aggregation, a commodity feature example is taken, the user feature is updated similarly, and the specific process is shown in the following formula:
where m represents a node characteristic, N represents a total number of goods or users, u i Indicates the ith user, v j And the j-th commodity is represented, H is the updated node characteristic, l is the l-th layer, and the multilayer convolutional neural network is related.
User and commodity characterization is initialized and input into the GCN encoder together with interaction information to optimize the characterization. Since the mass deviation of the user and the popular goods is generated in the interaction, the characterization of the interaction needs to be initialized by means of the initialization of the user and the goods. And initializing the interest prototype, optimizing the characteristics of the interest prototype per se, and optimizing the selected interest characteristics of the user better so as to enhance the final characteristics of the user. In order to better obtain the global relationship, the potential relationship between the interests in the original interest pool is represented more obviously, the super edge nodes are initialized, and the representation of the interest prototype is optimized in an auxiliary manner while the super edge nodes are optimized.
S3, learning an interest prototype feature vector, respectively initializing biased interaction features and unbiased interaction features, sending the biased interaction features and the unbiased interaction features into an interaction hypergraph convolution neural network to update a representation hypergraph of a global relationship, then sending the updated hypergraph into an interest hypergraph convolution neural network to finally obtain an updated interest prototype vector, wherein a formula 3 represents the initialization of the biased interaction features, the unbiased interaction features are similar, a formula 4 represents the detailed process of the updated hypergraph and the interest prototype in hypergraph convolution calculation, and different convolution layer numbers are determined by stacking multilayer nonlinear mapping.
Representing the biased interactive feature representation of the l-th layer, W (u) Represents a characterization of the super-edge node, δ represents the nonlinear transformation (attempting to use different activation functions such as ReLU, leakyReLU, etc.), (R) and (R) are expressed in terms of the degree of the activation function>
Representing the prototype representation of interest in the case of band bias.
And inputting the obtained interactive characteristics and the super edges of the updated user and commodity characteristics into a hypergraph convolutional neural network together, updating the interactive characteristics and the super edges, inputting the updated interaction characteristics and the super edges into another hypergraph neural network together with the interest, and finally optimizing and updating the interest characteristics:
1) And (3) two layers of iteration of the interactive hypergraph convolutional neural network are carried out, the biased interactive data and the unbiased interactive data are respectively updated to prepare for a subsequent comparison learning task, and meanwhile, the hyper-edge nodes can be updated so as to facilitate the optimization of the interest prototype.
2) The interest prototype hypergraph neural network iteratively updates the interest prototype using three-layer hypergraph convolution calculations. And the deviation generated by mass psychological interference when the user interest is captured is eliminated to a certain extent.
And S4, optimizing the multi-interest characteristics of the user, calculating the cosine similarity between the characteristics of the original user and the interest prototype characteristics of the large interest pool, and selecting the K interests which are closest to the user in the interest pool as the multi-interest characteristics of the user through personalized selection of a similarity threshold value as shown in a formula 5, namely selecting the K interests which are the most similar until the similarity interval with the K +1 th interest is the maximum.
The first K interest characteristics and the user multi-interest initial characteristics which are fused with the user and the original interest prototype characteristics are integrated and added to obtain the final multi-interest characteristics of the user, and the detailed process can refer to a formula 6,e u And e int Respectively representing a user initialization feature and an interest prototype initialization standard, and Z represents a user multi-interest feature fusing K interest prototypes.
The above processes are all subject to multiple iterations, with each feature vector's representation at the l-th level being stored for subsequent fusion, and residual results being considered in the fusion process, where Λ (u,l) Final characterization, Λ, representing the l-th layer that blends the user's multiple interests (v,l) For the product characterization, the final user and product characterization used for the score calculation is the sum of the results for each layer, ψ (u) Indicating the final embedded vector of the user, # (v) Represented as the final embedded vector for the good.
Λ (u,l) ＝Z (u,l) +H (u,l) +Λ (u,l-1) (7)
Λ (v,l) ＝H (v,l) +Λ (v,l-1) (8)
S5, because the commodity feature vector of the user can be subjected to the crowd psychology of the user caused by popular commodities in the learning optimization process, in order to eliminate deviation of the crowd psychology to the experimental result, the method and the device have the function of reducing the deviation by means of the adaptive sampling and comparison learning tasks. In the self-adaptive sampling process, the most popular data with bias and the least popular data with bias are respectively sampled, and the two batches of data are respectively sent to an interactive hypergraph neural network, so that the hyperedge representation representing the global relationship is optimized. In order to ensure adaptive depolarization, positive and negative pair are formed by the selected data with the maximum bias and the data with the minimum bias, a contrast learning task is constructed, contrast loss auxiliary learning is calculated, consistency of positive examples is maximized by using InfoNCE, greater weight is put on learning of the depolarization characteristics, consistency of negative columns is minimized, and a mathematical expression of the mathematical expression is shown as formula 10:
wherein s is a similarity function (cosine similarity or other similarity functions), the similarity between vectors is calculated, and τ is a temperature coefficient for automatically controlling the difficulty degree of the negative sample. E con Representing the optimized hypergraph in the biased interactive hypergraph convolutional neural network as a negative sample, and the same principle E int Expressed as positive samples, by separating the positive and negative samples to some extent by means of a contrast learning task, more weight is eventually replayed on the positive samples for optimization.
Wherein, the positive and negative sample pairs: the positive sample pair is an interactive characterization with a bias, and the negative sample pair is an interactive characterization with a bias, so that the distance between the positive sample pair and the negative sample pair is enlarged as much as possible. And normalizing the positive and negative sample pair characteristics, substituting the normalized positive and negative sample pair characteristics into an InfonCE expression to calculate the contrast loss of one sample, and adding the contrast losses of all samples to obtain the total contrast loss by dividing the total contrast loss by the number of the samples.
And S6, recommending and predicting. Finally, the model outputs a comprehensive user and commodity embedded vector, and a Margin-Loss function is used for obtaining the supervised Loss L of the recommended task margin The final model loss is obtained by adding the above unsupervised loss, β is a super parameter that can be adjusted to control the weight of the unsupervised loss generated by the contrast learning:
L＝L margin +β*L cl (11)
the loss minimization is the target of the neural network, the gradient of the loss to each parameter is calculated by using a gradient descent method, the gradient is reversely propagated to the network, and the model parameters are continuously updated. Inputting the updated network parameters of each training turn, user, commodity and interest prototype feature vector parameters into a network, calculating to obtain embedded vectors of the user (fusing user multi-interest characteristics) and the commodity in a testing stage, and using the prepared m: n, performing prediction scoring, and determining by the inner product of the two embedded vectors, wherein the higher the value is, the higher the preference degree of the user on the commodities is, and vice versa.
When the pre-measured time is calculated, the method and the device have the advantages that the other methods simply calculate and sort the fixed K interest characteristics of the user respectively, and the multiple characteristics originally of the user are separated to be subjected to inner product calculation with the commodity characteristics, so that the distinctiveness among the interests can be reflected, but the integrity of each user is lost, and the original characteristic expression of the user is also lost. The method and the system add the original characteristics and the multiple interest characteristics of the user together, and adaptively fuse the K multiple uncertain interests before calculating the score, thereby not only showing the distinctiveness among the user interests, but also showing the integrity of the user. The score sequence table is arranged according to descending order, the commodity corresponding to the maximum predicted score value is the commodity with the strongest user preference, the hit rate can be represented by the number of the positive examples of commodities in the first 20 scored commodities, namely one of the evaluation indexes of the invention, and meanwhile, the ranking information of the recommendation list is considered to relate to the other evaluation index, namely normalized breaking and losing accumulation gain, until the model training is completed with the best recommended prediction effect, accurate recommendation is carried out, and the commodity list which is possibly liked by the user is obtained.
In summary, compared with the prior art, the embodiment has at least the following advantages and beneficial effects:
(1) The invention starts from the fact that different users have a plurality of interest points, the unified user representation cannot reflect a plurality of interests of the users, and the number of the interests of the users is different. The existing multi-interest mining methods mostly use clustering algorithms, clustering can be performed only by means of designating interest numbers, and the multi-interest numbers of different users cannot be changed in a self-adaptive mode. Compared with the prior art, the method selects the K most similar interests of each user from a large-scale interest prototype pool until the similarity interval with K +1 is the maximum, adaptively learns the number of the multiple interests of each user, and finally fuses the multiple interest representations of each user to generate the multiple interest representations of the user.
(2) The hypergraph technology used by the invention can more accurately describe the multivariate relation, the characterization and mining capability of the nonlinear high-order relation of the hypergraph technology is stronger, more information can be mined by fully utilizing common binary relation data, the hypergraph technology can learn the relation between any two nodes, the traditional hypergraph construction is based on the prior relation between the nodes, the hypergraph used by the invention is self-adaptive, namely, the hyperedge nodes can be learned, and the global relation can be efficiently learned. Noise is introduced, robustness is increased for the model, the model is helped to learn stronger feature representation, noise of real data can be well processed, generalization performance of the model is enhanced, and an over-fitting phenomenon is prevented.
(3) The invention relates to a method for optimizing an interest prototype by utilizing a hypergraph neural network, which comprises the steps of firstly respectively updating an interactive characteristic with bias and an interactive characteristic with bias by two layers of hypergraph convolution neural networks, updating the overedge while updating the interactive characteristic, extracting the global potential relation of the existing interactive information, inputting the global potential relation and the interest prototype into another hypergraph neural network, and updating the interest characteristic. The interactive representation is obtained by optimizing the original representations of the user and the commodity on the basis of the existing interactive information by means of a user-commodity GCN encoder, and the problem of sparse data of the original interactive information can be solved to a certain extent by means of multiple layers of iteration.
(4) The method is characterized in that the depolarization self-adaptive sampling is skillfully designed, data is divided into three parts of band deviation, depolarization and fuzzy, the depolarization is enhanced and weakened by means of a contrast learning task, and a larger weight is automatically and implicitly replayed on the learning of the depolarization representation. And the data is subjected to InfonCE calculation, which is equivalent to the introduction of a supervision signal, so that deeper mutual supervision learning and better learning characteristic representation can be realized. And (4) comparing the different points of the different examples of the study focus study, zooming out the band deviation data, and zooming in the depolarization data.
The present embodiment further provides a device for recommending multiple interests of a depolarized user, including:
at least one processor;
at least one memory for storing at least one program;
when executed by the at least one processor, cause the at least one processor to implement the method described above.
The device for recommending multiple interests of a depolarized user according to the embodiment of the present invention can execute the method for recommending multiple interests of a depolarized user according to the embodiment of the present invention, can execute any combination of the method embodiments, and has corresponding functions and benefits of the method.
The embodiment of the application also discloses a computer program product or a computer program, which comprises computer instructions, and the computer instructions are stored in a computer readable storage medium. The computer instructions may be read by a processor of a computer device from a computer-readable storage medium, and executed by the processor to cause the computer device to perform the method illustrated in fig. 1.
The embodiment also provides a storage medium, which stores an instruction or a program capable of executing the method for recommending multiple interests of the depolarized user provided by the embodiment of the method of the present invention, and when the instruction or the program is run, any combination of the embodiments of the method can be executed to implement the steps, so that the method has the corresponding functions and advantages.
In alternative embodiments, the functions/acts noted in the block diagrams may occur out of the order noted in the operational illustrations. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality/acts involved. Furthermore, the embodiments presented and described in the flow charts of the present invention are provided by way of example in order to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and in which sub-operations described as part of larger operations are performed independently.
Furthermore, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the described functions and/or features may be integrated in a single physical device and/or software module, or one or more functions and/or features may be implemented in a separate physical device or software module. It will also be appreciated that a detailed discussion of the actual implementation of each module is not necessary for an understanding of the present invention. Rather, the actual implementation of the various functional modules in the apparatus disclosed herein will be understood within the ordinary skill of an engineer, given the nature, function, and internal relationship of the modules. Accordingly, those skilled in the art can, using ordinary skill, practice the invention as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are merely illustrative of and not intended to limit the scope of the invention, which is to be determined from the appended claims along with their full scope of equivalents.
The functions may be stored in a computer-readable storage medium if they are implemented in the form of software functional units and sold or used as separate products. Based on such understanding, the technical solution of the present invention or a part thereof which substantially contributes to the prior art may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, and various media capable of storing program codes.
The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.
It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
In the foregoing description of the specification, reference to the description of "one embodiment/example," "another embodiment/example," or "certain embodiments/examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (9)
1. A multi-interest recommendation method for a depolarized user is characterized by comprising the following steps:
collecting a data set of user commodity interaction records, and acquiring a training set according to the data set;
storing data in a sparse matrix form according to the interaction relation of a data set, and constructing graph structure data which can be processed by a graph convolution neural network;
learning a node feature vector on the user-commodity relation graph by using a graph convolution submodule;
node embedding vector learning: obtaining high-order semantic information by using a plurality of convolutional network layers, summing the characteristic output of each layer to obtain a commodity characteristic vector, and fusing user characteristics and multi-interest characteristics of a user to obtain a final embedded vector of the user;
optimizing the global relationship by using a hypergraph convolution neural network to obtain more potential semantic relationships;
and (3) performing super-edge characterization and interest prototype optimization: using multilayer hypergraph convolution network layer iteration to apply the optimized hyperedge global representation to the updating of the interest prototype, and finally enabling the existing interest representation to be established on the basis of the existing interactive information and the inherent potential semantics thereof;
obtaining a multi-interest representation: calculating the similarity of the user characteristics and the interest prototype characteristics of the large interest pool, sorting the similarity in a descending order, and selecting K interests which are closest to the user in the interest pool as the multi-interest characteristics of the user through personalized selection of a similarity threshold value;
depolarization assisted learning: acquiring deskew data through self-adaptive sampling, selecting data with the maximum bias for ensuring self-adaptive deskew, namely forming positive and negative pairs, constructing a comparison learning task, and calculating the comparison loss in the comparison learning task to assist in learning;
combining the supervision loss function with the comparison loss function, and performing back propagation by using an Adam optimizer to optimize network parameters; iteratively training the training set data until the model converges;
and (4) recommendation and prediction: and (4) carrying out scoring prediction on the embedded vectors of the user and the commodity finally learned by the model to obtain a recommended commodity sequence.
2. The method as claimed in claim 1, wherein the obtaining a training set according to a data set comprises:
filtering invalid users according to the condition that the commodity interaction number of the users is less than 2, and reserving the valid users and corresponding commodity nodes; dividing the data set, randomly selecting an interaction for the verification set and the test set of each user respectively, and taking the remaining interaction items as a training set; judging the prediction result, and carrying out negative sampling on the verification set and the test set;
the processing of the de-biased data consists in counting the popularity redefined threshold of each commodity, calculated as a proportion of the number of each user-interactive commodity, i.e. belonging to a deemed popular commodity if greater than the margin value, and being defined as a small commodity if less than the pool value.
3. The method according to claim 1, wherein learning the node feature vector on the user-commodity relationship graph comprises:
inputting the user and the commodity into a graph convolution neural network encoder together, and outputting the node characteristics in the respective relational graphs;
acquiring an interactive representation and a super edge from the updated user representation and commodity representation, inputting the interactive representation and the super edge into a hypergraph convolutional neural network, updating the interactive representation and the super edge, inputting the super edge into another hypergraph neural network together with the interest, and finally optimizing and updating the interest representation;
and adding and averaging the output of all layers and the initialized feature vector to obtain the feature vectors of the final user and the commodity.
4. The method according to claim 1, wherein the obtaining of the multi-interest representation comprises:
and obtaining the similarity between the user representation and the interests through cosine similarity, performing descending sorting according to the similarity, automatically selecting the most similar K interests until the similarity interval with the K +1 th interest is maximum, and obtaining the representation of the multiple interests of the user.
5. The method as claimed in claim 1, wherein the supervised Loss predicted by the model recommendation is obtained by using a Margin-Loss function for the feature vector finally obtained by the model.
6. The method of claim 1, wherein the depolarizing aids learning, comprising:
and constructing a comparison learning task through the bias data and the depolarization data obtained by self-adaptive sampling, eliminating the deviation generated by mass psychological interference when capturing the interest of the user, introducing a supervision signal, further optimizing the characterization vectors of the user and the commodity, and obtaining the unsupervised loss of model comparison learning.
7. The method as claimed in claim 1, wherein the objective loss function of the model is the sum of supervised and unsupervised losses, and the model parameters are continuously updated by using a gradient descent method until the objective loss reaches a minimum, so as to learn the high-quality expressions of the users and the commodities and realize accurate recommendation prediction.
8. A de-polarized user multi-interest recommendation device, comprising:
at least one processor;
at least one memory for storing at least one program;
when executed by the at least one processor, cause the at least one processor to implement the method of any one of claims 1-7.
9. A computer-readable storage medium, in which a program executable by a processor is stored, wherein the program executable by the processor is adapted to perform the method according to any one of claims 1 to 7 when executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310001232.8A CN115982467A (en) | 2023-01-03 | 2023-01-03 | Multi-interest recommendation method and device for depolarized user and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310001232.8A CN115982467A (en) | 2023-01-03 | 2023-01-03 | Multi-interest recommendation method and device for depolarized user and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115982467A true CN115982467A (en) | 2023-04-18 |
Family
ID=85969900
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310001232.8A Pending CN115982467A (en) | 2023-01-03 | 2023-01-03 | Multi-interest recommendation method and device for depolarized user and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115982467A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116167828A (en) * | 2023-04-25 | 2023-05-26 | 江苏亿友慧云软件股份有限公司 | Article recommendation method based on graph cooperation and contrast learning |
| CN116541716A (en) * | 2023-07-06 | 2023-08-04 | 深圳须弥云图空间科技有限公司 | Recommendation model training method and device based on sequence diagram and hypergraph |
| CN116894122A (en) * | 2023-07-06 | 2023-10-17 | 黑龙江大学 | Cross-view contrast learning group recommendation method based on hypergraph convolutional network |
-
2023
- 2023-01-03 CN CN202310001232.8A patent/CN115982467A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116167828A (en) * | 2023-04-25 | 2023-05-26 | 江苏亿友慧云软件股份有限公司 | Article recommendation method based on graph cooperation and contrast learning |
| CN116541716A (en) * | 2023-07-06 | 2023-08-04 | 深圳须弥云图空间科技有限公司 | Recommendation model training method and device based on sequence diagram and hypergraph |
| CN116894122A (en) * | 2023-07-06 | 2023-10-17 | 黑龙江大学 | Cross-view contrast learning group recommendation method based on hypergraph convolutional network |
| CN116894122B (en) * | 2023-07-06 | 2024-02-13 | 黑龙江大学 | Cross-view contrast learning group recommendation method based on hypergraph convolutional network |
| CN116541716B (en) * | 2023-07-06 | 2024-04-16 | 深圳须弥云图空间科技有限公司 | Recommendation model training method and device based on sequence diagram and hypergraph |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108920641B (en) | Information fusion personalized recommendation method | |
| CN111460130B (en) | Information recommendation method, device, equipment and readable storage medium | |
| CN109389151B (en) | Knowledge graph processing method and device based on semi-supervised embedded representation model | |
| CN115982467A (en) | Multi-interest recommendation method and device for depolarized user and storage medium | |
| WO2017193685A1 (en) | Method and device for data processing in social network | |
| CN115082142A (en) | Recommendation method, device and medium based on heterogeneous relational graph neural network | |
| CN110619081A (en) | News pushing method based on interactive graph neural network | |
| CN114519145A (en) | Sequence recommendation method for mining long-term and short-term interests of users based on graph neural network | |
| CN113918834B (en) | Graph convolution collaborative filtering recommendation method fusing social relations | |
| CN113918833A (en) | Product recommendation method realized through graph convolution collaborative filtering of social network relationship | |
| CN115878904A (en) | Intellectual property personalized recommendation method, system and medium based on deep learning | |
| CN112287170A (en) | Short video classification method and device based on multi-mode joint learning | |
| CN116401542A (en) | Multi-intention multi-behavior decoupling recommendation method and device | |
| CN113761359A (en) | Data packet recommendation method and device, electronic equipment and storage medium | |
| CN115358809A (en) | Multi-intention recommendation method and device based on graph comparison learning | |
| CN115618101A (en) | Streaming media content recommendation method and device based on negative feedback and electronic equipment | |
| CN113326384A (en) | Construction method of interpretable recommendation model based on knowledge graph | |
| CN115858919A (en) | Learning resource recommendation method and system based on project field knowledge and user comments | |
| Sridhar et al. | Content-Based Movie Recommendation System Using MBO with DBN. | |
| CN113590965B (en) | Video recommendation method integrating knowledge graph and emotion analysis | |
| CN114329051A (en) | Data information identification method, device, equipment, storage medium and program product | |
| CN112651487B (en) | Data recommendation method, system and medium based on graph collapse convolution neural network | |
| CN113590976A (en) | Recommendation method of space self-adaptive graph convolution network | |
| CN116821519A (en) | Intelligent recommendation method for system filtering and noise reduction based on graph structure | |
| CN116452293A (en) | Deep learning recommendation method and system integrating audience characteristics of articles |
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 |