Sequence recommendation method and device integrating inter-sequence and intra-sequence item correlation

Abstract

The present invention provides a sequence recommendation method and device that integrates inter-sequence and intra-sequence item correlations. The method comprises: Step 1: Using a graph neural network in the inter-sequence item correlation module to capture first-order item correlations from the user-item binary graph and the item-item co-occurrence graph, and modeling the inter-sequence item correlations; Step 2: Establishing a recurrent neural network and an attention network in the intra-sequence item correlation encoder to simulate the item correlation and duration dynamics within each individual sequence, and modeling the intra-sequence item correlations; Step 3: Using a pre-fusion approach to fuse the inter-sequence and intra-sequence item correlations to generate the final user interest; Step 4: Using the final user interest obtained in Step 3 to predict the user's preference for candidate items, calculate the ranking score of the recommended items, and complete the sequence recommendation task. The present invention can effectively improve the recommendation performance of the recommendation platform.

Description

Technical Field

[0001] The present invention relates to the field of sequence recommendation technology, and more particularly to a sequence recommendation method and device that integrates inter-sequence and intra-sequence item correlations. Background Art

[0002] Sequential recommendation involves predicting the items a user is most likely to perform an action on next based on their historical behavior. Specifically, the user's behavior record from the time they enter a recommendation service platform to the time they exit is called the user's historical behavior sequence. This historical behavior sequence is arranged in chronological order. By modeling this sequence data, we can learn about the user's interests and preferences. Generally speaking, a user's interests and preferences are determined by both their long-term and short-term interests. Long-term interests represent the user's more general interests, while short-term interests represent their real-time interests. By modeling these two sets of interest preference information and then performing a forward fusion, we can obtain the user's final interest representation. Finally, this interest representation is used to predict the items the user is likely to perform an action on next. Therefore, how to jointly model a user's long-term and short-term interests is a core issue in the sequential recommendation task. Summary of the Invention

[0003] In response to the above-mentioned problems in the prior art, embodiments of the present invention provide a sequence recommendation method and device that integrates the correlation between items within and between sequences.

[0004] In the first aspect, an embodiment of the present invention provides a sequence recommendation method that integrates inter-sequence and intra-sequence item correlations, including: Step 1: Using a graph neural network in the inter-sequence item correlation module to capture the first-order item correlation from the user-item binary graph and the item-item co-occurrence graph, and modeling the inter-sequence item correlation; Step 2: Establishing a recurrent neural network and an attention network in the intra-sequence item correlation encoder to simulate the item correlation and duration dynamics in each individual sequence, and modeling the intra-sequence item correlation; Step 3: Using a pre-fusion method to fuse the inter-sequence and intra-sequence item correlations to generate the final user interest; Step 4: Using the user's final interest obtained in step 3 to predict the user's preference for candidate items, calculate the ranking score of the items to be recommended, and complete the sequence recommendation task.

[0005] Based on the content of the above method embodiment, the sequence recommendation method for fusing inter-sequence and intra-sequence item correlations provided in the embodiment of the present invention, step 1 includes: step 1.1: learning inter-sequence item correlations by adopting user-item binary graph and item-item co-occurrence graph to obtain inter-sequence item correlation embedding vectors; step 1.2: saving the original item information by residual connection to obtain the embedding vectors of the original items; step 1.3: generating the embedding vectors of the integrated learning items by fusing the three types of item information obtained.

[0006] Based on the content of the above method embodiment, the sequence recommendation method for integrating inter-sequence and intra-sequence item correlation provided in the embodiment of the present invention, in step 1.1, the user-item bipartite graph is an undirected graph, which contains two types of nodes, namely user nodes and item nodes; the embedding vector of a node not only contains information about the node itself, but also includes graph structure information and neighbor node information. To obtain a node, the neighbor node information of the node is aggregated, including:

[0007]

[0008]

[0009] in, or For the project node i n Or user node u n The information after aggregation of k+1 layer neighbor nodes, and or and are the weight matrix and bias vector of the aggregation function at the kth layer, ρ is the pooling method of the aggregation function, σ is the activation function, and the information of the fusion node itself and the neighbor node information after aggregation include:

[0010]

[0011]

[0012] in, and or and The weight matrix and bias vector of the conversion function in the k-th layer project or user are respectively, [;] is the connection function, for an item node i n , and its hidden vector at the kth layer is obtained by aggregating the information of the k+1th layer neighbor nodes of the node.

[0013] Based on the content of the above method embodiment, the sequence recommendation method for integrating inter-sequence and intra-sequence item correlation provided in the embodiment of the present invention, in step 1.2, a residual connection component is introduced into the inter-sequence item correlation encoder module, and the residual connection method is used to save the information of the node itself, thereby preserving the original item. Its expression includes:

[0014]

[0015] in, It is project i n The original embedding vector of It is project i n The vector after the residual connection, W v ∈R d ×d and b v ∈R d are the weight matrix and bias vector of the residual connection function respectively; in step 1.3, GCN is fused through the information fusion function f(·) Bi and GCN Co The item representations learned in the two graph neural networks and the original item information representations retained in the residual connection are used to obtain the final item vector, which includes:

[0016]

[0017] in, It is project i n The items obtained after the information fusion function, the information fusion function f(·) has different meanings in different scenarios.

[0018] Based on the content of the above method embodiment, the sequence recommendation method provided in the embodiment of the present invention integrates the correlation between items within and between sequences. In step 2, the present invention learns the dependency relationship between items within the sequence through the GRU network and integrates the equality correlation between items between sequences learned by the previous module. The expression includes:

[0019]

[0020]

[0021]

[0022]

[0023] Among them, σ is the sigmoid function; The embedded vector representation of project i starting at time t after passing through the reset gate in the GRU network; is the embedding vector representation of project i starting at time t after information fusion; O h The user has an h-dimensional embedding vector representation weight matrix; The embedding vector of project i starting at time t after the update gate in the GRU network; O v is the user's v-dimensional embedding vector representation weight matrix; O is the user's embedding vector weight matrix; is the hidden vector of project i in GRU when it starts at time t; is the hidden vector of project i in GRU at time t-1; W h is the hyperparameter to be learned for the first model; W i is the hyperparameter to be learned for the second model; W is the hyperparameter to be learned for the third model; is the value calculated by the tanh function after the latent vector representation of project i at time t; tanh is the hyperbolic tangent function.

[0024] Based on the content of the above method embodiment, the sequence recommendation method for integrating the correlation between items in a sequence and within a sequence provided in the embodiment of the present invention, step 3 finally obtains the user's real-time interest preference based on step 2, and adopts the user preference vector and the original embedding vector of the item The inner product of is used as the score of the user's preference for the item, and the score is converted into the probability score of the user's behavior on the item through the softmax function. include:

[0025]

[0026] Where T is the matrix transpose operator.

[0027] Based on the content of the above method embodiment, the sequence recommendation method that integrates the correlation between items between sequences and within sequences provided in the embodiment of the present invention, in step 4, cross entropy is used as the loss function to calculate the training loss, describing the difference between the predicted value and the true value Specific forms include:

[0028]

[0029] Among them, n is the total number of samples. If the user has taken action on the item, set its label y n =1; otherwise, label y n =0.

[0030] In the second aspect, an embodiment of the present invention provides a sequence recommendation device that integrates inter-sequence and intra-sequence item correlations, including: a first main module, used to implement step 1: using a graph neural network in the inter-sequence item correlation module to capture the first-order item correlation from the user-item binary graph and the item-item co-occurrence graph, and modeling the inter-sequence item correlation; a second main module, used to implement step 2: establishing a recurrent neural network and an attention network in the intra-sequence item correlation encoder to simulate the item correlation and duration dynamics in each individual sequence, and modeling the intra-sequence item correlation; a third main module, used to implement step 3: using a pre-fusion method to fuse the inter-sequence and intra-sequence item correlations to generate the final user interest; a fourth main module, used to implement step 4: using the user's final interest obtained in step 3 to predict the user's preference for candidate items, calculate the ranking score of the items to be recommended, and complete the sequence recommendation task.

[0031] In a third aspect, an embodiment of the present invention provides an electronic device, including:

[0032] at least one processor; and

[0033] at least one memory communicatively coupled to the processor, wherein:

[0034] The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the sequence recommendation method that integrates inter-sequence and intra-sequence item correlations provided by any one of the various implementations of the first aspect.

[0035] In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions enable a computer to execute a sequence recommendation method that integrates inter-sequence and intra-sequence item correlations provided by any one of the various implementation methods of the first aspect.

[0036] The sequence recommendation method and device provided by the embodiment of the present invention integrates the correlation between items in a sequence and within a sequence. By modeling the correlation between items in a sequence and the correlation between items in a sequence, the final user interest is generated, the user's preference for candidate items is predicted, and the ranking score of the items to be recommended is calculated, which can effectively improve the recommendation performance of the recommendation platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 A flow chart of a sequence recommendation method that integrates inter-sequence and intra-sequence item correlations provided by an embodiment of the present invention;

[0039] Figure 2 A schematic diagram of the structure of a sequence recommendation device that integrates inter-sequence and intra-sequence item correlations provided by an embodiment of the present invention;

[0040] Figure 3 A schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, the technical features in the various embodiments or single embodiments provided by the present invention can be arbitrarily combined with each other to form a feasible technical solution. This combination is not restricted by the sequence of steps and / or structural composition mode, but must be based on the ability of ordinary technicians in this field to implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0042] The embodiment of the present invention provides a sequence recommendation method that integrates the correlation between items in a sequence and within a sequence. Figure 1 The method includes: step 1: using a graph neural network in the inter-sequence item correlation module to capture the first-order item correlation from the user-item binary graph and the item-item co-occurrence graph, and modeling the inter-sequence item correlation; step 2: establishing a recurrent neural network and an attention network in the intra-sequence item correlation encoder to simulate the item correlation and duration dynamics within each individual sequence, and modeling the intra-sequence item correlation; step 3: using a pre-fusion method to fuse the inter-sequence and intra-sequence item correlations to generate the final user interest; step 4: using the user's final interest obtained in step 3 to predict the user's preference for candidate items, calculate the ranking score of the items to be recommended, and complete the sequence recommendation task.

[0043] Based on the content of the above method embodiment, as an optional embodiment, the sequence recommendation method for fusing inter-sequence and intra-sequence item correlations provided in the embodiment of the present invention, step 1 includes: step 1.1: learning inter-sequence item correlations by adopting user-item binary graphs and item-item co-occurrence graphs to obtain inter-sequence item correlation embedding vectors; step 1.2: saving the original item information by residual connection to obtain the embedding vectors of the original items; step 1.3: generating the embedding vectors of the integrated learning items by fusing the three types of item information obtained.

[0044] Based on the content of the above method embodiment, as an optional embodiment, the sequence recommendation method for integrating inter-sequence and intra-sequence item correlation provided in the embodiment of the present invention, in step 1.1, the user-item bipartite graph is an undirected graph, which contains two types of nodes, namely user nodes and item nodes; the embedding vector of a node not only contains information about the node itself, but also includes graph structure information and neighbor node information. To obtain a node, the neighbor node information of the node is aggregated, including:

[0045]

[0046]

[0047] in, or For the project node i n Or user node u n The information after aggregation of k+1 layer neighbor nodes, and or and are the weight matrix and bias vector of the aggregation function at the kth layer, ρ is the pooling method of the aggregation function, σ is the activation function, and the information of the fusion node itself and the neighbor node information after aggregation include:

[0048]

[0049]

[0050] in, and or and The weight matrix and bias vector of the conversion function in the k-th layer project or user are respectively, [;] is the connection function, for an item node i n , and its hidden vector at the kth layer is obtained by aggregating the information of the k+1th layer neighbor nodes of the node.

[0051] Based on the content of the above method embodiment, as an optional embodiment, the sequence recommendation method that integrates inter-sequence and intra-sequence item correlations provided in the embodiment of the present invention, in step 1.2, introduces a residual connection component in the inter-sequence item correlation encoder module, and uses the residual connection method to save the node information itself, thereby preserving the original item. Its expression includes:

[0052]

[0053] in, It is project i n The original embedding vector of It is project i n The vector after the residual connection, W v ∈R d ×d and b v ∈R d are the weight matrix and bias vector of the residual connection function respectively; in step 1.3, GCN is fused through the information fusion function f(·) Bi and GCN Co The item representations learned in the two graph neural networks and the original item information representations retained in the residual connection are used to obtain the final item vector, which includes:

[0054]

[0055] in, It is project i n The items obtained after the information fusion function, the information fusion function f(·) has different meanings in different scenarios.

[0056] Based on the content of the above method embodiment, as an optional embodiment, the sequence recommendation method that integrates the correlation between items between sequences and within sequences provided in the embodiment of the present invention, in step 2, the present invention learns the dependency relationship between items within the sequence through the GRU network, and integrates the equality correlation between items between sequences learned by the previous sequence module, and its expression includes:

[0057]

[0058]

[0059]

[0060]

[0061] Among them, σ is the sigmoid function; The embedded vector representation of project i starting at time t after passing through the reset gate in the GRU network; is the embedding vector representation of project i starting at time t after information fusion; O h The user has an h-dimensional embedding vector representation weight matrix; The embedding vector of project i starting at time t after the update gate in the GRU network; O v is the user's v-dimensional embedding vector representation weight matrix; O is the user's embedding vector weight matrix; is the hidden vector of project i in GRU when it starts at time t; is the hidden vector of project i in GRU at time t-1; W h is the hyperparameter to be learned for the first model; W i is the hyperparameter to be learned for the second model; W is the hyperparameter to be learned for the third model; is the value calculated by the tanh function after the latent vector representation of project i at time t; tanh is the hyperbolic tangent function.

[0062] Based on the content of the above method embodiment, as an optional embodiment, the sequence recommendation method for integrating the correlation between items in a sequence and within a sequence provided in the embodiment of the present invention, step 3 finally obtains the user's real-time interest preference based on step 2, and adopts the user preference vector and the original embedding vector of the item The inner product of is used as the score of the user's preference for the item, and the score is converted into the probability score of the user's behavior on the item through the softmax function. include:

[0063]

[0064] Where T is the matrix transpose operator.

[0065] Based on the content of the above method embodiment, as an optional embodiment, the sequence recommendation method that integrates the correlation between items between sequences and within sequences provided in the embodiment of the present invention, in step 4, the cross entropy is used as the loss function to calculate the training loss, describing the difference between the predicted value and the true value Specific forms include:

[0066]

[0067] Among them, n is the total number of samples. If the user has taken action on the item, set its label y n =1; otherwise, label y n =0.

[0068] The sequence recommendation method provided by the embodiment of the present invention integrates the correlation between items in a sequence and within a sequence. By modeling the correlation between items in a sequence and the correlation between items in a sequence, the final user interest is generated, the user's preference for candidate items is predicted, and the ranking score of the items to be recommended is calculated, which can effectively improve the recommendation performance of the recommendation platform.

[0069] Furthermore, a neighborhood sampling technique is used to sample a certain number of neighboring nodes for graph convolutional network training. Specifically, based on the bipartite graph GCN Bi , the FIIS model uniformly randomly samples 10 neighbor nodes for each node. Based on the item-item co-occurrence relationship graph GCN Co ,The FIIS model adopts an importance sampling strategy, and randomly samples 10 neighbor nodes according to the edge weights.

[0070] Based on the above method, this embodiment also provides a sequence recommendation device that can automatically implement the method by fusing the equality correlation of items between sequences and within sequences. The device includes an information acquisition module, an inter-sequence item equality correlation encoder module, an intra-sequence item equality correlation encoder module, a fusion module, a prediction decoder module, an input display unit, and a control unit.

[0071] The inter-sequence item equality encoder module uses a graph neural network in the inter-sequence item equality module according to the content described in step S1 above to capture high-order item equality from the user-item binary graph and the item-item co-occurrence graph, model the inter-sequence item equality, and obtain the inter-sequence item equality embedding vector information;

[0072] The intra-sequence item equality encoder module, based on the content described in step S2 above and on the basis of step 1, establishes a recurrent neural network and an attention network in the intra-sequence item equality encoder to simulate the equality and duration dynamics of each individual sequence, model the intra-sequence item equality, and obtain the intra-sequence item equality embedding vector information;

[0073] The fusion module uses the pre-fusion method to fuse the equality correlations between and within sequences according to the content described in step S3 above, and generates the final user interest:

[0074] The prediction decoder module uses the user's final interest obtained in step S3 according to the content described in step S4 above to predict the user's preference for candidate items, calculate the ranking score of the recommended items, and complete the sequence recommendation task.

[0075] The input display unit is in communication with the information acquisition module, the inter-sequence item equality correlation encoder module, the intra-sequence item equality correlation encoder module, the fusion module, and the prediction decoder module, allowing the user to input control instructions and perform corresponding displays according to the control instructions.

[0076] The control unit is in communication with the information acquisition module, the inter-sequence item equality correlation encoder module, the intra-sequence item equality correlation encoder module, the fusion module, the prediction decoder module, and the input display unit to control their operations.

[0077] The implementation basis of each embodiment of the present invention is to implement it through programmed processing by a device with processor function. Therefore, in engineering practice, the technical solutions and functions of each embodiment of the present invention can be encapsulated into various modules. Based on this reality, on the basis of the above embodiments, an embodiment of the present invention provides a sequence recommendation device that integrates the correlation between items between sequences and within sequences. The device is used to execute the sequence recommendation method that integrates the correlation between items between sequences and within sequences in the above method embodiment. Figure 2 The device includes: a first main module, used to implement step 1: using a graph neural network in the inter-sequence item correlation module to capture the first-order item correlation from the user-item binary graph and the item-item co-occurrence graph, and modeling the inter-sequence item correlation; a second main module, used to implement step 2: establishing a recurrent neural network and an attention network in the intra-sequence item correlation encoder, simulating the item correlation and duration dynamics within each individual sequence, and modeling the intra-sequence item correlation; a third main module, used to implement step 3: using a pre-fusion method to fuse the inter-sequence and intra-sequence item correlations to generate the final user interest; a fourth main module, used to implement step 4: using the user's final interest obtained in step 3 to predict the user's preference for candidate items, calculate the ranking score of the items to be recommended, and complete the sequence recommendation task.

[0078] The sequence recommendation device provided by the embodiment of the present invention integrates the correlation between items in a sequence and within a sequence, and adopts Figure 2 Several modules in it can effectively improve the recommendation performance of the recommendation platform by modeling the correlation between items in the sequence and the correlation between items in the sequence, generating the final user interest, predicting the user's preference for candidate items, and calculating the ranking scores of the recommended items.

[0079] It should be noted that the device in the device embodiment provided by the present invention can be used to implement the method in the above-mentioned method embodiment as well as the method in other method embodiments provided by the present invention. The only difference is that the corresponding functional modules are set. The principle is basically the same as the principle of the above-mentioned device embodiment provided by the present invention. As long as those skilled in the art refer to the specific technical solutions in other method embodiments on the basis of the above-mentioned device embodiment, obtain the corresponding technical means and the technical solutions composed of these technical means by combining technical features, and ensure the practicality of the technical solutions, they can improve the device in the above-mentioned device embodiment to obtain the corresponding device class embodiment, thereby obtaining the corresponding device class embodiment for implementing the methods in other method class embodiments. For example:

[0080] Based on the content of the above-mentioned device embodiment, as an optional embodiment, the sequence recommendation device for fusing inter-sequence and intra-sequence item correlations provided in the embodiment of the present invention further includes: a first submodule, used to implement step 1 including: step 1.1: learning inter-sequence item correlations by adopting user-item binary graphs and item-item co-occurrence graphs to obtain inter-sequence item correlation embedding vectors; step 1.2: saving the original item information by residual connection to obtain the embedding vectors of the original items; step 1.3: generating the embedding vectors of the integrated learning items by fusing the three types of item information obtained.

[0081] Based on the content of the above device embodiment, as an optional embodiment, the sequence recommendation device for integrating inter-sequence and intra-sequence item correlations provided in the embodiment of the present invention further includes: a second submodule for implementing step 1.1, wherein the user-item bipartite graph is an undirected graph, which contains two types of nodes, namely user nodes and item nodes; the embedding vector of a node not only contains information about the node itself, but also includes graph structure information and neighbor node information. To obtain a node, the neighbor node information of the node is aggregated, including:

[0082]

[0083]

[0084] in, or For the project node i n Or user node u n The information after aggregation of k+1 layer neighbor nodes, and or and are the weight matrix and bias vector of the aggregation function at the kth layer, ρ is the pooling method of the aggregation function, σ is the activation function, and the information of the fusion node itself and the neighbor node information after aggregation include:

[0085]

[0086]

[0087] in, and or and The weight matrix and bias vector of the conversion function in the k-th layer project or user are respectively, [;] is the connection function, for an item node i n , and its hidden vector at the kth layer is obtained by aggregating the information of the k+1th layer neighbor nodes of the node.

[0088] Based on the content of the above device embodiment, as an optional embodiment, the sequence recommendation device for integrating inter-sequence and intra-sequence item correlations provided in the embodiment of the present invention further includes: a third submodule for implementing step 1.2, introducing a residual connection component into the inter-sequence item correlation encoder module, and using a residual connection to save the information of the node itself, thereby preserving the original item, and its expression includes:

[0089]

[0090] in, It is project i n The original embedding vector of It is project i n The vector after the residual connection, W v ∈R d ×d and b v ∈R d are the weight matrix and bias vector of the residual connection function respectively; in step 1.3, GCN is fused through the information fusion function f(·) Bi and GCN Co The item representations learned in the two graph neural networks and the original item information representations retained in the residual connection are used to obtain the final item vector, which includes:

[0091]

[0092] in, It is project i n The items obtained after the information fusion function, the information fusion function f(·) has different meanings in different scenarios.

[0093] Based on the content of the above device embodiment, as an optional embodiment, the sequence recommendation device for integrating inter-sequence and intra-sequence item correlations provided in the embodiment of the present invention further includes: a fourth submodule for implementing step 2, in which the present invention learns the item dependency within the sequence through the GRU network and integrates the inter-sequence item equality correlation learned by the previous module, and its expression includes:

[0094]

[0095]

[0096]

[0097]

[0098] Among them, σ is the sigmoid function; The embedded vector representation of project i starting at time t after passing through the reset gate in the GRU network; is the embedding vector representation of project i starting at time t after information fusion; O h The user has an h-dimensional embedding vector representation weight matrix; The embedding vector of project i starting at time t after the update gate in the GRU network; O v is the user's v-dimensional embedding vector representation weight matrix; O is the user's embedding vector weight matrix; is the hidden vector of project i in GRU when it starts at time t; is the hidden vector of project i in GRU at time t-1; W h is the hyperparameter to be learned for the first model; W i is the hyperparameter to be learned for the second model; W is the hyperparameter to be learned for the third model; is the value calculated by the tanh function after the latent vector representation of project i at time t; tanh is the hyperbolic tangent function.

[0099] Based on the content of the above device embodiment, as an optional embodiment, the sequence recommendation device for integrating inter-sequence and intra-sequence item correlation provided in the embodiment of the present invention further includes: a fifth submodule for implementing step 3 on the basis of step 2, and finally obtaining the user's real-time interest preference, using the user preference vector and the original embedding vector of the item The inner product of is used as the score of the user's preference for the item, and the score is converted into the probability score of the user's behavior on the item through the softmax function. include:

[0100]

[0101] Where T is the matrix transpose operator.

[0102] Based on the content of the above device embodiment, as an optional embodiment, the sequence recommendation device for integrating inter-sequence and intra-sequence item correlations provided in the embodiment of the present invention further includes: a sixth submodule for implementing step 4, using cross entropy as the loss function to calculate the training loss, describing the difference between the predicted value and the true value Specific forms include:

[0103]

[0104] Among them, n is the total number of samples. If the user has taken action on the item, set its label y n =1; otherwise, label y n =0.

[0105] The method of the embodiment of the present invention is implemented by electronic devices, so it is necessary to introduce the relevant electronic devices. Based on this purpose, the embodiment of the present invention provides an electronic device, such as Figure 3 As shown, the electronic device includes: at least one processor, a communications interface, at least one memory, and a communications bus, wherein the at least one processor, the communications interface, and the at least one memory communicate with each other via the communications bus. The at least one processor can call logic instructions in the at least one memory to execute all or part of the steps of the methods provided in the aforementioned method embodiments.

[0106] In addition, the logic instructions in the at least one memory mentioned above can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each method embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0107] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0108] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiment.

[0109] The flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. Based on this understanding, each box in the flowchart or block diagram can represent a module, program segment or part of the code, and the module, program segment or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, or sometimes in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0110] It should be noted that the terms "comprise," "include," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, the elements defined by the phrase "comprise..." do not preclude the presence of additional identical elements in the process, method, article, or apparatus comprising the elements.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A sequence recommendation method that integrates inter-sequence and intra-sequence item correlations, characterized in that: The method includes the following steps: Step 1: Using a graph neural network in the inter-sequence item correlation module to capture the first-order item correlation from the user-item binary graph and the item-item co-occurrence graph, and modeling the inter-sequence item correlation; Step 2: Establishing a recurrent neural network and an attention network in the intra-sequence item correlation encoder to simulate the item correlation and duration dynamics within each individual sequence, and modeling the intra-sequence item correlation; Step 3: Using a pre-fusion method to fuse the inter-sequence and intra-sequence item correlations to generate the final user interest; Step 4: Using the user's final interest obtained in step 3 to predict the user's preference for candidate items, calculate the ranking score of the recommended items, and complete the sequence recommendation task; Step 1 includes: Step 1.1: learning inter-sequence item correlation by using user-item binary graph and item-item co-occurrence graph to obtain inter-sequence item correlation embedding vector; Step 1.2: preserving original item information by residual connection to obtain original item embedding vector; Step 1.3: generating embedding vector of ensemble learning item by fusing the three types of item information; In step 1.1, the user-item bipartite graph is an undirected graph containing two types of nodes: user nodes and item nodes. The embedding vector of a node contains not only the information of the node itself, but also the graph structure information and neighbor node information. To obtain a node, the neighbor node information of the node is aggregated, including: in, or For this project node Or user node The information after aggregation of k+1 layer neighbor nodes, and or and , respectively the weight matrix and bias vector of the aggregation function at the kth layer, is the pooling method of the aggregation function, It is an activation function that integrates the node information and the aggregated neighbor node information, including: in, and or and The weight matrix and bias vector of the conversion function in the k-th layer project or user are respectively, [;] is the connection function, for an item node , its hidden vector at the kth layer is obtained by aggregating the information of the k+1th layer neighbor nodes of the node, is a user The original embedding vector of It is a project The original embedding vector of .

2. The sequence recommendation method integrating inter-sequence and intra-sequence item correlation according to claim 1, characterized in that: In step 1.2, a residual connection component is introduced into the inter-sequence item correlation encoder module. The residual connection method is used to preserve the information of the node itself, which plays the role of retaining the original item as information representation. Its expression includes: in, It is a project The original embedding vector of It is a project The vector after residual connection, and are the weight matrix and bias vector of the residual connection function respectively; in step 1.3, the bipartite graph is fused by the information fusion function f(·) and item-item co-occurrence graph The item representations learned in the two graph neural networks and the original item information representations retained in the residual connection are used to obtain the final item vector, which includes: in, It is a project The items obtained after the information fusion function, the information fusion function f(·) has different meanings in different scenarios.

3. The sequence recommendation method integrating inter-sequence and intra-sequence item correlation according to claim 2, characterized in that: In step 2, the dependency relationship between items within the sequence is learned through the GRU network, and the correlation between items between sequences learned by the previous module is integrated. Its expression includes: in, is the sigmoid function; The embedded vector representation of project i starting at time t after passing through the reset gate in the GRU network; is the embedding vector representation of project i starting at time t after information fusion; The user has an h-dimensional embedding vector representation weight matrix; The embedding vector of project i starting at time t after passing through the update gate in the GRU network; The user has a v-dimensional embedding vector representation weight matrix; is the user's embedding vector weight matrix; is the hidden vector of project i in GRU when it starts at time t; is the hidden vector of project i in GRU at time t-1; is the hyperparameter to be learned for the first model; is the hyperparameter to be learned for the second model; W is the hyperparameter to be learned for the third model; is the value calculated by the tanh function after the latent vector representation of project i at time t; tanh is the hyperbolic tangent function.

4. The sequence recommendation method integrating inter-sequence and intra-sequence item correlation according to claim 3, characterized in that: Step 3: Based on step 2, the user's real-time interest preference is finally obtained, and the user preference vector is used and the original embedding vector of the item The inner product of is used as the score of the user's preference for the item, and the score is converted into the probability score of the user's behavior on the item through the softmax function. ,include: Where T is the matrix transpose operator.

5. The sequence recommendation method integrating inter-sequence and intra-sequence item correlation according to claim 4, characterized in that: In step 4, cross entropy is used as the loss function to calculate the training loss, which describes the difference between the predicted value and the true value. Specific forms include: Among them, n is the total number of samples. If the user has taken action on the project, set its label Otherwise, the label .

6. A sequence recommendation device integrating inter-sequence and intra-sequence item correlations, based on the sequence recommendation method integrating inter-sequence and intra-sequence item correlations according to any one of claims 1 to 5, characterized in that: include: The first main module is used to implement step 1: using a graph neural network in the inter-sequence item correlation module to capture the first-order item correlation from the user-item binary graph and the item-item co-occurrence graph, and modeling the inter-sequence item correlation; the second main module is used to implement step 2: establishing a recurrent neural network and an attention network in the intra-sequence item correlation encoder to simulate the item correlation and duration dynamics within each individual sequence, and modeling the intra-sequence item correlation; the third main module is used to implement step 3: using a pre-fusion method to fuse the inter-sequence and intra-sequence item correlations to generate the final user interest; the fourth main module is used to implement step 4: using the user's final interest obtained in step 3 to predict the user's preference for candidate items, calculate the ranking score of the items to be recommended, and complete the sequence recommendation task.

7. An electronic device, characterized in that: include: At least one processor, at least one memory and a communication interface; wherein, The processor, memory and communication interface communicate with each other; The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the method according to any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium stores computer instructions, which cause the computer to execute the method of any one of claims 1 to 5.

Images