A time-aware graph neural network conversation recommendation method and system

Abstract

This invention belongs to the field of big data conversation recommendation technology, and provides a time-aware graph neural network-based conversation recommendation method and system. The method includes: acquiring a conversation sequence; constructing a first conversation graph based on the frequency of occurrence and the time interval between adjacent items in the conversation sequence, and a second conversation graph based on the time interval; learning complex item transition relationships in the first conversation graph to generate a first item representation vector; capturing multi-hop item transition relationships in the second conversation graph and generating a second item representation vector through weight selection; introducing soft attention mechanisms to distinguish the importance of different items based on the first and second item representation vectors, generating a first global embedding vector and a second global embedding vector; fusing the first and second global embedding vectors using a gated attention mechanism to learn the user's global interest preferences, obtaining a user preference representation vector, and thus obtaining a user conversation recommendation.

Description

Technical Field

[0001] This invention belongs to the field of big data conversation recommendation technology, and in particular relates to a time-aware graph neural network-based conversation recommendation method and system. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Previous research on session-based recommendation has primarily focused on capturing sequence transitions between consecutive items using recursive neural networks (RNNs) or modeling complex transitions between non-adjacent items using graph neural networks (GNNs). While these works have achieved encouraging results on session recommendation tasks, they suffer from two main problems: First, they do not consider the impact of time intervals between items on user interest. Generally, previous recommendation methods discard timestamps, retaining only the item order; these methods implicitly assume that the time intervals between all adjacent items in the sequence are the same, which is often unrealistic. Second, they do not combine temporal information with sequence patterns for joint recommendation. The degree of user interest shift is reflected not only in the frequency of adjacent items within a session but also in the length of the time interval between adjacent items—that is, the time a user lingers on a previous item while browsing the next one. Summary of the Invention

[0004] To address the technical problems existing in the background art, this invention provides a time-aware graph neural networks for session-based recommendation (T-SBR) method and system. It explores the impact of time-related information on user interests, constructs item graphs based on frequency of occurrence and item graphs based on time intervals, and learns item representations for the corresponding item graphs using a gated graph neural network (GGNN) and a graph convolutional network (GCN), respectively. Furthermore, it introduces soft attention mechanisms and gated attention mechanisms to learn the user's global interest preferences, thereby more accurately grasping the user's interests and intentions and further improving the performance of the recommendation method.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] The first aspect of the present invention provides a time-aware graph neural network-based session recommendation method.

[0007] A time-aware graph neural network-based session recommendation method includes:

[0008] Obtain the session sequence, and construct a first session graph based on the frequency of occurrence and the time interval between adjacent items in the session sequence, and a second session graph based on the time interval.

[0009] Learn the complex item transformation relationships in the first session graph and generate the first item representation vector; capture the multi-hop item transformation relationships in the second session graph and generate the second item representation vector by weight selection.

[0010] Based on the first item representation vector and the second item representation vector, a soft attention mechanism is introduced to distinguish the importance of different items, and a first global embedding vector and a second global embedding vector are generated respectively.

[0011] A gating attention mechanism is used to fuse the first global embedding vector and the second global embedding vector to learn the user's global interest preferences and obtain the user preference representation vector.

[0012] User session recommendations are obtained based on user preference representation vectors.

[0013] Furthermore, if the session sequence is a historical session sequence, the training process is also included: constructing a softmax function by performing an inner product operation using the user preference representation vector and a known set of candidate options; and training the learning parameters using the backpropagation algorithm based on the loss function calculated by the softmax function.

[0014] Furthermore, after training the learning parameters, the process also includes comparing the obtained user session recommendations with the actual user session behavior, updating the underlying data information, and optimizing the data weights of the network model.

[0015] Furthermore, before the training process, the session sequences are preprocessed, including data cleaning, missing data completion, data definition, and normalization.

[0016] Furthermore, a gated neural network is used to learn the complex item transformation relationships in the first session graph.

[0017] Furthermore, a multi-layer GCN network is used to aggregate the high-order neighborhood information of the project and capture the transformation relationship of multi-hop projects in the second session graph.

[0018] Furthermore, the process of generating the first global embedding vector and the second global embedding vector respectively includes: using a soft attention mechanism to distinguish the importance of different items and aggregating all node vectors in the first session graph to generate the first global embedding vector; using a soft attention mechanism to distinguish the importance of different items and aggregating all node vectors in the second session graph to generate the second global embedding vector.

[0019] A second aspect of the present invention provides a time-aware graph neural network-based conversation recommendation system.

[0020] A time-aware graph neural network-based conversation recommendation system includes:

[0021] The conversation graph construction module is configured to: obtain a conversation sequence, and construct a first conversation graph based on the frequency of occurrence and the time interval between adjacent items in the conversation sequence, and a second conversation graph based on the time interval.

[0022] The item representation vector acquisition module is configured to: learn complex item transformation relationships in the first session graph and generate a first item representation vector; capture multi-hop item transformation relationships in the second session graph and generate a second item representation vector by weight selection.

[0023] The global embedding vector learning module is configured to: based on the first item representation vector and the second item representation vector, introduce a soft attention mechanism to distinguish the importance of different items, and generate the first global embedding vector and the second global embedding vector respectively;

[0024] The user preference representation vector learning module is configured to: fuse the first global embedding vector and the second global embedding vector using a gated attention mechanism to learn the user's global interest preferences and obtain the user preference representation vector;

[0025] The recommendation module is configured to generate user session recommendations based on user preference representation vectors.

[0026] A third aspect of the present invention provides a computer-readable storage medium.

[0027] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the time-aware graph neural network-based session recommendation method as described in the first aspect above.

[0028] A fourth aspect of the present invention provides a computer device.

[0029] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the time-aware graph neural network-based session recommendation method as described in the first aspect above.

[0030] Compared with the prior art, the beneficial effects of the present invention are:

[0031] (1) Based on the user's historical session sequence, this invention fully considers the sequence information and time information of the session. Based on the occurrence frequency and time interval information between adjacent items, a session graph based on the occurrence frequency and a session graph based on the time interval are constructed respectively. Gated graph neural network and multi-layer GCN are used respectively to generate the item representation vector of the corresponding session graph.

[0032] (2) This invention takes into account the importance of different items to the recommendation, and introduces soft attention mechanism and gated attention mechanism to learn the user's global interest preferences, so as to more accurately grasp the user's interests and intentions and further improve the performance of the recommendation method. Attached Figure Description

[0033] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0034] Figure 1 This is an overall flowchart of the user session recommendation method provided in Embodiment 1 of the present invention;

[0035] Figure 2 The flowchart of the time-aware graph neural network-based session recommendation process provided in Embodiment 1 of the present invention is shown below.

[0036] Figure 3 This is a flowchart of a user session recommendation embodiment provided in Embodiment 1 of the present invention;

[0037] Figure 4 This is a schematic diagram of the user session recommendation system provided in Embodiment 2 of the present invention. Detailed Implementation

[0038] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0040] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0041] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems according to various embodiments of this disclosure. It should be noted that each block in a flowchart or block diagram may represent a module, segment, or portion of code, which may include one or more executable instructions for implementing the logical functions specified in the various embodiments. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutively represented blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, may be implemented using a dedicated hardware-based system that performs the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0042] Example 1

[0043] This embodiment provides a time-aware graph neural network-based session recommendation method. This embodiment uses the application of this method to a server as an example for illustration. It is understood that this method can also be applied to terminals, and can also be applied to systems including terminals, servers, and other components, and implemented through interaction between the terminal and the server. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communication, middleware services, domain name services, CDN security services, and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein. In this embodiment, the method includes the following steps:

[0044] Obtain the session sequence, and construct a first session graph based on the frequency of occurrence and the time interval between adjacent items in the session sequence, and a second session graph based on the time interval.

[0045] Learn the complex item transformation relationships in the first session graph and generate the first item representation vector; capture the multi-hop item transformation relationships in the second session graph and generate the second item representation vector by weight selection.

[0046] Based on the first item representation vector and the second item representation vector, a soft attention mechanism is introduced to distinguish the importance of different items, and a first global embedding vector and a second global embedding vector are generated respectively.

[0047] A gating attention mechanism is used to fuse the first global embedding vector and the second global embedding vector to learn the user's global interest preferences and obtain the user preference representation vector.

[0048] User session recommendations are obtained based on user preference representation vectors.

[0049] The following is a detailed description of the solution in this embodiment, such as... Figure 1 , Figure 2 and Figure 3 As shown, the session recommendation method described in this embodiment includes:

[0050] A. Perform data preprocessing on the user's historical session sequence data, including data cleaning, missing data completion, data definition and storage.

[0051] Specifically, using user purchase data from an online shopping platform as an application example, the acquired data was processed, and user data samples that were too small or too long were filtered to obtain user session data of uniform length. A detailed description of the user data is shown in Table 1.

[0052] Table 1. Basic statistical information of the dataset

[0053] Record quantity User click records 818479 Test data 351268 Training data 25898 Number of projects 40728 Average sequence length 6.69

[0054] B. Based on the historical information sequence of a certain online shopping platform, a session graph (Number-Graph) based on the occurrence frequency and time interval information between adjacent items were constructed respectively.

[0055] B1. For a Number-Graph, each session sequence S = {v1, v2, ..., v...} n} can be modeled as a session graph G b =(V b E b A b In this session graph, each node represents an item v. i Each edge (v i ,v i+1 )∈E bThis means that the user clicks on item v in session S. i Then click on project v i+1 A b Depend on and It consists of two parts, representing the weight matrices of the incoming and outgoing edges, respectively. To represent the relevance of an item to other items, considering that the item may exist in multiple session sequences, a normalized weight is assigned to each edge, calculated by dividing the number of times the edge appears in all sequences by the in-degree of the starting node of the edge;

[0056] B2. For a T-Graph, each session sequence S = {v1, v2, ..., v...} n} can be modeled as a session graph G t =(V t E t A t A t Depend on and It consists of two parts, representing the weight matrices of the incoming and outgoing edges, respectively. T-Graph further captures rich user interest information by assigning a weight primarily based on a time interval to each edge. Formally, for two consecutive items v in a session... i and v j An edge between them, edge weight The definition is as follows:

[0057]

[0058]

[0059] Where · represents the dot product operation, It is a learnable parameter, d T It's a dimension, a time effect. This shows the temporal correlation between the two timestamps;

[0060] C. Based on the constructed number-graph (primarily based on frequency of occurrence) conversation graph, a gated graph neural network is used to learn the complex item transformation relationships in the conversation graph, resulting in the item representation vector of the number-graph. The main formula is as follows:

[0061]

[0062]

[0063]

[0064]

[0065]

[0066] Where t is the timestamp, A b,i It is the i-th row in the matrix, corresponding to node v. i σ is the sigmoid function, and · is an element-wise operation. H and b are learnable parameter vectors. This is the item representation vector of session S. b,i and r b,i These represent the reset door and the update door, respectively.

[0067] D. Based on the constructed session graph with time intervals as the main component, a multi-layer GCN is used to avoid capturing the transformation relationships between multi-hop items, and a weight selection method is used to generate the item representation vector of the Time-Graph.

[0068] First, define the session S = {v1, v2, ..., v...} n Each item in} maps to an embedding sequence H = {H1, H2, ..., H}. n In this implementation, a simplified graph convolutional network (GCN) is then applied to learn item representations on the temporal graph. Since a single-layer GCN can only aggregate information about the item itself and its first-order neighbors, this implementation uses a multi-layer GCN to aggregate higher-order neighborhood information of items to capture the transition relationships between multi-hop items. To mitigate the over-smoothing problem caused by multi-layer GCNs, a weight selection method is used between the current layer representation and the initial layer representation. Formally, the information propagation process is as follows:

[0069]

[0070] α=W1(H (l) ||H (0) )

[0071] Where W1∈R d×2d Here, σ is a trainable parameter, and σ is the sigmoid activation function. After passing through the above L-layer GCN, the item representation vector is obtained.

[0072] E. Based on the Number-Graph item representation vector obtained in step C and the Time-Graph item representation vector obtained in step D, a soft attention mechanism is introduced to distinguish the importance of different items, thereby generating global embedding vectors for the corresponding session graphs respectively;

[0073] E1. Since different items in a session have varying degrees of importance to the user's interest, this implementation uses soft attention to differentiate the importance of different items and aggregates all node vectors in the N-Graph to generate the global embedding S of the session graph.n The formula is as follows:

[0074] α i =q T σ(W1v n +W2v i +b1)

[0075]

[0076] where q∈R d and W1, W2∈R d These are learnable parameters.

[0077] E2. Similarly, obtain the global session embedding representation S of the temporal session graph T-Graph. t The formula is shown below:

[0078]

[0079]

[0080] Where p∈R d and W3, W4∈R d These are learnable parameters.

[0081] F. Based on step E, obtain the global embedding vector of the conversation graph. Then, use a gated attention mechanism to fuse the two parts of the conversation representation to learn the user's global interest preferences, thereby obtaining the final user preference representation vector S. g :

[0082]

[0083] S g =g⊙S n +(1-g)⊙S t

[0084] Where W5∈R d b3 are both learnable parameters.

[0085] G. Based on the obtained user preference representation vector, a softmax function is constructed for each candidate item v. i ∈V calculate the recommendation score and recommendation probability The calculation method is shown in the formula:

[0086]

[0087]

[0088] in This represents the probability that an item will appear in the next interaction.

[0089] H. Using cross-entropy as the loss function, if y represents the true class distribution, the loss function is defined as follows:

[0090]

[0091] Then, the stochastic gradient descent optimizer is used for optimization.

[0092] User session recommendations were performed on the test samples, the recommendations were pushed out, and the results were compared with actual user session behavior. Table 2 describes the performance comparison of methods in user session recommendation:

[0093] Table 2 Performance Comparison

[0094]

[0095] Based on the results in Table 2, the user session recommendation model proposed in this embodiment outperforms other methods.

[0096] Example 2

[0097] This embodiment provides a time-aware graph neural network-based conversation recommendation system.

[0098] like Figure 4 As shown, a time-aware graph neural network-based conversation recommendation system includes:

[0099] The conversation graph construction module is configured to: obtain a conversation sequence, and construct a first conversation graph based on the frequency of occurrence and the time interval between adjacent items in the conversation sequence, and a second conversation graph based on the time interval.

[0100] The item representation vector acquisition module is configured to: learn complex item transformation relationships in the first session graph and generate a first item representation vector; capture multi-hop item transformation relationships in the second session graph and generate a second item representation vector by weight selection.

[0101] The global embedding vector learning module is configured to: based on the first item representation vector and the second item representation vector, introduce a soft attention mechanism to distinguish the importance of different items, and generate the first global embedding vector and the second global embedding vector respectively;

[0102] The user preference representation vector learning module is configured to: fuse the first global embedding vector and the second global embedding vector using a gated attention mechanism to learn the user's global interest preferences and obtain the user preference representation vector;

[0103] The recommendation module is configured to generate user session recommendations based on user preference representation vectors.

[0104] It should be noted that the conversation graph construction module, item representation vector acquisition module, global embedding vector learning module, user preference representation vector learning module, and recommendation module described above are the same examples and application scenarios implemented in the steps of Embodiment 1, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0105] Example 3

[0106] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the time-aware graph neural network-based session recommendation method as described in Embodiment 1 above.

[0107] Example 4

[0108] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the time-aware graph neural network-based session recommendation method as described in Embodiment 1 above.

[0109] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0110] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0111] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0112] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0113] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0114] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A time-aware graph neural network-based session recommendation method, characterized in that, include: Obtain the session sequence, and construct a first session graph based on the frequency of occurrence and the time interval between adjacent items in the session sequence, and a second session graph based on the time interval. The session graph based primarily on frequency of occurrence is called a Number-Graph; each session sequence It can be modeled as a session graph ; In this session diagram, each node represents a project. Each edge This means that the user is in the session S Click on the project Then click on the project ; Depend on and It consists of two parts, representing the weight matrix of the incoming and outgoing edges respectively; in order to represent the relevance of an item to other items, considering that the item may exist in multiple session sequences, a normalized weight is assigned to each edge, which is calculated by dividing the number of times the edge appears in all sequences by the in-degree of the starting node of the edge. The second session graph, which is primarily based on time intervals, is called a Time-Graph; each session sequence It can be modeled as a session graph ; Depend on and It consists of two parts, representing the weight matrices of the incoming and outgoing edges, respectively; the second session graph, which is mainly based on time intervals, further captures rich user interest information by assigning a weight mainly based on time intervals to each edge; formally, for two consecutive items in the session... and An edge between them, edge weight The definition is as follows: in, This represents the dot product operation. These are learnable parameters. It's a dimension, a time effect. This shows the temporal correlation between the two timestamps; Learn the complex item transformation relationships in the first session graph and generate the first item representation vector; capture the multi-hop item transformation relationships in the second session graph and generate the second item representation vector by weight selection. Based on the first item representation vector and the second item representation vector, soft attention mechanisms are introduced to distinguish the importance of different items, and first global embedding vector and second global embedding vector are generated respectively. Based on the obtained Number-Graph item representation vectors and Time-Graph item representation vectors, a soft attention mechanism is introduced to distinguish the importance of different items, thereby generating global embedding vectors for the corresponding session graphs. Since different items in a session have varying degrees of importance to the user's interest, this implementation uses soft attention to differentiate the importance of different items and aggregates all node vectors in the N-Graph to generate a global embedding of the session graph. The formula is as follows: in and These are learnable parameters; Similarly, obtain the global session embedding representation of the temporal session graph T-Graph. The formula is shown below: in and These are learnable parameters; A gating attention mechanism is used to fuse the first global embedding vector and the second global embedding vector to learn the user's global interest preferences and obtain the user preference representation vector. User session recommendations are obtained based on user preference representation vectors.

2. The time-aware graph neural network-based session recommendation method according to claim 1, characterized in that, If the session sequence is a historical session sequence, the training process is also included: constructing a softmax function by performing an inner product operation on the user preference representation vector and the known candidate option set; and training the learning parameters using the backpropagation algorithm based on the loss function calculated by the softmax function.

3. The time-aware graph neural network-based session recommendation method according to claim 2, characterized in that, After training the learning parameters, the process also includes comparing the obtained user session recommendations with the actual user session behavior, updating the underlying data information, and optimizing the data weights of the network model.

4. The time-aware graph neural network-based session recommendation method according to claim 2, characterized in that, Before training, the session sequences are preprocessed, including data cleaning, missing data completion, data definition, and normalization.

5. The time-aware graph neural network-based session recommendation method according to claim 1, characterized in that, A gated neural network is used to learn the complex item transformation relationships in the first session graph.

6. The time-aware graph neural network-based session recommendation method according to claim 1, characterized in that, A multi-layer GCN network is used to aggregate high-order neighborhood information of projects and capture the transformation relationship of multi-hop projects in the second session graph.

7. The time-aware graph neural network-based session recommendation method according to claim 1, characterized in that, The process of generating the first global embedding vector and the second global embedding vector includes: using a soft attention mechanism to distinguish the importance of different items and aggregating all node vectors in the first session graph to generate the first global embedding vector; using a soft attention mechanism to distinguish the importance of different items and aggregating all node vectors in the second session graph to generate the second global embedding vector.

8. A time-aware graph neural network-based conversation recommendation system, characterized in that, The time-aware graph neural network-based session recommendation method according to any one of claims 1-7 includes: The conversation graph construction module is configured to: obtain a conversation sequence, and construct a first conversation graph based on the frequency of occurrence and the time interval between adjacent items in the conversation sequence, and a second conversation graph based on the time interval. The item representation vector acquisition module is configured to: learn complex item transformation relationships in the first session graph and generate a first item representation vector; capture multi-hop item transformation relationships in the second session graph and generate a second item representation vector by weight selection. The global embedding vector learning module is configured to: based on the first item representation vector and the second item representation vector, introduce a soft attention mechanism to distinguish the importance of different items, and generate the first global embedding vector and the second global embedding vector respectively; The user preference representation vector learning module is configured to: fuse the first global embedding vector and the second global embedding vector using a gated attention mechanism to learn the user's global interest preferences and obtain the user preference representation vector; The recommendation module is configured to generate user session recommendations based on user preference representation vectors.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the time-aware graph neural network-based session recommendation method as described in any one of claims 1-7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the time-aware graph neural network-based session recommendation method as described in any one of claims 1-7.

Images