Project recommendation method based on multi-view comparative learning

Abstract

The invention provides a project recommendation method based on multi-view comparative learning, which comprises the following steps: in a segmentation clustering module, segmenting a user interaction project sequence into a plurality of cross subsequences in a sliding manner, obtaining initial coarse-grained intention embedding based on the cross subsequences, and clustering the initial coarse-grained intention embedding to obtain fine-grained intention embedding; in the fusion extraction module, a user cluster interest graph and a project cluster interest graph which can express the user intention are constructed, same-cluster interaction is carried out in the user cluster interest graph, different-cluster interaction is carried out on the user cluster interest graph and the project cluster interest graph, and user intention embedding in the global level is obtained; and in the comparison recommendation module, obtaining final coarse-grained intention embedding from the preliminary coarse-grained intention embedding through a layer attention network, and performing a recommendation task based on the final coarse-grained intention embedding to obtain a project recommendation list. The problems of insufficient embedding diversity and low recommendation accuracy in the prior art can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an item recommendation method based on multi-view contrast learning. Background Art

[0002] As one of the important tools for obtaining information in the era of big data, project recommendation systems have received extensive attention and research. Among them, sequential recommendation aims to analyze the project sequence characteristics of users and provide them with personalized project recommendations.

[0003] Current sequential recommendation models usually model users' immediate intentions to capture their potential needs, but most intent-based sequential recommendation models ignore the item interaction sequence information of neighboring users, which makes the item sequence embedding lack important neighbor intent information, resulting in insufficient embedding diversity. Moreover, existing intent-based recommendation models mostly do not make full use of the subsequence information within the item interaction sequence, which affects the accuracy of recommendations. Summary of the invention

[0004] To this end, an embodiment of the present invention proposes an item recommendation method based on multi-view contrastive learning to solve the problems of insufficient embedding diversity and low recommendation accuracy in the prior art.

[0005] According to an embodiment of the present invention, a project recommendation method based on multi-view contrastive learning is applied to a project recommendation system, wherein the project recommendation system includes a segmentation and clustering module, a fusion extraction module, and a contrast recommendation module; The method comprises: In the segmentation and clustering module, the user interaction item sequence is obtained, and the user interaction item sequence is slidingly segmented into multiple cross subsequences. A preliminary coarse-grained intent embedding is obtained based on the cross subsequences, and the preliminary coarse-grained intent embedding is clustered to obtain a fine-grained intent embedding. In the fusion extraction module, the user interaction project sequence is fused with the time interval information and location information to construct the user cluster interest graph and the project cluster interest graph that can express the user's intention. The same cluster interaction is performed within the user cluster interest graph, and the different cluster interaction is performed on the user cluster interest graph and the project cluster interest graph, so as to obtain the user intention embedding at the global level. In the comparative recommendation module, the preliminary coarse-grained intent embedding is used to obtain the final coarse-grained intent embedding through the layer attention network. The recommendation task is performed based on the final coarse-grained intent embedding to obtain a project recommendation list. The final coarse-grained intent embedding, fine-grained intent embedding, and global user intent embedding are subjected to multi-view comparative learning to optimize the model parameters of the comparative recommendation module.

[0006] The project recommendation method based on multi-view contrastive learning according to the embodiment of the present invention has the following beneficial effects: (1) To address the problem of lack of important neighbor intent information in existing intent-based recommendation models, the present invention considers combining time interval information with location information to construct user cluster interest graphs and project cluster interest graphs that can express user intent, performs intra-cluster interaction within the user cluster interest graph, and performs hetero-cluster interaction between the user cluster interest graph and the project cluster interest graph. By extracting user intent embedding at the global level through the above information fusion interaction, the diversity of embedding can be effectively enriched, which helps to improve model performance. (2) To address the problem that existing intent-based recommendation models do not fully utilize subsequence information within item interaction sequences, the present invention uses a layered attention network to dynamically process subsequence embeddings at different time periods to capture dynamic changes in user intent, which can further improve model performance. (3) Combining multi-view self-contrastive learning to fuse information from different views can maximize the performance improvement of the model. Since comprehensive modeling of user intent requires consideration of many factors, how to fuse information from multiple aspects becomes a key issue. After obtaining different information from multiple views, the present invention performs multi-view contrastive learning on the final coarse-grained intent embedding, fine-grained intent embedding, and global user intent embedding. It can capture the similarities and differences of sample pairs while alleviating data sparsity, and learn more discriminative features, thereby improving the generalization ability of the model and improving the accuracy of recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The above and / or additional aspects and advantages of the embodiments of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which: Figure 1 is a schematic diagram of the structure of a project recommendation system according to an embodiment of the present invention; Figure 2 4 is a flowchart of an item recommendation method based on multi-view contrastive learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0008] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 creative work are within the scope of protection of the present invention.

[0009] The embodiment of the present invention proposes a project recommendation method based on multi-view contrast learning, which is applied to the project recommendation system. Figure 1 , the project recommendation system includes a segmentation and clustering module, a fusion extraction module and a comparison recommendation module.

[0010] See also Figure 2 , the method comprises steps S1 to S3: S1, in the segmentation and clustering module, obtain the user interaction item sequence, slide-segment the user interaction item sequence into multiple cross subsequences, obtain preliminary coarse-grained intent embedding based on the cross subsequences, and cluster the preliminary coarse-grained intent embedding to obtain fine-grained intent embedding.

[0011] The user's interactive behavior mainly depends on his intention, and different intentions are often hidden in their historical interactive item sequences. The present invention slide-splits the interactive item sequence into multiple cross subsequences and clusters them as units to obtain the user's intentions at different levels in the sequence.

[0012] Different subsequences with the same target item (i.e., the last item of the sequence is the same) may represent the same intention. To study the subsequence information of the same target item in the user interaction item sequence, we use sliding splitting DS The operation splits the original sequence into multiple crossover subsequences.

[0013] After performing a sliding split operation on all user sequences, all obtained subsequences are placed into the target item set to construct a coarse-grained self-supervisory signal.

[0014] Specifically, in the segmentation and clustering module, the user interaction item sequence is obtained s u , will be split by sliding DS Actions sequence user interaction items s u Split into multiple cross subsequences, and obtain preliminary coarse-grained intent embedding based on the cross subsequences. The expression is as follows: ; in, For the r The initial coarse-grained intent embedding corresponding to each item, express s u The target item is The target item is the last item in the sequence. For the encoder, It has the characteristics of efficiently processing item sequence information, adapting to different data densities, and capturing rich semantic information.

[0015] The initial coarse-grained intent embedding of the user's intent has been obtained through the subsequences of the same target item. However, in actual scenarios, the user intent represented by the same target item may belong to different intent categories in different interaction sequences. Therefore, it is necessary to distinguish the user's interaction item sequence more finely to deeply explore the real user intent hidden in the item sequence. Clustering can classify them into several categories based on the similarities and differences between embeddings, thereby distinguishing the user's real intent represented by the user's interaction item sequence.

[0016] In the segmentation and clustering module, K-means clustering is performed on the preliminary coarse-grained intent embedding to obtain the clustering results. F , ,in, f h Indicates h Cluster centers, H Represents the total number of types of user intent, embeds the obtained cluster centers as fine-grained intents, and defines the distribution function Q ( f h ),like s u Belong to h clusters, then Q ( f h )=1, indicating that the user's potential intention is f h ;like s u Not in h clusters, then Q ( f h )=0, indicating that the user's potential intention is not f h .

[0017] S2, in the fusion extraction module, the time interval information and location information of the user interaction project sequence are fused to construct a user cluster interest graph and a project cluster interest graph that can express the user's intention. The same cluster interaction is performed within the user cluster interest graph, and the different cluster interaction is performed on the user cluster interest graph and the project cluster interest graph to obtain the user intention embedding at the global level.

[0018] Among them, in sequential recommendation, by analyzing the behavior and preferences of neighbor users, the potential intentions of target users can be more accurately inferred, and more personalized content can be recommended accordingly. The present invention combines time interval information to construct user cluster and project cluster interest graphs to capture dynamic neighbor user intention information, and then fuses information within users (same cluster) and between users and projects (different clusters) to ensure effective processing of neighbor information, provide global comparative learning signals for the model, and make user intention embedding more diverse.

[0019] In the process of studying user intentions, the information of neighbor users associated with the target user is extremely valuable. However, the association information between these neighbor users and the target user is usually scattered, which undoubtedly increases the difficulty of mining key information between user interaction project sequences. In order to more effectively extract effective information between sequences, by connecting specific users or projects in the form of clusters in the interest graph, the originally scattered interaction sequence is transformed into a more compact cluster interest graph. In addition, the user's intention is not consistent and often changes over time. Therefore, the present invention combines the time interval information to establish a self-attention mechanism, so as to distinguish the user's intentions in different time periods based on the interest graph.

[0020] Specifically, in the fusion extraction module, an undirected graph is used G c represents all user interactions with items, G c ={ V c , E c},in, V c is a node set, E c is the edge set, when the node set V c Nodes in v c When the subject is a user, the node v c Represents user embedding; when the node set V c Nodes in v c When the subject is a project, the node v c Represents item embedding. By observing the user's historical interaction item sequence, it can be found that multiple related items will frequently appear together at different time points. These items are generally highly correlated with the target item and contribute highly to the current intention pointed to by the target item, while those items with low correlation with the target item contribute relatively little to the current intention. In view of this, the present invention uses time interval and location information to model user intentions and constructs a self-attention mechanism to distinguish the diverse intentions of users in this way.

[0021] Specifically, the user interaction item sequence is fused with the time interval information and the position information to obtain the k Project embedding for projects e k , the expression is: ; Among them, tanh is the hyperbolic tangent activation function, is the weight matrix in the fusion extraction module, For an undirected graph G c The initial embedding of the items obtained, is the bias vector in the fusion extraction module, For the k The position embedding vector of each position, For the k The time interval characteristics of each location; The item embeddings of all items form the item representation matrix e ; According to the direct relevance of the items, the embedding of each item is updated to obtain the following formula: ; ; ; in, represents the updated item representation matrix, e i Indicates i Project embedding for each project, e j Indicates j Project embedding for each project, Indicates i Project and j The correlation between the items is calculated. W Q and W K denote the input projections for query and key respectively, T represents the transpose operation, represents the scale factor, Indicates i Project and k The correlation between the items is calculated. K Indicates the total number of items, represents the normalized exponential function, Indicates updated , n Indicates the number of items associated with the target item, W V An input projection representing the model value; Based on the updated item representation matrix Construct user cluster interest graphs and item cluster interest graphs that can express user intentions.

[0022] In this way, by integrating time interval and location factors to model user dynamic intentions, calculating the correlation between items and distinguishing different intentions in the user interaction item sequence, the user intention information that changes over time is effectively captured. A user cluster interest graph with users as the main body is constructed, from which the user adjacency matrix and user embedding can be obtained. Similarly, a project cluster interest graph can be constructed with projects as the main body, thereby obtaining the project adjacency matrix and project embedding.

[0023] The interest graph integrated with user intent greatly enriches the project embedding, but the user embedding is relatively simple and does not contain relevant information about the project. Therefore, the present invention takes users and projects as the center respectively, and performs same-cluster interaction and different-cluster interaction between the user cluster interest graph obtained by integrating user intent and the project cluster interest graph.

[0024] In order to make each node in the user cluster interest graph fully interact with each other without increasing the complexity of the model too much, the node embedding is input into the Multilayer Perceptron (MLP) to obtain the interaction results between the nodes.

[0025] In the fusion extraction module, the same cluster interaction is performed in the user cluster interest graph, and the expression is as follows: ; in, z k is the node interaction value after aggregation, is the MLP function, Represents nodes in the user cluster interest graph k The embedding Represents nodes in the user cluster interest graph j Embedded, N k A node-set representing an item.

[0026] MLP can explicitly model the interaction information between user nodes and effectively extract the neighbor information in the graph structure. However, the interaction between two user nodes in the user cluster interest graph does not necessarily mean that the two nodes are similar, because the intentions of interacting with the same item in different situations are different. Therefore, the characteristics of the user interaction items must be considered when modeling user intentions. Next, let the user cluster interest graph and the item cluster interest graph complete the information interaction.

[0027] The project node embedding has completed information interaction with the associated nodes, and the node contains information of other associated nodes. In order to integrate the features in the project node into the user node, the node matching is implemented using a cross-interaction method between the user cluster interest graph and the project cluster interest graph. Similar to the same-cluster message passing, the Hadamard product is used to aggregate one node of the graph with all nodes in the other graph. This can increase the correlation of similar features between the user node and the project node, and reduce the correlation of dissimilar features. Therefore, if the user node attribute has a higher matching score on the project node attribute, it is determined that they have similar feature information.

[0028] Specifically, the user cluster interest graph and the project cluster interest graph are subjected to heterogeneous cluster interaction, and the expression is as follows: ; in, is the relevant value of node features after aggregation, represents the Hadamard product, Represents the updated item representation matrix Project Node b Embedded, A set of nodes representing the interest graph of item clusters.

[0029] In order to make the user node integrate the message transmission results and node matching results, the gated recurrent unit (GRU) is used to aggregate the above interaction results to the user's initial node. Specifically, the following formula is used to obtain the global level user intention embedding ; ; ; ; ; in, To reset the gate neuron, is the initial representation of the user node, is the input gate neural unit, , , is the weight matrix of the gated recurrent unit, , , is the bias vector of the gated recurrent unit, is the updated embedding representation of the user node, and concat is the linking operation.

[0030] S3, in the comparative recommendation module, the preliminary coarse-grained intent embedding is used to obtain the final coarse-grained intent embedding through the layer attention network, and the recommendation task is performed based on the final coarse-grained intent embedding to obtain the project recommendation list. The final coarse-grained intent embedding, fine-grained intent embedding, and global user intent embedding are subjected to multi-view comparative learning to optimize the model parameters of the comparative recommendation module.

[0031] Among them, when exploring user intentions, a contrastive learning task framework is constructed from coarse-grained, fine-grained and global levels. The representation ability of the model can be improved by comparing contrast signals from different perspectives. It can not only make full use of the subsequences and cluster center embeddings of the same target items in the project sequence, but also integrate neighbor user embeddings, thereby enhancing the model's understanding of user intentions and improving recommendation accuracy.

[0032] At the coarse-grained level, by using sequence embedding after layer-by-layer attention networks and comprehensively considering the changing information of user intentions hidden in subsequences, coarse-grained contrastive learning is performed on samples, which helps to grasp the user's initial intentions more comprehensively. At the fine-grained level, through fine-grained contrastive learning, it can be found that users' interactive behaviors under the same target item will show different intentions due to different contexts. At the global level, contrastive learning integrates neighbor user information to discover information that has significant and multi-faceted impacts on target users, and more comprehensively models the user's intention distribution.

[0033] In real business scenarios, user intentions are not static, but change over time. In order to initially grasp user intentions, the cross subsequences encoded after the sliding split of the user interaction item sequence have been obtained in step S1, but using only the cross subsequences as samples for comparative learning and the subject of performing the recommendation task is not enough to represent the user's comprehensive intentions. This is because after the cross subsequences are arranged in chronological order, the subsequences in different time periods represent the dynamic changes in user interests in different periods, and only using the embedding of a certain period for operation will lose the dynamic change information of user intentions.

[0034] The layer attention network can effectively capture the dynamics of user intentions, and model the intentions in detail through a layered structure, and can adaptively assign weights according to the user's behavior at different time periods, so that the model can focus on the items that the user is most likely to be interested in at a certain time period, thereby grasping the changing information of user intentions and improving the accuracy of recommendations. Therefore, the present invention uses a layer attention network to process the user's dynamic intentions. Specifically, in the comparison recommendation module, the preliminary coarse-grained intention embedding is obtained through the layer attention network to obtain the final coarse-grained intention embedding, which satisfies the following conditional formula: ; in, For the rThe final coarse-grained intent embedding corresponding to each item.

[0035] In sequential recommendation, the prediction of the next item can be regarded as a classification task based on all items. Therefore, the final coarse-grained intent embedding is used to calculate the probability of the next item that the user interacts with. Specifically, the recommendation task is performed based on the final coarse-grained intent embedding to obtain a list of recommended items that satisfies the following conditional formula: ; in, represents the predicted score of the item, and M represents the item embedding matrix.

[0036] Loss function for recommendation tasks for: ; in, represents the correct prediction result of the sequence, Represents the incorrect prediction results of the sequence.

[0037] Since different subsequences with the same target item may represent the same intent, we hope that the intent representations of two subsequences with the same target item are closer in the latent space. Specifically, we use a contrastive learning framework to avoid the false negative sample problem. Suppose there is a sequence s 1. First, randomly select a sequence s 1Sequences with the same target item s 2. Calculate the coarse-grained intent embeddings of the two sequences respectively, and use contrastive learning to make the two coarse-grained intent embeddings close to each other in the latent space. Specifically, the coarse-grained contrast loss function is for: ; in, For sequence s 1 coarse-grained embedding, For random sampling and sequence s 1Sequences with the same target item s 2 coarse-grained embedding, is the contrast loss function.

[0038] In actual scenarios, users' interactive behaviors under the same target item will show different intentions due to different contexts. For example, the intention of male users to buy razors is a daily need, while the intention of female users to buy razors is to give them as gifts. It can be seen that using only coarse-grained contrastive learning is not enough for the model to recognize the differences in user interactive behaviors in different scenarios. It is also necessary to identify the user's true intentions in a more fine-grained way in some way. After performing a clustering operation on the subsequence embedding, the present invention obtains fine-grained intention embedding and uses it as a fine-grained contrastive learning signal for contrastive learning. In this way, the model can further distinguish the true intentions of subsequences of the same target item by using similar attributes embedded in the cluster.

[0039] pass query The operation obtains the fine-grained intent embedding of the two subsequences. Specifically, the expression is as follows: ; ; in, f c1 Indicates distance The nearest cluster center, f c2 Indicates distance The nearest cluster center, query Represents a query operation.

[0040] To avoid the problem of introducing false negative samples, a fine-grained contrast loss function is designed : .

[0041] At the global level, in order to utilize the neighbor user information that has a significant impact on the target user, it is necessary not only to accurately capture and understand the user's intention, but also to effectively process the neighbor information to enhance the diversity of the fused information in the user intention embedding. Through contrastive learning, the model can learn the similarities and differences with the neighbor user intentions in the feature space, which helps to more accurately identify the target user's true intention. In this way, combining contrastive learning and supervised learning tasks can further improve the recommendation performance while learning richer user and item embeddings.

[0042] According to step S2, the user cluster interest graph and the item cluster interest graph are constructed and then information fusion interaction is implemented to obtain a global embedding with time information. Next, this is used as a global contrast learning signal for contrast learning, so that the model can effectively use neighbor information to make accurate judgments on the user's true intentions. Similarly, in order to avoid the introduction of false negative samples, a global contrast loss function is designed , the expression is as follows: ; in, express The global level user intention embedding of the user, express User intent embedding at the global level for the user.

[0043] The present invention uses a multi-task learning paradigm to jointly optimize the prediction task and the other three auxiliary learning objectives, and constructs the total loss function L of the comparison recommendation module, as shown in the following formula: ; in, for The weight coefficient of for The weight coefficient of for The weight coefficient of .

[0044] According to the above-mentioned project recommendation method based on multi-view contrastive learning, the following beneficial effects are achieved: (1) To address the problem of lack of important neighbor intent information in existing intent-based recommendation models, the present invention considers combining time interval information with location information to construct user cluster interest graphs and project cluster interest graphs that can express user intent, performs intra-cluster interaction within the user cluster interest graph, and performs hetero-cluster interaction between the user cluster interest graph and the project cluster interest graph. By extracting user intent embedding at the global level through the above information fusion interaction, the diversity of embedding can be effectively enriched, which helps to improve model performance. (2) To address the problem that existing intent-based recommendation models do not fully utilize subsequence information within item interaction sequences, the present invention uses a layered attention network to dynamically process subsequence embeddings at different time periods to capture dynamic changes in user intent, which can further improve model performance. (3) Combining multi-view self-contrastive learning to fuse information from different views can maximize the performance improvement of the model. Since comprehensive modeling of user intent requires consideration of many factors, how to fuse information from multiple aspects becomes a key issue. After obtaining different information from multiple views, the present invention performs multi-view contrastive learning on the final coarse-grained intent embedding, fine-grained intent embedding, and global user intent embedding. It can capture the similarities and differences of sample pairs while alleviating data sparsity, and learn more discriminative features, thereby improving the generalization ability of the model and improving the accuracy of recommendations.

[0045] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0046] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. An item recommendation method based on multi-view contrastive learning, characterized in that: Applied to a project recommendation system, the project recommendation system includes a segmentation and clustering module, a fusion extraction module and a comparison and recommendation module; The method comprises: In the segmentation and clustering module, the user interaction item sequence is obtained, and the user interaction item sequence is slidingly segmented into multiple cross subsequences. A preliminary coarse-grained intent embedding is obtained based on the cross subsequences, and the preliminary coarse-grained intent embedding is clustered to obtain a fine-grained intent embedding. In the fusion extraction module, the user interaction project sequence is fused with the time interval information and location information to construct the user cluster interest graph and the project cluster interest graph that can express the user's intention. The same cluster interaction is performed within the user cluster interest graph, and the different cluster interaction is performed on the user cluster interest graph and the project cluster interest graph, so as to obtain the user intention embedding at the global level. In the comparative recommendation module, the preliminary coarse-grained intent embedding is used to obtain the final coarse-grained intent embedding through the layer attention network. The recommendation task is performed based on the final coarse-grained intent embedding to obtain a project recommendation list. The final coarse-grained intent embedding, fine-grained intent embedding, and global user intent embedding are subjected to multi-view comparative learning to optimize the model parameters of the comparative recommendation module.

2. The project recommendation method based on multi-view contrastive learning according to claim 1, characterized in that: In the segmentation and clustering module, obtain the user interaction item sequence s u , will be split by sliding DS Actions sequence user interaction items s u Split into multiple cross subsequences, and obtain preliminary coarse-grained intent embedding based on the cross subsequences. The expression is as follows: ； in, For the r The initial coarse-grained intent embedding corresponding to each item, express s u The target item is The target item is the last item in the sequence. For the encoder.

3. The project recommendation method based on multi-view contrastive learning according to claim 2, characterized in that: In the segmentation and clustering module, K-means clustering is performed on the preliminary coarse-grained intent embedding to obtain a clustering result set. F , ,in, f h Indicates h Cluster centers, H Represents the total number of types of user intent, embeds the obtained cluster centers as fine-grained intents, and defines the distribution function Q ( f h ),like s u Belong to h clusters, then Q ( f h )=1, indicating that the user's potential intention is f h ;like s u Not in h clusters, then Q ( f h )=0, indicating that the user's potential intention is not f h .

4. The project recommendation method based on multi-view contrastive learning according to claim 3, characterized in that: In the fusion extraction module, an undirected graph is used G c represents all user interactions with items, G c ={ V c , E c },in, V c is a node set, E c is the edge set, when the node set V c Nodes in v c When the subject is a user, the node v c Represents user embedding; when the node set V c Nodes in v c When the subject is a project, the node v c Represents item embedding, and fuses the time interval information and position information of the user interaction item sequence to obtain the first k Project embedding for projects e k , the expression is: ； Among them, tanh is the hyperbolic tangent activation function, is the weight matrix in the fusion extraction module, For an undirected graph G c The initial embedding of the items obtained, is the bias vector in the fusion extraction module, For the k The position embedding vector of each position, For the k The time interval characteristics of each location; The item embeddings of all items form the item representation matrix e ; According to the direct relevance of the items, the embedding of each item is updated to obtain the following formula: ； ； ； in, represents the updated item representation matrix, e i Indicates i Project embedding for each project, e j Indicates j Project embedding for each project, Indicates i Project and j The correlation between the items is calculated. W Q and W K denote the input projections for query and key respectively, T represents the transpose operation, represents the scale factor, Indicates i Project and k The correlation between the items is calculated. K Indicates the total number of items, represents the normalized exponential function, Indicates updated , n Indicates the number of items associated with the target item, W V An input projection representing the model value; Based on the updated item representation matrix Construct user cluster interest graphs and item cluster interest graphs that can express user intentions.

5. The project recommendation method based on multi-view contrastive learning according to claim 4, characterized in that: In the fusion extraction module, the same cluster interaction is performed in the user cluster interest graph, and the expression is as follows: ； in, z k is the node interaction value after aggregation, is the MLP function, Represents nodes in the user cluster interest graph k The embedding Represents nodes in the user cluster interest graph j The embedding N k A node set representing an item; The heterogeneous cluster interaction between the user cluster interest graph and the project cluster interest graph is expressed as follows: ； in, is the relevant value of node features after aggregation, represents the Hadamard product, Represents the updated item representation matrix Project Node b The embedding A set of nodes representing the interest graph of item clusters; The global user intent embedding is obtained using the following formula: ; ； ； ； ； in, To reset the gate neuron, is the initial representation of the user node, is the input gate neural unit, , , is the weight matrix of the gated recurrent unit, , , is the bias vector of the gated recurrent unit, is the updated embedding representation of the user node, and concat is the linking operation.

6. The project recommendation method based on multi-view contrastive learning according to claim 5, characterized in that: In the comparative recommendation module, the preliminary coarse-grained intent embedding is passed through the layer attention network to obtain the final coarse-grained intent embedding, satisfying the following conditional formula: ； in, For the r The final coarse-grained intent embedding corresponding to each item; Based on the final coarse-grained intent embedding, we perform the recommendation task and obtain a list of recommended items that satisfies the following conditions: ； in, represents the predicted score of the item, and M represents the item embedding matrix.

7. The project recommendation method based on multi-view contrastive learning according to claim 6, characterized in that: The total loss function L of the comparison recommendation module satisfies the following conditional formula: ； ； ； ； ； ； ； in, is the loss function for the recommendation task, is the coarse-grained contrast loss function, is the fine-grained contrast loss function, is the global contrast loss function, for The weight coefficient of for The weight coefficient of for The weight coefficient of represents the correct prediction result of the sequence, represents the wrong prediction result of the sequence, is the contrast loss function, For sequence s 1 coarse-grained embedding, For random sampling and sequence s 1Sequences with the same target item s 2 coarse-grained embedding, f c1 Indicates distance The nearest cluster center, f c2 Indicates distance The nearest cluster center, query Indicates a query operation. express The global level user intention embedding of the user, express User intent embedding at the global level for the user.

Images