Group sequence recommendation method and system considering long-term and short-term interests of user

Abstract

The invention discloses a group sequence recommendation method and system considering long and short term interests of a user, and belongs to the technical field of deep learning. Most existing group recommendation methods are interactive-driven, multi-attribute scoring data of users and dynamic preferences of the users evolved along with time are often ignored, and when user preferences in a group are fused, correlation between auxiliary interaction information and the influence of the correlation on the group preferences are not fully considered, so that the performance of a recommendation system is limited. On the basis, a group sequence recommendation method considering long and short term interests of the users is provided, overall scores, multi-attribute scores and corresponding long and short term sequence information of the users in the group are comprehensively utilized, overall preferences and long and short term preferences of the group are mined, and group characterization is optimized and aggregated in combination with comparative learning and an attention mechanism; therefore, the effect of the group recommendation system is effectively enhanced, group users are assisted to better make decisions and choices, and the user satisfaction is improved.

Description

Technical Field

[0001] The present invention belongs to the field of deep learning technology, and specifically relates to a group sequence recommendation method and system that takes into account users' long-term and short-term interests. Background Art

[0002] Group recommendation systems aim to provide groups of users with items that meet their needs by analyzing user-item interaction information. Sequential recommendation methods, by taking time into account, can more accurately capture user preferences and generate more personalized recommendation lists. In sequential recommendation methods, user preferences can be divided into two categories: long-term preferences and short-term preferences. Long-term preferences are generally obtained by mining users' global behavior sequences and reflect their stable interests over time, while short-term preferences are obtained by mining users' short-term interaction sequences and reflect their immediate interests that change over time. By integrating long-term and short-term preferences, user preferences can be modeled more comprehensively and accurately. To further improve recommendation performance, contrastive learning has been introduced into sequential recommendation methods to optimize user representations.

[0003] However, current research on sequential recommendation methods typically recommends individual users based on their overall ratings, without considering the user's multi-attribute rating sequence or the needs of the user group. Furthermore, existing contrastive learning methods mostly use data augmentation techniques to split the original sequence into multi-view sequences. This processing approach can disrupt the inherent patterns of the original sequence, resulting in poor recommendation results. Furthermore, existing group recommendation methods are generally interaction-driven, often ignoring users' multi-attribute rating data and their dynamic preferences that evolve over time, making them unable to provide accurate recommendation services for groups of users. Summary of the Invention

[0004] The purpose of the present invention is to solve the above problems and propose a group sequence recommendation method and system that takes into account users' long-term and short-term interests. Based on the comprehensive consideration of users' overall ratings, multi-attribute ratings and time series information, the group preferences are comprehensively and meticulously modeled. By constructing long-term and short-term comparative learning, the consistency of group users in the overall rating sequence and multi-attribute rating sequence is maximized. In this way, the common preferences of group users are further learned and optimized while maintaining the integrity of the original sequence, thereby effectively enhancing the effect of group recommendation.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A group sequence recommendation method considering users' long-term and short-term interests includes the following steps:

[0007] Step 1: Obtain the initial embedding representation of the group by aggregating the ID embeddings of users in the group;

[0008] Step 2: Build a group sequence item graph by integrating the overall rating sequence of users in the group, and use a graph convolutional network to capture the multi-hop information of users in the group to generate the overall group preference;

[0009] Step 3: Use the attention mechanism to analyze and measure the relationship between the overall rating sequence and the multi-attribute rating sequence of users in the group, and model the group's long-term and short-term preferences;

[0010] Step 4: Integrate the group's overall preference with the long-term and short-term preferences to form the group's comprehensive preference, and calculate the inner product of the comprehensive preference and the candidate item to obtain the project prediction score;

[0011] Step 5: Establish contrastive learning by maximizing the consistency between the long-term and short-term preferences of the overall ratings and the long-term and short-term preferences of the multi-attribute ratings of the users in the group, and then optimize the method using the total loss function obtained by mixing the recommendation loss function and the contrast loss function.

[0012] Furthermore, the implementation method of step 1 is as follows:

[0013] Use Embedding encoding for group g l The user ID in the group is initially embedded. l The user vector in is represented as Group g l The user vectors in the group are spliced ​​and reduced in dimension to obtain the group g l The initial embedding representation of

[0014] Furthermore, the implementation method of step 2 is as follows:

[0015] The learning process of the group's overall preference is achieved by integrating the group g l The interaction sequence of users within the group is used to construct a group sequence project graph, and then based on the connection relationship between items in the group sequence project graph, the adjacency matrix of the sequence project graph is established. The specific calculation formula is as follows:

[0016]

[0017] In order to deeply explore the characteristics of the node itself and learn more accurate node features, Add the unit matrix and perform normalization. The specific calculation formula is as follows:

[0018]

[0019] in, is the normalized sequence item graph adjacency matrix, D -1 / 2 is a diagonal matrix, I E is the identity matrix;

[0020] Then, the initial embedding representation of the group is obtained through the graph convolutional network. Perform linear transformation to obtain the overall preference of the group The calculation formula is as follows:

[0021]

[0022] in, is the overall preference of the group at layer t-1, For group g l The initial embedding representation, W t is the trainable parameter matrix of the tth layer.

[0023] Furthermore, the implementation method of step 3 is as follows:

[0024] The user's interaction records within one month are considered as the user's short-term sequence, and the user's interaction records within one year are considered as the long-term sequence for analysis;

[0025] In order to learn the long-term preference of the group, based on the in-depth mining of the user's overall rating sequence and multi-attribute rating sequence, the attention mechanism is used to learn and aggregate the long-term preference characteristics of the users in the group to obtain the group g l Long-term preference for the overall rating sequence and long-term preferences on multi-attribute rating sequences

[0026] Group g l Long-term preference for the overall rating sequence The calculation formula is as follows:

[0027]

[0028]

[0029] in, Represents a user Long-term sequence information on the overall score, Represents a user Query matrix on the overall score, Represents a user Key matrix on the overall score, Represents a user The Value matrix on the overall rating, represents the learnable transformation matrix;

[0030] Group g l Long-term preferences over multi-attribute rating sequences The calculation formula is as follows:

[0031]

[0032] in, Represents a user In attribute a v Long-term sequence information on Represents a user In attribute a v The weight on Indicates group g l In attribute a v Long-term preference on long represents the group's long-term preference influence weight matrix, b long represents the group’s long-term preference bias vector;

[0033] In-depth analysis of recent interaction sequences of users within a group can effectively capture the group's current needs and ensure that recommended items are consistent with the group's current interests;

[0034] The user's overall rating sequence and multi-attribute rating sequence are used to learn the short-term preferences of users in the group, and the user weights are obtained through the attention mechanism to comprehensively model the decision-making process and obtain the group g l Short-term preference for overall rating series and its short-term preference on multi-attribute rating sequences

[0035] Group g l Short-term preference for overall rating series The calculation formula is as follows:

[0036]

[0037]

[0038] in, Represents a user Short-term sequence information on the overall score, Represents a user Query matrix on the overall score, Represents a user Key matrix on the overall score, Represents a user The Value matrix on the overall rating, represents the learnable transformation matrix;

[0039] Group g l Short-term preferences on multi-attribute rating sequences The calculation formula is as follows:

[0040]

[0041] in, Represents a user In attribute a v Short-term sequence information on Represents a user In attribute a v The weight on Indicates group g l In attribute a v Short-term preference on W short represents the group short-term preference influence matrix, b short Represents the group's short-term preference bias vector.

[0042] Furthermore, the implementation method of generating the prediction score in step 4 is as follows:

[0043] Initially embed the group into Group overall preference Long-term preference for the group's overall rating sequence Long-term preference of group multi-attribute rating sequence Short-term preference of group overall rating sequence and short-term preference of group multi-attribute rating sequence Weighted summation to obtain the group comprehensive preference g l , the specific calculation formula is as follows:

[0044]

[0045] The group's comprehensive preference g l With candidate project i j Perform inner product operation to obtain group g l For candidate project i j Prediction score of The specific calculation formula is as follows:

[0046]

[0047] Furthermore, the specific optimization method in step 5 is as follows:

[0048] After obtaining the predicted score, use the total loss function Optimize the method;

[0049] The total loss function consists of two parts: the recommendation loss function and contrastive learning loss function

[0050] The recommendation loss function The calculation formula is as follows:

[0051]

[0052] Where M represents the number of candidate items in the group, y lj represents the group g in the training set l For Project i j The actual rating, Indicates g l to i j 's prediction score;

[0053] In the recommendation loss function, the average rating of the items by users in the group is used to represent the actual rating of the group on the items;

[0054] In order to obtain high-quality group embedding representation, we use contrastive learning to maximize the consistency between the long-term and short-term preferences of the overall ratings of users in the group and the long-term and short-term preferences of the multi-attribute ratings. The higher the similarity, the better the recommendation effect.

[0055] The contrastive learning loss function The calculation formula is as follows:

[0056]

[0057] Among them, cos() is the cosine similarity function, For group g l The similarity between the long-term preference of the overall rating sequence and the long-term preference of the multi-attribute rating sequence, For group g l The similarity between the short-term preference of the overall rating sequence and the short-term preference of the multi-attribute rating sequence, τ is the coefficient;

[0058] By minimizing the total loss function To optimize the method:

[0059]

[0060] Where λ is the coefficient and π is the regularization weight of the learning parameter set Θ.

[0061] A group sequence recommendation system that considers users' long-term and short-term interests, including the following modules:

[0062] Group initial embedding representation module: Use Embedding encoding to represent the group g l The initial embedding of the user ID is represented by the group g l The user vector in is represented as The user vectors in the group are concatenated and dimensionally reduced to obtain the group g l The initial embedding representation of

[0063] Group overall preference learning module: integrating group g lThe interaction sequence of users within the group is used to construct a group sequence project graph. Based on the connection relationship between items in the group sequence project graph, the adjacency matrix of the sequence project graph is established. right After adding the identity matrix and normalizing, the initial embedding representation of the group is obtained through the graph convolutional network. Perform linear transformation to obtain the overall preference of the group

[0064] Group long-term and short-term preference learning module: Use the attention mechanism to learn the long-term preference characteristics of users in the group from the user's overall rating sequence and multi-attribute rating sequence, and obtain the group g through aggregation l Long-term preference for the overall rating sequence and its long-term preference on multi-attribute rating sequences At the same time, the attention mechanism is used to obtain user weights and more comprehensively model the decision-making process to obtain group g l Short-term preference for overall rating series and its short-term preference on multi-attribute rating sequences

[0065] Prediction scoring module: Initially embedding the group Group overall preference Long-term preference for the group's overall rating sequence Long-term preference of group multi-attribute rating sequence Short-term preference of group overall rating sequence and short-term preference of group multi-attribute rating sequence Perform weighted summation to obtain the group comprehensive preference g l , by combining the group's comprehensive preference g l With candidate project i j Perform inner product operation to obtain group g l For candidate project i j Prediction score of

[0066] Method optimization module: After obtaining the prediction score, the method is optimized using the total loss function obtained by mixing the recommendation loss function and the contrastive learning loss function;

[0067] Recommendation list generation module: The system generates a corresponding item recommendation list for the recommended group based on the comprehensive embedding representation of the recommended group.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] 1. This invention can more comprehensively and accurately model the common preferences of group users, provide the group with a list of recommended items that meet their needs, and thus assist the group users in making decisions faster and better, thereby improving user satisfaction and experience, and effectively enhancing the effectiveness of group recommendations;

[0070] 2. The present invention can be used in various network applications such as e-commerce, social media, online education, and online medical care to improve the quality and efficiency of service networks, better meet the real needs of intelligence, groupization, and diversification in the big data era, and further unleash the economic and market value of recommendation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 Schematic diagram of the process of the present invention;

[0072] Figure 2 Schematic diagram of the framework of the method recommended by the present invention;

[0073] Figure 3 Schematic diagram of the structure of the system recommended by the present invention. DETAILED DESCRIPTION

[0074] The present invention will be further described below with reference to the accompanying drawings and examples.

[0075] like Figure 1-3 As shown, a group sequence recommendation method considering users' long-term and short-term interests includes the following steps:

[0076] Step 1: Get the initial embedding representation of the group by aggregating the ID embeddings of the users in the group. The implementation method is as follows:

[0077] The present invention uses Embedding coding to encode the group g l The user ID in the group is initially embedded. l The user vector in is represented as Group g l The user vectors in the group are spliced ​​and reduced in dimension to obtain the group g l The initial embedding representation of

[0078] Step 2: Build a group sequence item graph by integrating the overall rating sequence of users in the group, and use a graph convolutional network to capture the multi-hop information of users in the group to generate the overall group preference. The implementation method is as follows:

[0079] The learning process of the group's overall preference is achieved by integrating the group g l The interaction sequence of users within the group is used to construct a group sequence project graph, and then based on the connection relationship between items in the group sequence project graph, the adjacency matrix of the sequence project graph is established. The specific calculation formula is as follows:

[0080]

[0081] In order to deeply explore the characteristics of the node itself and learn more accurate node features, Add the unit matrix and perform normalization. The specific calculation formula is as follows:

[0082]

[0083] in, is the normalized sequence item graph adjacency matrix, D -1 / 2 is a diagonal matrix, I E is the identity matrix;

[0084] Then, the initial embedding representation of the group is obtained through the graph convolutional network. Perform linear transformation to obtain the overall preference of the group The calculation formula is as follows:

[0085]

[0086] in, is the overall preference of the group at layer t-1, For group g l The initial embedding representation, W t is the trainable parameter matrix of the tth layer.

[0087] Step 3: Use the attention mechanism to analyze and measure the relationship between the overall rating sequence and the multi-attribute rating sequence of users in the group, and model the group's long-term and short-term preferences. The implementation method is as follows:

[0088] Long-term sequences usually involve more persistent user preferences and habits, generally user behavior records spanning several months or even years, while short-term sequences reflect users' recent behavioral trends and interests, generally user behavior records spanning the past few days, weeks, or a month.

[0089] The present invention analyzes the user's interaction records within one month as the user's short-term sequence and the user's interaction records within one year as the long-term sequence;

[0090] In order to learn the long-term preference of the group, based on the in-depth mining of the user's overall rating sequence and multi-attribute rating sequence, the attention mechanism is used to learn and aggregate the long-term preference characteristics of the users in the group to obtain the group g l Long-term preference for the overall rating sequence and long-term preferences on multi-attribute rating sequences

[0091] Group g l Long-term preference for the overall rating sequence The calculation formula is as follows:

[0092]

[0093] in, Represents a user Long-term sequence information on the overall score, Represents a user Query matrix on the overall score, Represents a user Key matrix on the overall score, Represents a user The Value matrix on the overall rating, represents the learnable transformation matrix;

[0094] Group g l Long-term preferences over multi-attribute rating sequences The calculation formula is as follows:

[0095]

[0096] in, Represents a user In attribute a v Long-term sequence information on Represents a user In attribute a v The weight on Indicates group g l In attribute a v Long-term preference on long represents the group's long-term preference influence weight matrix, b long represents the group’s long-term preference bias vector;

[0097] In-depth analysis of the recent interaction sequences of users in the group can effectively capture the current needs of the group and ensure that the recommended items are consistent with the current interests of the group. This paper uses the user's overall rating sequence and the multi-attribute rating sequence to learn the short-term preferences of users in the group, and obtains the user weight through the attention mechanism, so as to more comprehensively model the decision-making process and obtain the group g l Short-term preference for overall rating series and its short-term preference on multi-attribute rating sequences

[0098] Group g l Short-term preference for overall rating series The calculation formula is as follows:

[0099]

[0100] in, Represents a user Short-term sequence information on the overall score, Represents a user Query matrix on the overall score, Represents a user Key matrix on the overall score, Represents a user The Value matrix on the overall rating, represents the learnable transformation matrix;

[0101] Group g l Short-term preferences on multi-attribute rating sequences The calculation formula is as follows:

[0102]

[0103] in, Represents a user In attribute a v Short-term sequence information on Represents a user In attribute a v The weight on Indicates group g l In attribute a v Short-term preference on W short represents the group short-term preference influence matrix, b short Represents the group's short-term preference bias vector.

[0104] Step 4: Integrate the group's overall preference with the long-term and short-term preferences to form the group's comprehensive preference. Calculate the inner product between the comprehensive preference and the candidate item to obtain the project prediction score. The implementation method is as follows:

[0105] Initially embed the group into Group overall preference Long-term preference for the group's overall rating sequence Long-term preference of group multi-attribute rating sequence Short-term preference of group overall rating sequence and short-term preference of group multi-attribute rating sequence Weighted summation to obtain the group comprehensive preference g l , the specific calculation formula is as follows:

[0106]

[0107] The group's comprehensive preference g l With candidate project i j Perform inner product operation to obtain group g lFor candidate project i j Prediction score of The specific calculation formula is as follows:

[0108]

[0109] Step 5: Establish contrastive learning by maximizing the consistency of the long-term and short-term preferences of the overall ratings and the long-term and short-term preferences of the multi-attribute ratings of the users in the group. Optimize the method using the total loss function obtained by mixing the recommendation loss function and the contrastive loss function. The specific optimization method is as follows:

[0110] After obtaining the predicted score, use the total loss function Optimize the method;

[0111] The total loss function consists of two parts: the recommendation loss function and contrastive learning loss function

[0112] The recommendation loss function The calculation formula is as follows:

[0113]

[0114] Where M represents the number of candidate items in the group, y lj represents the group g in the training set l For Project i j The actual rating, Indicates g l to i j 's prediction score;

[0115] In the recommendation loss function, the average rating of the items by users in the group is used to represent the actual rating of the group on the items;

[0116] Since the groups in the dataset are obtained through clustering, the actual ratings of the groups on the items do not exist. Therefore, the average rating of the users in the group on the items is used in the recommendation loss function to represent the actual ratings of the groups on the items.

[0117] In order to obtain high-quality group embedding representation, we use contrastive learning to maximize the consistency of the long-term and short-term preferences of the overall ratings and the long-term and short-term preferences of the multi-attribute ratings of users in the group. The higher the similarity, the better the recommendation effect.

[0118] The contrastive learning loss function The calculation formula is as follows:

[0119]

[0120] Among them, cos() is the cosine similarity function, For group gl The similarity between the long-term preference of the overall rating sequence and the long-term preference of the multi-attribute rating sequence, For group g l The similarity between the short-term preference of the overall rating sequence and the short-term preference of the multi-attribute rating sequence, τ is the coefficient;

[0121] By minimizing the total loss function To optimize the method:

[0122]

[0123] Where λ is the coefficient and π is the regularization weight of the learning parameter set Θ.

[0124] A group sequence recommendation system that considers users' long-term and short-term interests, including the following modules:

[0125] Group initial embedding representation module: Use Embedding encoding to represent the group g l The initial embedding of the user ID is represented by the group g l The user vector in is represented as The user vectors in the group are concatenated and dimensionally reduced to obtain the group g l The initial embedding representation of

[0126] Group overall preference learning module: integrating group g l The interaction sequence of users within the group is used to construct a group sequence project graph. Based on the connection relationship between items in the group sequence project graph, the adjacency matrix of the sequence project graph is established. right After adding the identity matrix and normalizing, the initial embedding representation of the group is obtained through the graph convolutional network. Perform linear transformation to obtain the overall preference of the group

[0127] Group long-term and short-term preference learning module: Use the attention mechanism to learn the long-term preference characteristics of users in the group from the user's overall rating sequence and multi-attribute rating sequence, and obtain the group g through aggregation l Long-term preference for the overall rating sequence and its long-term preference on multi-attribute rating sequences At the same time, the attention mechanism is used to obtain user weights and more comprehensively model the decision-making process to obtain group g l Short-term preference for overall rating series and its short-term preference on multi-attribute rating sequences

[0128] Prediction scoring module: Initially embedding the group Group overall preference Long-term preference for the group's overall rating sequence Long-term preference of group multi-attribute rating sequence Short-term preference of group overall rating sequence and short-term preference of group multi-attribute rating sequence Perform weighted summation to obtain the group comprehensive preference g l ,

[0129]

[0130] By combining the group's preferences g l With candidate project i j Perform inner product operation to obtain group g l For candidate project i j Prediction score of

[0131]

[0132] Method optimization module: After obtaining the prediction score, use the total loss function To optimize the method, the total loss function consists of two parts: the recommended loss function and contrastive learning loss function Using the minimized total loss function To optimize the method:

[0133]

[0134] Where λ is the coefficient and π is the regularization weight of the learning parameter set Θ;

[0135] Recommendation list generation module: The system generates a corresponding item recommendation list for the recommended group based on the comprehensive embedding representation of the recommended group.

Claims

1. A group sequence recommendation method considering users' long-term and short-term interests, characterized in that: The following steps are involved: Step 1: Get the initial embedding representation of the group by aggregating the ID embeddings of the users in the group; Step 2: Construct a group sequence item graph by integrating the overall rating sequence of users in the group, and use a graph convolutional network to capture the multi-hop information of users in the group to generate the overall preference of the group; Step 3: Use the attention mechanism to analyze and measure the relationship between the overall rating sequence and the multi-attribute rating sequence of users in the group, and model the long-term and short-term preferences of the group; Step 4: Integrate the group's overall preference with the long-term and short-term preferences to form the group's comprehensive preference, and calculate the inner product between it and the candidate item to obtain the project prediction score; Step 5: Contrastive learning is established by maximizing the consistency of the long-term and short-term preferences of the overall ratings of users in the group and the long-term and short-term preferences of the multi-attribute ratings, and then the method is optimized using the total loss function obtained by mixing the recommendation loss function and the contrast loss function.

2. The method for recommending group sequences considering the long-term and short-term interests of users according to claim 1, characterized in that: The implementation method of step 1 is as follows: Use Embedding encoding for group g l The user ID in the group g is initially embedded. l The user vector in is represented as Group g l The user vectors in the group are concatenated and reduced in dimension to obtain the group g l The initial embedding representation of 3. The group sequence recommendation method considering the long-term and short-term interests of users according to claim 1, characterized in that: The implementation method of step 2 is as follows: The learning process of the group's overall preference is achieved by integrating the group g l The interaction sequence of users within the group is used to construct a group sequence project graph, and then based on the connection relationship between items in the group sequence project graph, the adjacency matrix of the sequence project graph is established. The specific calculation formula is as follows: In order to deeply explore the characteristics of the node itself and learn more accurate node features, Add the unit matrix and perform normalization. The specific calculation formula is as follows: in, is the normalized sequence item graph adjacency matrix, is a diagonal matrix, I E is the identity matrix; Then, the initial embedding representation of the group is obtained through the graph convolutional network. Perform linear transformation to obtain the overall preference of the group The calculation formula is as follows: in, is the overall preference of the group at the t-1th layer, For group g l The initial embedding representation, W t is the trainable parameter matrix of the tth layer.

4. The method for recommending group sequences considering users' long-term and short-term interests according to claim 1, characterized in that: The implementation method of step 3 is as follows: The user's interaction records within one month are regarded as the user's short-term series, and the user's interaction records within one year are regarded as the long-term series for analysis; In order to learn the long-term preferences of the group, based on the in-depth mining of the user's overall rating sequence and multi-attribute rating sequence, the attention mechanism is used to learn and aggregate the long-term preference characteristics of the users in the group to obtain the group g l Long-term preference for overall rating sequence and long-term preferences over multi-attribute rating sequences Group l Long-term preference for overall rating sequence The calculation formula is as follows: in, Indicates user Long-term sequence information on the overall score, Indicates user Query matrix on the overall score, Indicates user Key matrix on the overall score, Indicates user The Value matrix on the overall rating, represents a learnable transformation matrix; Group l Long-term preferences over multi-attribute rating sequences The calculation formula is as follows: in, Indicates user In attribute a v Long-term sequence information on Indicates user In attribute a v The weight on Indicates group g l In attribute a v The long-term preference on long represents the weight matrix of group long-term preference influence, b long represents the group's long-term preference bias vector; In-depth analysis of the recent interaction sequences of users in the group can effectively capture the current needs of the group and ensure that the recommended items are consistent with the current interests of the group; The user's overall rating sequence and multi-attribute rating sequence are used to learn the short-term preferences of users in the group, and the user weights are obtained through the attention mechanism to comprehensively model the decision-making process and obtain the group g l Short-term preference over overall rating series and its short-term preference on multi-attribute rating sequences Group l Short-term preference over overall rating series The calculation formula is as follows: in, Indicates user Short-term sequence information on the overall score, Indicates user Query matrix on the overall score, Indicates user Key matrix on the overall score, Indicates user The Value matrix on the overall rating, represents a learnable transformation matrix; Group l Short-term preference over multi-attribute rating sequences The calculation formula is as follows: in, Indicates user In attribute a v Short-term sequence information on Indicates user In attribute a v The weight on Indicates group g l In attribute a v Short-term preference on short represents the group short-term preference influence matrix, b short Represents the group's short-term preference bias vector.

5. The method for recommending group sequences considering users' long-term and short-term interests according to claim 1, characterized in that: The implementation method of generating the prediction score in step 4 is as follows: Initially embed the group into Group overall preference Long-term preference for group overall rating sequence Long-term preference of group multi-attribute rating sequence Short-term preference of group overall rating sequence and short-term preference of group multi-attribute rating sequences Weighted summation to obtain the group comprehensive preference g l , the specific calculation formula is as follows: The group's comprehensive preference g l With candidate project i j Perform inner product operation to obtain group g l For candidate projects j Prediction score of The specific calculation formula is as follows:

6. The method for recommending group sequences considering users' long-term and short-term interests according to claim 1, characterized in that: The specific optimization method in step 5 is as follows: After obtaining the prediction score, use the total loss function Optimize the method; The total loss function consists of two parts: the recommended loss function And contrastive learning loss function The recommendation loss function The calculation formula is as follows: Where M represents the number of candidate items in the group, y lj represents the group g in the training set l For Project I j The actual rating of Indicates g l to i j 's prediction score; In the recommendation loss function, the average rating of the items by users in the group is used to represent the actual rating of the group on the items; In order to obtain high-quality group embedding representation, contrastive learning is established by maximizing the consistency of the long-term and short-term preferences of the overall ratings of users in the group and the long-term and short-term preferences of multi-attribute ratings. The higher the similarity, the better the recommendation effect. The contrastive learning loss function The calculation formula is as follows: Among them, cos() is the cosine similarity function, For group g l The similarity between the long-term preference of the overall rating sequence and the long-term preference of the multi-attribute rating sequence, For group g l The similarity between the short-term preference of the overall rating sequence and the short-term preference of the multi-attribute rating sequence, τ is the coefficient; By minimizing the total loss function To optimize the method: Among them, λ is the coefficient and π is the regularization weight of the learning parameter set Θ.

7. A group sequence recommendation system considering users' long-term and short-term interests, characterized in that: Includes the following modules: Group initial embedding representation module: Use Embedding encoding to represent the group g l The initial embedding of the user ID in group g l The user vector in is represented as The user vectors in the group are concatenated and dimensionally reduced to obtain the group g l The initial embedding representation of Group overall preference learning module: integrating group g l The interaction sequence of users within the group is used to construct a group sequence project graph. Based on the connection relationship between items in the group sequence project graph, an adjacency matrix of the sequence project graph is established. right After adding the identity matrix and normalizing, the initial embedding representation of the group is obtained through the graph convolutional network. Perform a linear transformation to obtain the overall preference of the group Group long-term and short-term preference learning module: The attention mechanism is used to learn the long-term preference characteristics of users in the group from the user's overall rating sequence and multi-attribute rating sequence, and the group g is obtained by aggregation. l Long-term preference for overall rating sequence and its long-term preference on multi-attribute rating sequences At the same time, the attention mechanism is used to obtain user weights and model the decision-making process more comprehensively to obtain group g l Short-term preference over overall rating series and its short-term preference on multi-attribute rating sequences Prediction and scoring module: Initially embedding the group into a representation Group overall preference Long-term preference for group overall rating sequence Long-term preference of group multi-attribute rating sequence Short-term preference of group overall rating sequence and short-term preference of group multi-attribute rating sequences Perform weighted summation to obtain the group comprehensive preference g l , by comprehensively analyzing the group preference g l With candidate project i j Perform inner product operation to obtain group g l For candidate projects j Prediction score of Method optimization module: After obtaining the prediction score, the method is optimized using the total loss function obtained by mixing the recommendation loss function and the contrastive learning loss function; Recommendation list generation module: The system generates a corresponding item recommendation list for the recommended group based on the comprehensive embedding representation of the recommended group.

Images