Session recommendation method based on de-noising edge sensing network

Abstract

The invention provides a session recommendation method based on a de-noising edge sensing network, which belongs to the technical field of session recommendation, and comprises the following steps: mining items in all historical session data and sequence hopping neighbors of the items, and constructing a frequency sensing graph by utilizing co-occurrence frequencies of the items and the sequence hopping neighbors of the items; according to the information carried by the edge, obtaining a project embedding matrix, distributing learning parameters for the edge by using a de-noising edge sensing network, obtaining project representation learning, and obtaining project representation through calculation; a soft attention mechanism is utilized to obtain session characterization; according to the session representation and the item embedding matrix, calculating to obtain a ground live one-hot coding vector, and according to the ground live one-hot coding vector and the independent contribution of each item, judging the next target item after the user inputs the session to finish session recommendation; the technical problems that an existing session recommendation algorithm neglects the modeling problem of the frequency relation between items and cannot effectively process noise are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of conversation recommendation, and in particular relates to a conversation recommendation method based on a denoising edge perception network. Background Art

[0002] Recommender systems are essential for online platforms, extracting users' preferences from their profiles and behaviors to combat information overload. However, users are often anonymous and transient on websites, such as news and streaming platforms, making session-based recommendations crucial, which focus on predicting the next item based on anonymous interaction sequences.

[0003] Traditional session-based recommendation algorithms model complex item transition relationships by assuming that the next item is only related to the last one or by leveraging deep learning to extract potential information and complex relationships from raw data. However, traditional methods only aggregate information from adjacent items to facilitate learning item representations, and most ignore the impact of potential noise neighbors caused by user misclicks or drifting interests. A small number of algorithms take noise issues into account and remove noise items from input sessions. However, due to the short input sequence characteristics of session recommendations, denoising strategies that work directly on the input will result in sparse data, and this type of strategy requires accurate extraction of the user's overall intent, which may cause items to be incorrectly classified as noise. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a conversation recommendation method based on a denoising edge perception network, which solves the technical problems that the existing conversation recommendation algorithm ignores the modeling problem of the frequency relationship between items and cannot effectively process noise.

[0005] In order to achieve the above objectives, the technical solution adopted by the present invention is: a conversation recommendation method based on a denoising edge perception network, comprising the following steps:

[0006] S1. Mining all items and their hop-order neighbors in the historical conversation data, and constructing a frequency perception graph containing each item using the co-occurrence frequency of each item and its hop-order neighbors;

[0007] S2. According to the information carried by the edge in the frequency perception graph, the item embedding matrix is ​​obtained, and the learning parameters are assigned to the edge using the denoising edge perception network to obtain the item representation learning and obtain the item representation through calculation;

[0008] S3. Based on the item representation, the soft attention mechanism is used to obtain the session representation;

[0009] S4. Based on the session representation and the item embedding matrix, the one-hot encoding vector of the ground truth is calculated. Based on the one-hot encoding vector of the ground truth and the individual contribution of each item, combined with the current item after the user inputs the session, the next item is predicted to complete the session recommendation.

[0010] The beneficial effects of the present invention are as follows: the present invention takes sparse consideration of adjacent items, eliminates the influence of potential noise neighbors, improves the accuracy of conversation recommendations, and refines the connection relationship between items in the input conversation by using a frequency perception map, thereby improving the representativeness of the items and making the recommendation method more accurate and comprehensive. The present invention can be used on multiple platforms with a set of networks and is not limited to recommendations on a specific platform, thereby improving the practicality of conversation recommendations and enhancing the generalizability of conversation recommendations, and achieving the ability to accurately recommend the next item based on the conversation even if the user accidentally touches it.

[0011] Furthermore, the S1 comprises the following steps:

[0012] S101, mining items and their hopping neighbors in all historical conversation data, combining items and their hopping neighbors into adjacent item pairs, calculating the number of adjacent item pairs in all historical conversation data, and using the number of adjacent item pairs as the weight of the edge in the frequency perception graph;

[0013] S102, normalizing the edge weight of each item according to the edge weight to obtain a frequency representation of the edge weight of each item;

[0014] S103, classifying edge types according to the frequency representation of edge weights of each item, and combining the frequency representation of edge weights of hopping neighbors of each item to obtain the co-occurrence frequency of each item and its hopping neighbors, and constructing a frequency perception map containing each item.

[0015] The beneficial effect of the above further scheme is: the present invention constructs a frequency perception graph in the denoising edge perception network to understand the relative importance of each neighbor node to the central item, thereby improving the noise item perception ability of the denoising edge perception network.

[0016] Furthermore, the S2 comprises the following steps:

[0017] S201, obtaining a specific input session according to information carried by the edge in the frequency perception graph, initializing the vector form of the input session to obtain a vectorized session, and combining the vectorized sessions to obtain a project embedding matrix;

[0018] S202, using the edge-aware graph attention subnetwork in the denoising edge-aware network to assign learnable parameters to the edge;

[0019] S203, using vectorized sessions and learnable parameters combined with leaky rectified linear units, calculating the attention score in the semantic space from item to item neighbor, using a self-attention mechanism based on an α-entmax activation function, and using the attention score to calculate a sparse weight score;

[0020] S204, obtaining the final attention weight according to the sparse weight score of each item;

[0021] S205. Linearly combine the final attention weight and the vectorized session associated with the final attention weight to calculate the item representation.

[0022] The beneficial effects of the above further scheme are as follows: the present invention uses the edge-aware graph attention subnetwork in the denoising edge-aware network, and utilizes the edge-aware graph attention subnetwork to more accurately model the effect of the graph structure according to the information carried by the edges, assigns different learning parameters to the edges, refines the learning process, improves the accuracy of item representation learning, and uses a self-attention mechanism to sparsely consider the learning parameters, eliminating the influence of noise items on central items, so that the conversation recommendation method of the present invention can effectively handle noise.

[0023] Furthermore, the S3 includes the following steps:

[0024] S301, calculating the average value of the item representations, and representing the user's interaction information as the average value of the item representations;

[0025] S302: Use the attention mechanism to obtain the weight of each item, and linearly combine the weight of each item with the item representation to calculate the session representation.

[0026] Furthermore, the expression of the session representation is as follows:

[0027]

[0028] Among them, R u represents the session representation of session u, n represents the length of the session, i represents the i-th item in the session, δ i represents the weight of the i-th item, represents the i-th item representation of session u, represents the i-th interaction information of user in session u, b1, W1, and W2 all represent learnable parameters; b1∈R d ,W1,W2∈R d×d , R d represents a set of points in d-dimensional real space, R d×d Represents a set of d×d matrices of real numbers, where d represents the dimension.

[0029] Furthermore, the expression of the one-hot encoding vector of the ground truth is as follows:

[0030] y i =softmax((R u ) T E i )

[0031] Among them, y i represents the one-hot encoded vector of the ground truth, softmax(·) represents the normalized exponential function, and R u represents the session representation of session u, E i Represents the item embedding matrix.

[0032] The beneficial effects of the above further scheme are as follows: the present invention improves the ability to distinguish and obtain the individual contribution of each item by adopting a soft attention mechanism to obtain session representation, and improves the accuracy of the prediction of the next item by combining the individual contribution of each item in the current session into the prediction of the next item by utilizing the session representation. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The present invention is a flow chart of the method.

[0034] Figure 2 : is the overall structure diagram of the denoising edge perception network in this embodiment. DETAILED DESCRIPTION

[0035] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0036] Before describing this embodiment, the following terms are explained:

[0037] EA-GAT: Edge-Aware Graph Attention Network;

[0038] D-EAN: Denoising Edge-Aware Network;

[0039] α-entmax: α-entropy maximization normalization;

[0040] Sigmoid: S-shaped curve function;

[0041] Softmax: Normalized exponential function.

[0042] Example

[0043] In this embodiment, the present invention provides a session recommendation method based on a denoising edge-aware network, which takes the establishment of a frequency-aware graph as the basic idea, captures more detailed connection relationships according to the global frequency of the connection between items and adjacent items in the user input session, uses the edge-aware graph attention subnetwork (EA-GAT) in the denoising edge-aware network (D-EAN) to identify edge connection relationships and process neighbor noise, and uses a soft attention mechanism to identify and combine harvesters, and combines the individual contributions of each item in the current session into the next item prediction, solving the problem of recommendation bottlenecks caused by the inability to effectively process noise;

[0044] In order to solve the problem of ignoring the frequency relationship between items in the conversation, which leads to the inability to obtain the relationship with adjacent items, a frequency perception graph was established. When using the edge perception graph attention network to learn the item representation process, parameters are assigned to each edge according to the type of connected edges in the frequency perception graph. A soft attention mechanism is used to obtain the conversation representation for prediction. Finally, after softmax, the unique hot encoding of each category is obtained. The most likely next target item is determined based on the unique hot encoding to complete the conversation recommendation based on the denoising edge perception network.

[0045] like Figure 1 As shown, the present invention provides a conversation recommendation method based on a denoising edge perception network, and its implementation method is as follows:

[0046] S1. Mining all items and their hop-order neighbors in the historical session data, and constructing a frequency perception graph containing each item using the co-occurrence frequency of each item and its hop-order neighbors. The specific steps are as follows:

[0047] S101, mining items and their hopping neighbors in all historical conversation data, combining items and their hopping neighbors into adjacent item pairs, calculating the number of adjacent item pairs in all historical conversation data, and using the number of adjacent item pairs as the weight of the edge in the frequency perception graph;

[0048] S102, normalizing the edge weight of each item according to the edge weight to obtain a frequency representation of the edge weight of each item;

[0049] S103, classifying edge types according to the frequency representation of edge weights of each item, and combining the frequency representation of edge weights of hopping neighbors of each item to obtain the co-occurrence frequency of each item and its hopping neighbors, and constructing a frequency perception map containing each item.

[0050] In this embodiment, a frequency perception graph of each item is constructed using the co-occurrence frequencies of items and their hopping neighbors mined from all historical session data. First, the number of adjacent item pairs in all sessions is calculated, and then the number of adjacent item pairs is used as the weight of the edge in the frequency perception graph. On this basis, the edge weight of each item is normalized to obtain its frequency representation, and the edge type is classified according to the frequency. By constructing a frequency perception graph, the relative importance of each neighbor node to the central item can be understood to perceive the noise item; wherein, the edge types are divided into four categories: ultra-high frequency edge, high frequency edge, medium frequency edge and low frequency edge.

[0051] S2. According to the information carried by the edge in the frequency perception graph, the item embedding matrix is ​​obtained, and the denoising edge perception network is used to assign learning parameters to the edge, and the item representation learning is obtained and the item representation is obtained through calculation. The specific steps are as follows:

[0052] S201, obtaining a specific input session according to information carried by the edge in the frequency perception graph, initializing the vector form of the input session to obtain a vectorized session, and combining the vectorized sessions to obtain a project embedding matrix;

[0053] S202, using the edge-aware graph attention subnetwork in the denoising edge-aware network to assign learnable parameters to the edge;

[0054] S203, using vectorized sessions and learnable parameters combined with leaky rectified linear units, calculating the attention score in the semantic space from item to item neighbor, using a self-attention mechanism based on an α-entmax activation function, and using the attention score to calculate a sparse weight score;

[0055] S204, obtaining the final attention weight according to the sparse weight score of each item;

[0056] S205. Linearly combine the final attention weight and the vectorized session associated with the final attention weight to calculate the item representation.

[0057] In this embodiment, Figure 2 As shown, Figure 2 The overall architecture of the denoising edge-aware network is used. The EA-GAT subnetwork in the denoising edge-aware network can more accurately model the effect of the graph structure according to the information carried by the edges, assign different learning parameters to the edges to refine the learning process, and thus obtain item representation learning, and obtain item representation;

[0058] In the item representation learning process of the EA-GAT subnetwork, the influence of adjacent noise items in the EA-GAT learning process is taken into account, and α-entmax is used to sparsely consider the learning parameters to minimize the weight score of the noise item on the central item.

[0059] In this embodiment, the session information carried by the edge, the specific input session Initialize the vector form of the input session as The vectorized sessions are combined into a matrix form as the item embedding matrix E i , the attention score of item i to its neighbors in the semantic space is calculated considering semantic and structural noise, and the calculation formula is as follows:

[0060]

[0061] in, represents the attention score in the semantic space from item i to item j, LeakyReLU(·) represents the rectified linear unit with leakage, r i,j represents the edge e between item i and item j in the frequency-aware graph ij Type, represents the edge e assigned to item i and item j ij The learnable parameters of R d×d represents a vector space, represents the vectorized session for project i, represents the vectorized session of item j, ⊙ represents matrix dot product;

[0062] The self-attention mechanism based on α-entmax is used to completely eliminate the influence of noise and calculate the weight score of sparse consideration, the sparse weight score, which is calculated as follows:

[0063]

[0064] in, represents the sparse weight score, represents a hyperparameter;

[0065] The final attention weight is obtained from the sparse weight score, and then the features associated with the final attention weight are linearly combined to calculate the item representation The calculation formula of project representation is as follows:

[0066]

[0067] in, represents the item representation of item i, represents the final attention weight from item i to item j, represents the vectorized session for project j, Represents the neighbor node set of node i in session u, including all nodes connected to node i.

[0068] S3. Based on the item representation, the soft attention mechanism is used to obtain the session representation. The specific steps are as follows:

[0069] S301, calculating the average value of the item representations, and representing the user's interaction information as the average value of the item representations;

[0070] S302: Use the attention mechanism to obtain the weight of each item, and linearly combine the weight of each item with the item representation to calculate the session representation.

[0071] In this embodiment, the generated project representation A soft attention mechanism is used to obtain session representations for discrimination and a harvester is used to combine the individual contributions of each item in the current session into the prediction of the next item, as follows:

[0072] The user's interaction information is expressed as the average value of the item representation, and the expression is as follows:

[0073]

[0074] in, represents the user's i-th interaction information, m represents the total number of items, represents the i-th item representation;

[0075] The attention mechanism is used to obtain the weight of each item. The calculation expression is as follows:

[0076]

[0077] Where i represents the i-th item in the session, δ i represents the weight of the i-th item, represents the i-th interaction information of user in session u, represents the i-th item representation of session u, b1, W1, and W2 all represent learnable parameters; b1∈R d ,W1,W2∈R d×d , R d represents a set of points in d-dimensional real space, R d×d represents a set of matrices of d×d real numbers, where d represents the dimension;

[0078] And use linear combination to obtain the session representation of session u. The calculation expression is as follows:

[0079]

[0080] Among them, R u represents the session representation of session u, n represents the length of the session, i represents the i-th item in the session, δ irepresents the weight of the i-th item, represents the i-th item representation of session u.

[0081] In this embodiment, U represents a session set, V = {v1, v1, ..., v m} represents the unique set derived from the session set, m represents the total number of items, represents a specific session, u∈U represents a session, n represents the session length, represents the interaction term at time step p, Represents a vectorized session, which is combined into a matrix form as the project embedding matrix E i , by a specific session Embedded into vector space.

[0082] S4. Based on the session representation and the item embedding matrix, the one-hot encoding vector of the ground truth is calculated. Based on the one-hot encoding vector of the ground truth and the individual contribution of each item, combined with the current item after the user inputs the session, the next item is predicted to complete the session recommendation.

[0083] In this example, the session representation and the item embedding matrix are multiplied, and softmax is used to scale the result. The calculation formula is as follows:

[0084] y i =softmax((R u ) T E i )

[0085] Among them, y i represents the one-hot encoded vector of the ground truth, softmax(·) represents the normalized exponential function, and R u represents the session representation of session u, E i represents the item embedding matrix;

[0086] The calculated one-hot encoding is used to determine the user's most likely next target item. Based on the one-hot encoding vector of the ground truth and the individual contribution of each item, and combined with the current item after the user enters the session, the next item is predicted to complete the session recommendation based on the denoising edge perception network.

Claims

1. A conversation recommendation method based on a denoising edge-aware network, characterized in that: The following steps are involved: S1. Mining all items and their hop-order neighbors in the historical conversation data, and constructing a frequency perception graph containing each item using the co-occurrence frequency of each item and its hop-order neighbors; S2. According to the information carried by the edge in the frequency perception graph, the item embedding matrix is ​​obtained, and the learning parameters are assigned to the edge using the denoising edge perception network to obtain the item representation learning and obtain the item representation through calculation; S3. Based on the item representation, the soft attention mechanism is used to obtain the session representation; S4. Based on the session representation and the item embedding matrix, the one-hot encoding vector of the ground truth is calculated. Based on the one-hot encoding vector of the ground truth and the individual contribution of each item, combined with the current item after the user inputs the session, the next item is predicted to complete the session recommendation.

2. The conversation recommendation method based on denoising edge perception network according to claim 1, characterized in that: The S1 comprises the following steps: S101, mining items and their hopping neighbors in all historical conversation data, combining items and their hopping neighbors into adjacent item pairs, calculating the number of adjacent item pairs in all historical conversation data, and using the number of adjacent item pairs as the weight of the edge in the frequency perception graph; S102, normalizing the edge weight of each item according to the edge weight to obtain a frequency representation of the edge weight of each item; S103, classifying edge types according to the frequency representation of edge weights of each item, and combining the frequency representation of edge weights of hopping neighbors of each item to obtain the co-occurrence frequency of each item and its hopping neighbors, and constructing a frequency perception map containing each item.

3. The conversation recommendation method based on denoising edge perception network according to claim 1, characterized in that: The S2 comprises the following steps: S201, obtaining a specific input session according to information carried by the edge in the frequency perception graph, initializing the vector form of the input session to obtain a vectorized session, and combining the vectorized sessions to obtain a project embedding matrix; S202, using the edge-aware graph attention subnetwork in the denoising edge-aware network to assign learnable parameters to the edge; S203, using vectorized sessions and learnable parameters combined with leaky rectified linear units, calculating the attention score in the semantic space from item to item neighbor, using a self-attention mechanism based on an α-entmax activation function, and using the attention score to calculate a sparse weight score; S204, obtaining the final attention weight according to the sparse weight score of each item; S205. Linearly combine the final attention weight and the vectorized session associated with the final attention weight to calculate the item representation.

4. The conversation recommendation method based on denoising edge perception network according to claim 1, characterized in that: The S3 comprises the following steps: S301, calculating the average value of the item representations, and representing the user's interaction information as the average value of the item representations; S302: Use the attention mechanism to obtain the weight of each item, and linearly combine the weight of each item with the item representation to calculate the session representation.

5. The conversation recommendation method based on denoising edge perception network according to claim 4 is characterized in that: The expression of the session representation is as follows: Among them, R u represents the session representation of session u, n represents the length of the session, i represents the i-th item in the session, δ i represents the weight of the i-th item, represents the i-th item representation of session u, represents the i-th interaction information of user in session u, b1, W1, and W2 all represent learnable parameters; W1,W2∈R d×d , R d represents a set of points in d-dimensional real space, R d×d Represents a set of d×d matrices of real numbers, where d represents the dimension.

6. The conversation recommendation method based on denoising edge perception network according to claim 1, characterized in that: The expression of the one-hot encoded vector of the ground truth is as follows: and i =softmax((R u ) T AND i ) Among them, y i represents the one-hot encoded vector of the ground truth, softmax(·) represents the normalized exponential function, and R u represents the session representation of session u, E i Represents the item embedding matrix.

Images