Project recommendation method, device, equipment and storage medium

Abstract

The present application relates to artificial intelligence and digital medical treatment, and provides a project recommendation method, device, equipment and storage medium. The method generates an interaction matrix based on the interaction information of an initial user with an initial project, generates a user walk sequence and a project walk sequence based on the interaction matrix, inputs the user walk sequence into a user representation model to obtain a user representation vector, processes the project walk sequence based on a project representation model to obtain a project representation vector, generates a user context vector based on the multidimensional information of the target user, generates a project context vector based on the multidimensional features of each initial project, generates an interested project vector based on the project representation vector and the project context vector, and accurately generates a target project based on the user context vector, the user representation vector and the interested project vector. In addition, the present application also relates to blockchain technology, and the target project can be stored in the blockchain.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and digital medical technology, and in particular to a project recommendation method, apparatus, device and storage medium. Background Technology

[0002] In the fintech sector, e-commerce is becoming increasingly prevalent, and with the development of artificial intelligence, proactively recommending products to users based on their preferences has become the mainstream approach. However, current product recommendation solutions typically analyze users based on static characteristics. As user data changes over time, it becomes difficult to accurately determine user preferences, leading to inaccurate product recommendations. Summary of the Invention

[0003] In view of the above, it is necessary to provide a project recommendation method, apparatus, device, and storage medium that can solve the technical problem of how to improve the accuracy of project recommendation generation.

[0004] On the one hand, the present invention proposes a project recommendation method, the project recommendation method comprising:

[0005] An interaction matrix is ​​generated based on the interaction information of initial users with multiple initial items, wherein the initial users include target users;

[0006] Based on the interaction matrix, a user walk sequence is generated for each initial user, and a project walk sequence is generated for each initial project.

[0007] The user walking sequence is input into a pre-trained user representation model to obtain a user representation vector, and the item walking sequence is processed based on a pre-trained item representation model to obtain an item representation vector.

[0008] A user context vector is generated based on the multi-dimensional information of the target user, and a project context vector is generated based on the multi-dimensional features of each initial project.

[0009] Generate the target user's interest item vector based on the item representation vector and the item context vector;

[0010] The target items of the target user are generated based on the user context vector, the user representation vector, and the item of interest vector.

[0011] According to a preferred embodiment of the present invention, generating the user walk sequence for each initial user based on the interaction matrix includes:

[0012] A user graph is constructed based on the interaction matrix, and the user graph includes user nodes of multiple initial users;

[0013] Starting from any user node, the probability of any user node moving to any of the other user nodes is calculated based on the node distance between the user node and the other user nodes in the user graph and the interaction matrix.

[0014] Compare the walk probability with the maximum value with the preset probability threshold;

[0015] If the maximum walk probability is greater than or equal to the preset probability threshold, then the remaining user nodes corresponding to the maximum walk probability are taken as the next walk node of any user node, and the initial user sequence is obtained.

[0016] When the number of nodes in the initial user sequence is greater than or equal to a preset number, the initial user sequence is determined as the user roaming sequence.

[0017] According to a preferred embodiment of the present invention, calculating the probability of any user node moving to the other user nodes based on the node distances between any user node and the other user nodes in the user graph and the interaction matrix includes:

[0018] Extract a first vector corresponding to any user node and a second vector corresponding to the remaining user nodes from the interaction matrix;

[0019] Calculate the user similarity between any user node and the other user nodes based on the first vector and the second vector;

[0020] The wander probability is calculated based on the node distance and the user similarity. The formula for calculating the wander probability is as follows:

[0021]

[0022] in, This represents the walk probability. Let d represent the user similarity, d represent the node distance, p>0, q>0.

[0023] According to a preferred embodiment of the present invention, before inputting the user walk sequence into a pre-trained user representation model to obtain a user representation vector, the method further includes:

[0024] Multiple user walk sequences are input into a pre-constructed learning network to obtain the prediction vector and prediction bias for each user walk sequence;

[0025] Based on the prediction vector, the prediction bias, and the number of times any two user nodes co-occur in multiple user walk sequences, calculate the co-occurrence loss value of any two user nodes;

[0026] Calculate the sum of the multiple co-occurrence loss values ​​to obtain the total loss value;

[0027] The learning network is adjusted until the total loss value converges, thus obtaining the user representation model.

[0028] According to a preferred embodiment of the present invention, the multi-dimensional information includes discrete information, continuous information, text information, and image information of the target user, and the generation of a user context vector based on the multi-dimensional information of the target user includes:

[0029] The discrete information is vector-encoded to obtain discrete features;

[0030] The continuous information is normalized to obtain continuous features;

[0031] Feature extraction is performed on the text information to obtain text features;

[0032] The image information is convolved to obtain image features;

[0033] Feature interaction processing is performed on the discrete features, continuous features, text features, and image features to obtain the user context vector. The formula for calculating the user context vector is as follows:

[0034]

[0035] Among them, e c Let L represent the user context vector, L represent the total number of features among the discrete features, the continuous features, the text features, and the image features, and x represent the total number of features among the discrete features, the continuous features, the text features, and the image features. i and x i′ Each of the discrete features, continuous features, text features, and image features represents any one of them. and Let b represent the transpose weights of any of the features. i and b i′ Each of these represents the bias of any of the aforementioned features.

[0036] According to a preferred embodiment of the present invention, generating the target user's interest item vector based on the item representation vector and the item context vector includes:

[0037] Based on the interaction information, the first interaction item for the target user is selected from the plurality of initial items;

[0038] Identify similar users of the target user based on the user representation vector;

[0039] The project representation vector and the project context vector of the first interactive project are fused to obtain the first fused feature.

[0040] The item representation vector and the item context vector of the second interactive project of the similar users are fused to obtain the second fused feature;

[0041] The first fusion feature and the second fusion feature are interactively processed to obtain the vector of the item of interest.

[0042] According to a preferred embodiment of the present invention, generating the target items of the target user based on the user context vector, the user representation vector, and the item of interest vector includes:

[0043] Interactive features are obtained by interactively processing the user context vector, the user representation vector, and the item of interest vector.

[0044] Attention features are obtained by performing attention analysis on the interaction features.

[0045] The attention features are classified and predicted to obtain the target item.

[0046] On the other hand, the present invention also proposes a project recommendation device, the project recommendation device comprising:

[0047] A generation unit is used to generate an interaction matrix based on the interaction information of an initial user with multiple initial items, wherein the initial user includes a target user;

[0048] The generation unit is further configured to generate a user walk sequence for each initial user based on the interaction matrix, and to generate a project walk sequence for each initial project based on the interaction matrix;

[0049] The input unit is used to input the user walking sequence into the pre-trained user representation model to obtain the user representation vector, and to process the item walking sequence based on the pre-trained item representation model to obtain the item representation vector.

[0050] The generation unit is also used to generate a user context vector based on the multi-dimensional information of the target user, and to generate a project context vector based on the multi-dimensional features of each initial project;

[0051] The generation unit is further configured to generate the target user's interest item vector based on the item representation vector and the item context vector;

[0052] The generation unit is further configured to generate the target items of the target user based on the user context vector, the user representation vector, and the item of interest vector.

[0053] On the other hand, the present invention also proposes an electronic device, the electronic device comprising:

[0054] Memory, which stores computer-readable instructions; and

[0055] The processor executes computer-readable instructions stored in the memory to implement the recommended method of the project.

[0056] On the other hand, the present invention also proposes a computer-readable storage medium storing computer-readable instructions that are executed by a processor in an electronic device to implement the recommended method for the project.

[0057] As can be seen from the above technical solutions, this application generates the user walk sequence and the item walk sequence through the interaction matrix, and then accurately generates the corresponding user representation vector and item representation vector through the user walk sequence and the item walk sequence. Furthermore, the item context vector generated by the multi-dimensional features can be combined with the item representation vector to generate the target user's interest item vector. Furthermore, the user context vector and the user representation vector generated by the multi-dimensional information can accurately generate the target items that the target user is actually interested in. In addition, since this application analyzes the target user in conjunction with similar users, it can accurately locate the target user's preference direction, thereby further improving the accuracy of the target items. Simultaneously, this application can generate the target items by combining the interaction information of the initial user with the initial items. Since it is not necessary to extract and analyze the interaction information of all recommended items, the generation efficiency of the target items can be improved. Attached Figure Description

[0058] Figure 1 This is a flowchart of a preferred embodiment of the method recommended in this invention.

[0059] Figure 2 This is a schematic diagram of the user map in the project recommendation method of this invention.

[0060] Figure 3 This is a functional block diagram of a preferred embodiment of the device recommended in this invention.

[0061] Figure 4 This is a schematic diagram of the structure of an electronic device that implements the project recommendation method of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0063] like Figure 1 The diagram shown is a flowchart of a preferred embodiment of the method recommended by this invention. The order of the steps in this flowchart can be changed, and some steps can be omitted, depending on different requirements.

[0064] The project recommendation method described above can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0065] Foundational artificial intelligence technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0066] The recommended method described in the project is applied to one or more electronic devices. The electronic device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored computer-readable instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0067] The electronic device can be any electronic product that can interact with the user, such as a personal computer, tablet computer, smartphone, personal digital assistant (PDA), game console, interactive network television (IPTV), smart wearable device, etc.

[0068] The electronic devices may include network devices and / or user devices. The network devices include, but are not limited to, single network electronic devices, groups of multiple network electronic devices, or cloud computing-based systems consisting of a large number of hosts or network electronic devices.

[0069] The network in which the electronic device is located includes, but is not limited to: the Internet, wide area network, metropolitan area network, local area network, virtual private network (VPN), etc.

[0070] 101. Generate an interaction matrix based on the interaction information of the initial user with multiple initial items, wherein the initial user includes the target user.

[0071] In at least one embodiment of the present invention, the initial user can be any user who has browsed, clicked, or purchased any initial item at any point in history, and the initial user includes target users who need item recommendations. The initial user can also be a patient user in the field of digital healthcare, and the initial user includes target users who need healthcare item recommendations.

[0072] The plurality of initial projects may include project products in the fintech field, project products under the e-commerce sales model, and health care projects in the digital healthcare field, etc. This application does not limit the specific products of the plurality of initial projects.

[0073] The interactive information includes the initial user's browsing, clicking, and purchasing actions on any initial item.

[0074] The interaction matrix is ​​used to characterize the initial user's interactions with the multiple initial items.

[0075] In at least one embodiment of the present invention, the electronic device generates an interaction matrix based on the interaction information of an initial user with multiple initial items, including:

[0076] Obtain the project operation information of the initial user for each initial project;

[0077] The interaction matrix is ​​constructed based on the project operation information.

[0078] For example, the number of initial users is N, and the number of initial projects is M. If an initial user interacts with any initial project, the matrix element corresponding to that initial user and that initial project is 1; if an initial user does not interact with any initial project, the matrix element corresponding to that initial user and that initial project is 0. The resulting interaction matrix is ​​Y∈{0,1}.N×M .

[0079] 102. Generate a user walk sequence for each initial user based on the interaction matrix, and generate a project walk sequence for each initial project based on the interaction matrix.

[0080] In at least one embodiment of the present invention, each user walk sequence starts from the node of the corresponding initial user, and the nodes of the remaining users are the sequence of walk nodes. For example, the user walk sequence for the initial user A can be ACDRGK.

[0081] Each item walk sequence starts with the node of the corresponding initial item, and the nodes of the remaining items form a sequence of walk nodes. For example, the user walk sequence for the initial item 1 could be 1-5-3-6-7-2.

[0082] In at least one embodiment of the present invention, the electronic device generates a user walk sequence for each initial user based on the interaction matrix, including:

[0083] A user graph is constructed based on the interaction matrix, and the user graph includes user nodes of multiple initial users;

[0084] Starting from any user node, the probability of any user node moving to any of the other user nodes is calculated based on the node distance between the user node and the other user nodes in the user graph and the interaction matrix.

[0085] Compare the walk probability with the maximum value with the preset probability threshold;

[0086] If the maximum walk probability is greater than or equal to the preset probability threshold, then the remaining user nodes corresponding to the maximum walk probability are taken as the next walk node of any user node, and the initial user sequence is obtained.

[0087] When the number of nodes in the initial user sequence is greater than or equal to a preset number, the initial user sequence is determined as the user roaming sequence.

[0088] The user graph includes user nodes corresponding to multiple initial users, and the node distance between any two user nodes. For example... Figure 2 As shown, Figure 2 This is a schematic diagram of the user map in the recommendation method of this invention. Figure 2In the graph, the user map includes user node A, user node B, user node C, user node D, and user node E. The node distance between user node A and user node B is 2, the node distance between user node A and user node C is 1, the node distance between user node A and user node D is 1, the node distance between user node A and user node E is 2, the node distance between user node B and user node C is 3, the node distance between user node B and user node D is 1, the node distance between user node B and user node E is 4, the node distance between user node C and user node D is 2, the node distance between user node C and user node E is 1, and the node distance between user node D and user node E is 3.

[0089] The preset probability threshold and the preset quantity can be set according to actual needs. For example, the preset probability threshold can be 0.7 and the preset quantity can be 6, etc.

[0090] The user graph can be accurately constructed using the interaction matrix. The node distances can then be accurately obtained from the user graph. Based on the node distances and the interaction matrix, the probability of any user node moving to other user nodes can be accurately quantified. By comparing the maximum probability with a preset probability threshold, and comparing the number of nodes in the initial user sequence with a preset number, the length of the user movement sequence can be controlled as needed, thus preventing the sequence from becoming too long and affecting the efficiency of item recommendations.

[0091] Specifically, the electronic device constructs a user graph based on the interaction matrix, including:

[0092] The number of joint operations performed by any initial user and the other initial users on the multiple initial items is counted based on the interaction matrix.

[0093] A similarity ratio between any initial user and the remaining initial users is generated based on the number of common operations and the number of projects in the plurality of initial projects.

[0094] Calculate the node distance between any initial user and the other initial users based on the similarity ratio;

[0095] The user graph is constructed by using multiple initial users as user nodes, where the node distance is the distance between the user node corresponding to any initial user and the user nodes corresponding to the other initial users.

[0096] The number of joint operations refers to the number of projects that any initial user and the other initial users interact with simultaneously.

[0097] The node distance can be a value obtained by multiplying the reciprocal of the similarity ratio. For example, if the similarity ratio between initial user A and initial user C is 3 / 5, the similarity ratio between initial user A and initial user D is 2 / 5, and the similarity ratio between initial user C and initial user D is 1 / 5, after multiplying the reciprocal of the similarity ratio, the node distance between initial user A and initial user C is 2, the node distance between initial user A and initial user D is 3, and the node distance between initial user C and initial user D is 6. Simultaneously, the node distance indicates the distance between the user node corresponding to any initial user and the user nodes corresponding to the other initial users in the user graph. Continuing the above example, initial user A and initial user C are 2 nodes apart in the user graph, initial user A and initial user D are 3 nodes apart, and initial user C and initial user D are 6 nodes apart.

[0098] The interaction matrix allows for a direct statistical analysis of the number of common operations performed by any initial user and other initial users on the multiple initial items. Based on the number of common operations and the number of items, the similarity ratio can be accurately quantified, thereby enabling the accurate generation of the node distance and improving the accuracy of the user graph construction.

[0099] Specifically, the electronic device calculates the probability of any user node moving to any of the other user nodes based on the node distances between any user node and the other user nodes in the user graph and the interaction matrix, including:

[0100] Extract a first vector corresponding to any user node and a second vector corresponding to the remaining user nodes from the interaction matrix;

[0101] Calculate the user similarity between any user node and the other user nodes based on the first vector and the second vector;

[0102] The wander probability is calculated based on the node distance and the user similarity. The formula for calculating the wander probability is as follows:

[0103]

[0104] in, This represents the walk probability. Let d represent the user similarity, d represent the node distance, p>0, q>0.

[0105] Wherein, the first vector is used to represent the interaction of the user corresponding to any user node with the multiple initial items, and the second vector is used to represent the interaction of the users corresponding to the other user nodes with the multiple initial items. The first vector and the second vector can be column vectors with M matrix elements.

[0106] The user similarity can be the dot product of the first vector and the second vector.

[0107] The cases where d≠0 and d≠1 include the cases where the node distance d≥2 between any user node and the other user nodes in the user graph, and the cases where any user node and the other user nodes are not connected in the user graph.

[0108] The interaction matrix allows for the rapid extraction of the first and second vectors, thereby improving the efficiency of generating user similarity. Furthermore, by combining the node distance and user similarity, the wander probability can be accurately quantified.

[0109] In at least one embodiment of the present invention, the method of generating the item walk sequence is similar to the method of generating the user walk sequence, and will not be described in detail here.

[0110] 103. Input the user walking sequence into the pre-trained user representation model to obtain the user representation vector, and process the item walking sequence based on the pre-trained item representation model to obtain the item representation vector.

[0111] In at least one embodiment of the present invention, the user representation model is used to perform vector representation of the initial user by combining the initial user's trips to multiple identical initial items. The user representation model is the learning network corresponding to the convergence of the total loss value.

[0112] The project representation model is used to perform vector representation of the multiple initial projects.

[0113] In at least one embodiment of the present invention, before inputting the user walk sequence into a pre-trained user representation model to obtain a user representation vector, the method further includes:

[0114] Multiple user walk sequences are input into a pre-constructed learning network to obtain the prediction vector and prediction bias for each user walk sequence;

[0115] Based on the prediction vector, the prediction bias, and the number of times any two user nodes co-occur in multiple user walk sequences, calculate the co-occurrence loss value of any two user nodes;

[0116] Calculate the sum of the multiple co-occurrence loss values ​​to obtain the total loss value;

[0117] The learning network is adjusted until the total loss value converges, thus obtaining the user representation model.

[0118] The learning network includes weight matrices corresponding to multiple different node distances.

[0119] The co-occurrence count refers to the number of times any two user nodes appear simultaneously in multiple user walk sequences.

[0120] Specifically, the formula for calculating the co-occurrence loss value is as follows:

[0121]

[0122] Among them, L GE U represents the co-occurrence loss value. n u l Represents any two user nodes. Let represent the prediction vectors corresponding to any two user nodes. This represents the prediction bias corresponding to any two user nodes. This indicates the number of times the two events occur simultaneously.

[0123] By combining the prediction vector, the prediction bias, and the number of co-occurrences, the co-occurrence loss value of any two user nodes can be accurately quantified, thereby improving the accuracy of the total loss value. Then, the learning network can be adjusted using the total loss value to improve the accuracy of the user representation model for each initial user.

[0124] In at least one embodiment of the present invention, the training method of the project representation model is similar to that of the user representation model, and will not be described in detail here.

[0125] 104. Generate a user context vector based on the multi-dimensional information of the target user, and generate a project context vector based on the multi-dimensional features of each initial project.

[0126] In at least one embodiment of the present invention, the multi-dimensional information includes discrete information, continuous information, text information, and image information of the target user. The multi-dimensional information includes the target user's physical examination results, etc., wherein the discrete information includes the target user's gender, the continuous information includes the target user's age, the text information includes the target user's evaluation information, such as the target user's evaluation of a certain health care program, and the image information includes the target user's facial image, the target user's CT image, etc.

[0127] The multi-dimensional features include feature information such as the project type of the initial project.

[0128] In at least one embodiment of the present invention, the electronic device generates a user context vector based on the multi-dimensional information of the target user, including:

[0129] The discrete information is vector-encoded to obtain discrete features;

[0130] The continuous information is normalized to obtain continuous features;

[0131] Feature extraction is performed on the text information to obtain text features;

[0132] The image information is convolved to obtain image features;

[0133] Feature interaction processing is performed on the discrete features, continuous features, text features, and image features to obtain the user context vector. The formula for calculating the user context vector is as follows:

[0134]

[0135] Among them, e c Let L represent the user context vector, L represent the total number of features among the discrete features, the continuous features, the text features, and the image features, and x represent the total number of features among the discrete features, the continuous features, the text features, and the image features. i and x i′ Each of the discrete features, continuous features, text features, and image features represents any one of them. and Let b represent the transpose weights of any of the features. i and b i′ Each of these represents the bias of any of the aforementioned features.

[0136] The transpose weight and the bias can be set according to actual needs.

[0137] By processing the multi-dimensional information in different ways, the corresponding feature information can be accurately generated. Then, feature interaction processing can be performed on the discrete features, the continuous features, the text features, and the image features, which can improve the accuracy of the user context vector.

[0138] In at least one embodiment of the present invention, the method of generating the project context vector is similar to the method of generating the user context vector, and will not be described in detail here.

[0139] 105. Generate the target user's interest item vector based on the item representation vector and the item context vector.

[0140] In at least one embodiment of the present invention, the item of interest vector is used to indicate the vector corresponding to the item of interest of the target user. The item corresponding to the item of interest vector can be any item among the plurality of initial items, or the item corresponding to the item of interest vector can be a different item from the plurality of initial items.

[0141] In at least one embodiment of the present invention, the electronic device generates the target user's interest item vector based on the item representation vector and the item context vector, including:

[0142] Based on the interaction information, the first interaction item for the target user is selected from the plurality of initial items;

[0143] Identify similar users of the target user based on the user representation vector;

[0144] The project representation vector and the project context vector of the first interactive project are fused to obtain the first fused feature.

[0145] The item representation vector and the item context vector of the second interactive project of the similar users are fused to obtain the second fused feature;

[0146] The first fusion feature and the second fusion feature are interactively processed to obtain the vector of the item of interest.

[0147] The first interactive item refers to the initial item for the target user to perform interactive operations such as clicking.

[0148] The first fusion feature refers to a feature that simultaneously includes the project representation vector and the project context vector of the first interactive project.

[0149] The second interactive item refers to the initial item for similar users to perform interactive operations such as clicking.

[0150] The second fusion feature refers to a feature that simultaneously includes the project representation vector and the project context vector of the second interactive project.

[0151] By combining the first interaction items of the target user and the second interaction items of similar users, the vector of items of interest can be comprehensively and accurately quantified.

[0152] Specifically, the user representation vector includes an initial representation vector and a target representation vector of the target user. The electronic device identifies similar users of the target user based on the user representation vector, including:

[0153] Based on the target user, extract the target representation vector from the user representation vector;

[0154] Calculate the dot product of the target representation vector and the initial representation vector;

[0155] The initial user corresponding to the initial representation vector whose value is greater than the preset result is determined as the similar user.

[0156] The preset result can be set according to actual needs.

[0157] Specifically, the electronic device performs vector fusion on the item representation vector and the item context vector of the first interactive item to obtain the first fused feature, which includes:

[0158] Obtain the first set weight corresponding to the project representation vector and the second set weight corresponding to the project context vector;

[0159] Based on the first set weight and the second set weight, the project representation vector and the project context vector are weighted and summed to obtain the second fusion feature.

[0160] In other embodiments, the generation method of the second fusion feature is similar to that of the first fusion feature, and will not be described in detail here.

[0161] In at least one embodiment of the present invention, the way in which the electronic device performs interactive processing on the first fused feature and the second fused feature to obtain the vector of interest is similar to the way in which the electronic device performs feature interactive processing on the discrete feature, the continuous feature, the text feature and the image feature to obtain the user context vector, and this application will not elaborate further on this.

[0162] 106. Generate the target items of the target user based on the user context vector, the user representation vector, and the item of interest vector.

[0163] It should be emphasized that, to further ensure the privacy and security of the aforementioned target projects, they can also be stored in a node of a blockchain.

[0164] In at least one embodiment of the present invention, the target item is a recommended item for the target user. The target item can be any one of the plurality of initial items, or it can be an item that is different from all of the plurality of initial items. The target item can be a healthcare item that needs to be recommended in the field of digital healthcare.

[0165] In at least one embodiment of the present invention, the electronic device generates the target items of the target user based on the user context vector, the user representation vector, and the item of interest vector, including:

[0166] Interactive features are obtained by interactively processing the user context vector, the user representation vector, and the item of interest vector.

[0167] Attention features are obtained by performing attention analysis on the interaction features.

[0168] The attention features are classified and predicted to obtain the target item.

[0169] By performing interactive processing on the user context vector, the user representation vector, and the interest item vector, the feature comprehensiveness of the interaction features can be improved. Furthermore, by performing attention analysis on the interaction features, the feature concentration of the attention features can be improved, thereby enhancing the prediction accuracy of the target item.

[0170] Specifically, the way in which the electronic device performs interactive processing on the user context vector, the user representation vector, and the item of interest vector to obtain interactive features is similar to the way in which the electronic device performs feature interactive processing on the discrete features, the continuous features, the text features, and the image features to obtain the user context vector, and this application will not elaborate further on this.

[0171] Specifically, the electronic device performs classification and prediction processing on the attention features to obtain the target items, including:

[0172] The attention features are activated to obtain activation features;

[0173] The activation features are mapped to obtain the target item.

[0174] This application is applied in the field of digital healthcare. For example, when the initial users include patient user A, patient user B, patient user C, and patient user D, and assuming the target user is patient user A, the multiple initial items are health care items that the multiple initial users have interacted with. The multiple initial items may include health care item 1, health care item 2, and health care item 3. The interaction information includes the browsing information and purchase records of the multiple initial users on the health care items. Through this application, health care item 2 can be generated to be recommended to patient user A.

[0175] As can be seen from the above technical solutions, this application generates the user walk sequence and the item walk sequence through the interaction matrix, and then accurately generates the corresponding user representation vector and item representation vector through the user walk sequence and the item walk sequence. Furthermore, the item context vector generated by the multi-dimensional features can be combined with the item representation vector to generate the target user's interest item vector. Furthermore, the user context vector and the user representation vector generated by the multi-dimensional information can accurately generate the target items that the target user is actually interested in. In addition, since this application analyzes the target user in conjunction with similar users, it can accurately locate the target user's preference direction, thereby further improving the accuracy of the target items. Simultaneously, this application can generate the target items by combining the interaction information of the initial user with the initial items. Since it is not necessary to extract and analyze the interaction information of all recommended items, the generation efficiency of the target items can be improved.

[0176] like Figure 3 The diagram shown is a functional block diagram of a preferred embodiment of the project recommendation device of the present invention. The project recommendation device 11 includes a generation unit 110, an input unit 111, a calculation unit 112, and an adjustment unit 113. The module / unit referred to in this invention is a series of computer-readable instruction segments that can be acquired by the processor 13 and perform a fixed function, and is stored in the memory 12. In this embodiment, the functions of each module / unit will be described in detail in subsequent embodiments.

[0177] The generation unit 110 is used to generate an interaction matrix based on the interaction information of an initial user with multiple initial items, wherein the initial user includes the target user;

[0178] The generation unit 110 is further configured to generate a user walk sequence for each initial user based on the interaction matrix, and to generate a project walk sequence for each initial project based on the interaction matrix.

[0179] The input unit 111 is used to input the user walking sequence into the pre-trained user representation model to obtain the user representation vector, and to process the item walking sequence based on the pre-trained item representation model to obtain the item representation vector.

[0180] The generation unit 110 is also used to generate a user context vector based on the multi-dimensional information of the target user, and to generate a project context vector based on the multi-dimensional features of each initial project.

[0181] The generation unit 110 is further configured to generate the target user's interest item vector based on the item representation vector and the item context vector;

[0182] The generation unit 110 is further configured to generate the target items of the target user based on the user context vector, the user representation vector, and the item of interest vector.

[0183] In at least one embodiment of the present invention, the generation unit 110 is further configured to construct a user graph based on the interaction matrix, wherein the user graph includes user nodes of a plurality of the initial users;

[0184] Starting from any user node, the probability of any user node moving to any of the other user nodes is calculated based on the node distance between the user node and the other user nodes in the user graph and the interaction matrix.

[0185] Compare the walk probability with the maximum value with the preset probability threshold;

[0186] If the maximum walk probability is greater than or equal to the preset probability threshold, then the remaining user nodes corresponding to the maximum walk probability are taken as the next walk node of any user node, and the initial user sequence is obtained.

[0187] When the number of nodes in the initial user sequence is greater than or equal to a preset number, the initial user sequence is determined as the user roaming sequence.

[0188] In at least one embodiment of the present invention, the generation unit 110 is further configured to extract a first vector corresponding to any user node and a second vector corresponding to the remaining user nodes from the interaction matrix;

[0189] Calculate the user similarity between any user node and the other user nodes based on the first vector and the second vector;

[0190] The wander probability is calculated based on the node distance and the user similarity. The formula for calculating the wander probability is as follows:

[0191]

[0192] in, This represents the walk probability. Let d represent the user similarity, d represent the node distance, p>0, q>0.

[0193] In at least one embodiment of the present invention, the input unit 111 is further configured to input a plurality of user walk sequences into a pre-constructed learning network to obtain a prediction vector and a prediction bias for each user walk sequence;

[0194] The calculation unit 112 is used to calculate the co-occurrence loss value of any two user nodes based on the prediction vector, the prediction bias, and the number of times any two user nodes co-occur in multiple user walk sequences.

[0195] The calculation unit 112 is also used to calculate the sum of multiple co-occurrence loss values ​​to obtain a total loss value;

[0196] The adjustment unit 113 is used to adjust the learning network until the total loss value converges to obtain the user representation model.

[0197] In at least one embodiment of the present invention, the generation unit 110 is further configured to perform vector encoding on the discrete information to obtain discrete features;

[0198] The continuous information is normalized to obtain continuous features;

[0199] Feature extraction is performed on the text information to obtain text features;

[0200] The image information is convolved to obtain image features;

[0201] Feature interaction processing is performed on the discrete features, continuous features, text features, and image features to obtain the user context vector. The formula for calculating the user context vector is as follows:

[0202]

[0203] Among them, e c Let L represent the user context vector, L represent the total number of features among the discrete features, the continuous features, the text features, and the image features, and x represent the total number of features among the discrete features, the continuous features, the text features, and the image features. i and x i′ Each of the discrete features, continuous features, text features, and image features represents any one of them. and Let b represent the transpose weights of any of the features. i and bi′ Each of these represents the bias of any of the aforementioned features.

[0204] In at least one embodiment of the present invention, the generation unit 110 is further configured to select the first interactive item of the target user from the plurality of initial items based on the interactive information;

[0205] Identify similar users of the target user based on the user representation vector;

[0206] The project representation vector and the project context vector of the first interactive project are fused to obtain the first fused feature.

[0207] The item representation vector and the item context vector of the second interactive project of the similar users are fused to obtain the second fused feature;

[0208] The first fusion feature and the second fusion feature are interactively processed to obtain the vector of the item of interest.

[0209] In at least one embodiment of the present invention, the generation unit 110 is further configured to perform interactive processing on the user context vector, the user representation vector and the interest item vector to obtain interactive features;

[0210] Attention features are obtained by performing attention analysis on the interaction features.

[0211] The attention features are classified and predicted to obtain the target item.

[0212] As can be seen from the above technical solutions, this application generates the user walk sequence and the item walk sequence through the interaction matrix, and then accurately generates the corresponding user representation vector and item representation vector through the user walk sequence and the item walk sequence. Furthermore, the item context vector generated by the multi-dimensional features can be combined with the item representation vector to generate the target user's interest item vector. Furthermore, the user context vector and the user representation vector generated by the multi-dimensional information can accurately generate the target items that the target user is actually interested in. In addition, since this application analyzes the target user in conjunction with similar users, it can accurately locate the target user's preference direction, thereby further improving the accuracy of the target items. Simultaneously, this application can generate the target items by combining the interaction information of the initial user with the initial items. Since it is not necessary to extract and analyze the interaction information of all recommended items, the generation efficiency of the target items can be improved.

[0213] like Figure 4The diagram shown is a schematic representation of the structure of an electronic device that implements the project recommendation method of the present invention.

[0214] In one embodiment of the present invention, the electronic device 1 includes, but is not limited to, a memory 12, a processor 13, and computer-readable instructions stored in the memory 12 and executable on the processor 13, such as a project recommendation program.

[0215] Those skilled in the art will understand that the schematic diagram is merely an example of electronic device 1 and does not constitute a limitation on electronic device 1. It may include more or fewer components than shown in the diagram, or combine certain components, or different components. For example, electronic device 1 may also include input / output devices, network access devices, buses, etc.

[0216] The processor 13 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 13 is the computing core and control center of the electronic device 1, connecting various parts of the electronic device 1 through various interfaces and lines, and executing the operating system of the electronic device 1, as well as various installed application programs and program code.

[0217] For example, the computer-readable instructions can be divided into one or more modules / units, which are stored in the memory 12 and executed by the processor 13 to complete the present invention. The one or more modules / units can be a series of computer-readable instruction segments capable of performing a specific function, which describe the execution process of the computer-readable instructions in the electronic device 1. For example, the computer-readable instructions can be divided into a generation unit 110, an input unit 111, a calculation unit 112, and an adjustment unit 113.

[0218] The memory 12 can be used to store the computer-readable instructions and / or modules. The processor 13 implements various functions of the electronic device 1 by running or executing the computer-readable instructions and / or modules stored in the memory 12 and calling the data stored in the memory 12. The memory 12 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. The memory 12 may include non-volatile and volatile memory, such as: hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other storage devices.

[0219] The memory 12 can be the external memory and / or internal memory of the electronic device 1. Furthermore, the memory 12 can be a physical memory, such as a memory module, a TF card (Trans-flash Card), etc.

[0220] If the modules / units integrated in the electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by instructing related hardware through computer-readable instructions. The computer-readable instructions can be stored in a computer-readable storage medium, and when executed by a processor, the computer-readable instructions can implement the steps of the various method embodiments described above.

[0221] The computer-readable instructions include computer-readable instruction code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer-readable instruction code, recording medium, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), and random access memory (RAM).

[0222] The blockchain referred to in this invention is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0223] Combination Figure 1 The memory 12 in the electronic device 1 stores computer-readable instructions to implement a project recommendation method, and the processor 13 can execute the computer-readable instructions to achieve the following:

[0224] An interaction matrix is ​​generated based on the interaction information of initial users with multiple initial items, wherein the initial users include target users;

[0225] Based on the interaction matrix, a user walk sequence is generated for each initial user, and a project walk sequence is generated for each initial project.

[0226] The user walking sequence is input into a pre-trained user representation model to obtain a user representation vector, and the item walking sequence is processed based on a pre-trained item representation model to obtain an item representation vector.

[0227] A user context vector is generated based on the multi-dimensional information of the target user, and a project context vector is generated based on the multi-dimensional features of each initial project.

[0228] Generate the target user's interest item vector based on the item representation vector and the item context vector;

[0229] The target items of the target user are generated based on the user context vector, the user representation vector, and the item of interest vector.

[0230] Specifically, the specific implementation method of the processor 13 for the above-mentioned computer-readable instructions can be found in [reference needed]. Figure 1 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0231] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0232] The computer-readable storage medium stores computer-readable instructions, which, when executed by the processor 13, are used to perform the following steps:

[0233] An interaction matrix is ​​generated based on the interaction information of initial users with multiple initial items, wherein the initial users include target users;

[0234] Based on the interaction matrix, a user walk sequence is generated for each initial user, and a project walk sequence is generated for each initial project.

[0235] The user walking sequence is input into a pre-trained user representation model to obtain a user representation vector, and the item walking sequence is processed based on a pre-trained item representation model to obtain an item representation vector.

[0236] A user context vector is generated based on the multi-dimensional information of the target user, and a project context vector is generated based on the multi-dimensional features of each initial project.

[0237] Generate the target user's interest item vector based on the item representation vector and the item context vector;

[0238] The target items of the target user are generated based on the user context vector, the user representation vector, and the item of interest vector.

[0239] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0240] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0241] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0242] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or devices described may also be implemented by a single unit or device through software or hardware. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0243] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A project recommendation method characterized by comprising: The item recommendation method comprises: generating an interaction matrix based on interaction information of a plurality of initial items by an initial user, the initial user comprising a target user; generating a user walk sequence of each initial user based on the interaction matrix, and generating an item walk sequence of each initial item based on the interaction matrix; inputting the user walk sequence into a pre-trained user representation model to obtain a user representation vector, and processing the item walk sequence based on a pre-trained item representation model to obtain an item representation vector; generate a user context vector based on multi-dimensional information of the target user, and generate an item context vector based on multi-dimensional features of each initial item, the multi-dimensional information including discrete information, continuous information, text information, and image information of the target user, the generating of the user context vector based on the multi-dimensional information of the target user including: vector encoding the discrete information to obtain discrete features; normalizing the continuous information to obtain continuous features; feature extraction on the text information to obtain text features; convolution processing on the image information to obtain image features; and feature interaction processing on the discrete features, the continuous features, the text features, and the image features to obtain the user context vector, a calculation formula of the user context vector being: ; wherein, represents the user context vector, represents a total number of features in the discrete features, the continuous features, the text features, and the image features, and respectively represent any feature in the discrete features, the continuous features, the text features, and the image features, and respectively represent a transpose weight of the any feature, and respectively represent a bias of the any feature. generating an interested item vector of the target user based on the item representation vector and the item context vector; generating a target item of the target user based on the user context vector, the user representation vector, and the interested item vector.

2. The item recommendation method according to claim 1, wherein The generating of the user walk sequence of each initial user based on the interaction matrix comprises: constructing a user graph according to the interaction matrix, the user graph comprising user nodes of a plurality of initial users; calculating a walk probability of any user node walking to the remaining user nodes according to the node distance of the any user node and the remaining user nodes in the user graph and the interaction matrix; comparing the maximum value of the walk probability with a preset probability threshold value; if the maximum value of the walk probability is greater than or equal to the preset probability threshold value, the remaining user node corresponding to the maximum value of the walk probability is taken as the next walk node of the any user node, and an initial user sequence is obtained; when the number of nodes in the initial user sequence is greater than or equal to a preset number, the initial user sequence is determined as the user walk sequence.

3. The item recommendation method according to claim 2, wherein The calculating of the walk probability of the any user node walking to the remaining user nodes according to the node distance of the any user node and the remaining user nodes in the user graph and the interaction matrix comprises: extracting a first vector corresponding to the any user node and a second vector corresponding to the remaining user nodes from the interaction matrix; calculating the user similarity of the any user node and the remaining user nodes according to the first vector and the second vector; calculating the walk probability according to the node distance and the user similarity, and the calculation formula of the walk probability is: ； wherein, denotes the walk probability, denotes the user similarity, denotes the node distance, , .

4. The item recommendation method according to claim 1, wherein Before inputting the user walk sequence into the pre-trained user representation model to obtain the user representation vector, the method further comprises: inputting a plurality of user walk sequences into a pre-constructed learning network to obtain a prediction vector and a prediction bias of each user walk sequence; calculating a co-occurrence loss value of any two user nodes according to the prediction vector, the prediction bias, and the co-occurrence frequency of the any two user nodes in a plurality of user walk sequences; calculating the sum of a plurality of co-occurrence loss values to obtain a total loss value; adjusting the learning network until the total loss value converges to obtain the user representation model.

5. The item recommendation method according to claim 1, wherein The generating of the interested item vector of the target user based on the item representation vector and the item context vector comprises: selecting, based on the interaction information, a first interaction item of the target user from the plurality of initial items; identifying a similar user of the target user according to the user representation vector; vector fusing a project representation vector of the first interaction item and a project context vector of the first interaction item to obtain a first fused feature; vector fusing a project representation vector of a second interaction item of the similar user and a project context vector of the second interaction item to obtain a second fused feature; interacting the first fused feature and the second fused feature to obtain the item of interest vector.

6. The item recommendation method according to claim 1, wherein The generating the target item of the target user based on the user context vector, the user representation vector and the item of interest vector includes: interacting the user context vector, the user representation vector and the item of interest vector to obtain an interaction feature; performing attention analysis on the interaction feature to obtain an attention feature; performing classification prediction processing on the attention feature to obtain the target item.

7. An item recommendation device characterized by comprising: The item recommendation apparatus includes: a generating unit configured to generate an interaction matrix based on interaction information of an initial user to a plurality of initial items, the initial user including a target user; the generating unit is further configured to generate a user walk sequence of each initial user based on the interaction matrix, and generate a project walk sequence of each initial item based on the interaction matrix; an input unit configured to input the user walk sequence to a pre-trained user representation model to obtain a user representation vector, and process the project walk sequence based on a pre-trained project representation model to obtain a project representation vector; The generation unit is further configured to generate a user context vector based on multi-dimensional information of the target user, and generate an item context vector based on multi-dimensional features of each initial item, the multi-dimensional information including discrete information, continuous information, text information and image information of the target user, the generation of the user context vector based on the multi-dimensional information of the target user including: vector encoding the discrete information to obtain discrete features; normalizing the continuous information to obtain continuous features; feature extraction on the text information to obtain text features; convolution processing on the image information to obtain image features; and feature interaction processing on the discrete features, the continuous features, the text features and the image features to obtain the user context vector, a calculation formula of the user context vector being: ; wherein, represents the user context vector, represents a total number of features in the discrete features, the continuous features, the text features and the image features, and respectively represent any feature in the discrete features, the continuous features, the text features and the image features, and respectively represent a transpose weight of the any feature, and respectively represent a bias of the any feature. the generating unit is further configured to generate an item of interest vector of the target user based on the project representation vector and a project context vector; the generating unit is further configured to generate a target item of the target user based on the user context vector, the user representation vector and the item of interest vector.

8. An electronic device, comprising: The electronic device includes: a memory storing computer readable instructions; and a processor executing the computer readable instructions stored in the memory to implement the item recommendation method of any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer readable storage medium stores computer readable instructions, and the computer readable instructions are executed by a processor in an electronic device to implement the item recommendation method of any one of claims 1 to 6. The computer readable storage medium stores computer readable instructions, and the computer readable instructions are executed by a processor in an electronic device to implement the item recommendation method of any one of claims 1 to 6.

Images