Artificial intelligence-based project recommendation method, apparatus, device, and medium

Abstract

The embodiment of the application relates to the financial field and the intelligent medical treatment technical field, and discloses a project recommendation method, device and equipment based on artificial intelligence and a medium, the method comprises the following steps: obtaining a historical interaction sequence of a user; determining an intention vector of the user through a double-intention network according to the historical interaction projects arranged in time sequence and the classification tree data of each historical interaction project; obtaining a target target recommendation project in a target recommendation project set through an inference network based on the historical interaction sequence and basic data of the target recommendation project set, and obtaining a target project vector of the target target recommendation project; and calculating a recommendation score of the target target recommendation project based on the intention vector and the target project vector. The intention vector of the user is determined through the double-intention network, the project vector of the target target recommendation project is selected, and the intention vector and the project vector are iterated, so that the target recommendation project can be recommended, the user can obtain the target recommendation project in time, and user experience is improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, finance, and smart healthcare, and particularly to a project recommendation method, apparatus, device, and medium based on artificial intelligence. Background Technology

[0002] In the fields of finance and smart healthcare, as technology advances and user needs constantly evolve, new projects are often launched to interact with users. For example, in the financial sector, banks launch various financial programs as the economy develops; similarly, in the field of smart healthcare, medical institutions launch various medical programs as user needs change. These newly launched projects often offer better results and provide users with a superior experience.

[0003] In recent years, many sequential recommendation methods have transformed the user's historical behavior sequence into a general vector representation to recommend the next item to the user. Currently, graph neural networks can only recommend items that already exist in the user's historical sequence in sequential recommendation. Therefore, it is very difficult for existing sequential recommendation systems to recommend new items, resulting in a poor user experience. Summary of the Invention

[0004] In view of this, the present invention provides an artificial intelligence-based project recommendation method, apparatus, device and medium to solve the problem that recommending new projects is difficult in the prior art, resulting in a poor user experience.

[0005] To achieve one or more of the above objectives or other objectives, the present invention proposes an artificial intelligence-based item recommendation method, comprising: obtaining a user's historical interaction sequence, wherein the historical interaction sequence includes historical interaction items arranged in chronological order, attribute set information of each historical interaction item, and classification tree data of each historical interaction item;

[0006] The user's intent vector is determined through a dual intent network based on historical interaction items arranged in chronological order and classification tree data of each historical interaction item.

[0007] Based on the historical interaction sequence and the basic data of the item set to be recommended, the target item to be recommended in the item set to be recommended is obtained through the inference network, and the target item vector of the target item to be recommended is obtained. The basic data includes the attribute set information of each item to be recommended and the classification tree data of each item to be recommended. The item set to be recommended includes items to be recommended, and the items to be recommended are items that have not interacted with the user.

[0008] The recommendation score for the target item to be recommended is calculated based on the intent vector and the target item vector, and the item recommendation is completed based on the recommendation score.

[0009] On the other hand, this application provides an artificial intelligence-based project recommendation device, the device comprising:

[0010] The data acquisition module is used to acquire the user's historical interaction sequence, which includes historical interaction items arranged in chronological order, attribute set information of each historical interaction item, and classification tree data of each historical interaction item.

[0011] The intent determination module is used to determine the user's intent vector through a dual intent network based on historical interaction items arranged in chronological order and classification tree data of each historical interaction item.

[0012] The selection module is used to obtain the target recommended items in the recommended item set through an inference network based on the historical interaction sequence and the basic data of the recommended item set, and to obtain the target item vector of the target recommended item. The basic data includes the attribute set information of each recommended item and the classification tree data of each recommended item. The recommended item set includes recommended items, and the recommended items are items that have not interacted with the user.

[0013] The recommendation module is used to calculate the recommendation score of the target item to be recommended based on the intent vector and the target item vector, and to complete the item recommendation based on the recommendation score.

[0014] On the other hand, this application provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory through the bus. When the machine-readable instructions are executed by the processor, they are used to obtain a user's historical interaction sequence. The historical interaction sequence includes historical interaction items arranged in chronological order, attribute set information of each historical interaction item, and classification tree data of each historical interaction item.

[0015] The user's intent vector is determined through a dual intent network based on historical interaction items arranged in chronological order and classification tree data of each historical interaction item.

[0016] Based on the historical interaction sequence and the basic data of the item set to be recommended, the target item to be recommended in the item set to be recommended is obtained through the inference network, and the target item vector of the target item to be recommended is obtained. The basic data includes the attribute set information of each item to be recommended and the classification tree data of each item to be recommended. The item set to be recommended includes items to be recommended, and the items to be recommended are items that have not interacted with the user.

[0017] The recommendation score for the target item to be recommended is calculated based on the intent vector and the target item vector, and the item recommendation is completed based on the recommendation score.

[0018] On the other hand, this application provides a computer-readable storage medium storing a computer program that is executed by a processor to obtain a user's historical interaction sequence. The historical interaction sequence includes historical interaction items arranged in chronological order, attribute set information of each historical interaction item, and classification tree data of each historical interaction item.

[0019] The user's intent vector is determined through a dual intent network based on historical interaction items arranged in chronological order and classification tree data of each historical interaction item.

[0020] Based on the historical interaction sequence and the basic data of the item set to be recommended, the target item to be recommended in the item set to be recommended is obtained through the inference network, and the target item vector of the target item to be recommended is obtained. The basic data includes the attribute set information of each item to be recommended and the classification tree data of each item to be recommended. The item set to be recommended includes items to be recommended, and the items to be recommended are items that have not interacted with the user.

[0021] The recommendation score for the target item to be recommended is calculated based on the intent vector and the target item vector, and the item recommendation is completed based on the recommendation score.

[0022] Implementing the embodiments of the present invention will have the following beneficial effects:

[0023] Based on the historical interaction items arranged in chronological order and the classification tree data of each historical interaction item, the user's intent vector is determined through a dual intent network. That is, the dual intent network uses a soft attention mechanism to enhance the user's current interest, making the recognition of the user's intent more accurate. Based on the historical interaction sequence and the basic data of the set of items to be recommended, the target items to be recommended in the set of items to be recommended are obtained through an inference network. That is, the item vector of the target item to be recommended is selected, breaking the closed loop of the historical interaction sequence. Based on the intent vector and the target item vector, the recommendation score of the target item to be recommended is calculated, and the item recommendation is completed according to the recommendation score. That is, the intent vector and the item vector are iterated to ensure that the items to be recommended can be recommended, and the user can obtain the recommended items in a timely manner, thereby improving the user experience. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] in:

[0026] Figure 1 This is an application scenario diagram of an artificial intelligence-based project recommendation method provided in an embodiment of this application;

[0027] Figure 2 This is a flowchart of an artificial intelligence-based project recommendation method provided in an embodiment of this application;

[0028] Figure 3 This is a schematic diagram of the structure of an artificial intelligence-based project recommendation device provided in an embodiment of this application;

[0029] Figure 4 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention;

[0030] Figure 5 This is another structural schematic diagram of a computer device according to one embodiment of the present invention.

[0031] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0032] Figure 7 This is a schematic diagram of the structure of a storage medium provided in an embodiment of this application. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] The AI-based project recommendation method provided in this invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network. The server can obtain the user's historical interaction sequence, which includes historical interaction items arranged in chronological order, attribute set information for each historical interaction item, and classification tree data for each historical interaction item. Based on the historical interaction items arranged in chronological order and the classification tree data for each historical interaction item, the server determines the user's intent vector through a dual intent network. Based on the historical interaction sequence and the basic data of the item set to be recommended, the server obtains the target items to be recommended in the item set to be recommended through an inference network, and obtains the target item vector of the target item to be recommended. The basic data includes attribute set information and classification tree data for each item to be recommended. The item set to be recommended includes items to be recommended, which are items that have not interacted with the user. Based on the intent vector and the target item vector, the server calculates the recommendation score for the target item to be recommended, and completes the item recommendation based on the recommendation score. Based on the historical interaction items arranged in chronological order and the classification tree data of each historical interaction item, a dual intent network is used to determine the user's intent vector. Specifically, the dual intent network employs a soft attention mechanism to enhance the user's current interest, making the recognition of the user's intent more accurate. Based on the historical interaction sequence and the basic data of the item set to be recommended, a reasoning network is used to obtain the target item to be recommended from the item set. This involves selecting the item vector of the target item to be recommended, breaking the closed loop of the historical interaction sequence. Based on the intent vector and the target item vector, a recommendation score is calculated for the target item to be recommended, and the item recommendation is completed based on the recommendation score. In other words, the intent vector and item vector are iterated to ensure that the item to be recommended is recommended, allowing the user to access the recommended item promptly and improving the user experience. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0035] For example, in the medical field, obtaining a user's historical interaction sequence includes inquiry items, consultation items, and medication purchase items. When making item recommendations, existing technologies only recommend items to users according to the historical interaction sequence, and items that are not in the historical interaction sequence will not be recommended. For example, the admission item for hospitalization does not appear in the historical interaction sequence, but in real-world scenarios, hospitalization is often something that users need to do after finishing a consultation. Therefore, the admission item needs to be recommended to better meet user needs.

[0036] To alleviate the computational burden on the server, the AI-based project recommendation method provided in this embodiment of the invention can also be applied to... Figure 1The client in the middle.

[0037] like Figure 2 As shown in the figure, this application provides an artificial intelligence-based project recommendation method, including:

[0038] S101. Obtain the user's historical interaction sequence, which includes historical interaction items arranged in chronological order, attribute set information of each historical interaction item, and classification tree data of each historical interaction item.

[0039] The project recommendation method provided in this application can be applied to project recommendation devices or engines in various scenarios. The project recommendation process is usually implemented through a server, which can transmit data with the user's client in real time. For example, when the server receives a project recommendation request (user interaction request) from the client, it recommends projects for the user, that is, it obtains the user's historical interaction sequence. The historical interaction sequence includes historical interaction items arranged in chronological order, attribute set information of each historical interaction item, and classification tree data of each historical interaction item.

[0040] For example, a user's historical sequence of interactions with an interactive item is S = {v1, v2, v3, ..., v...} n}, each historical interactive item v i Sort by time, each historical interaction item has a tree with three different category levels, denoted as t. i1 ,t i2 ,t i3 The categories are arranged from most granular to least granular. Additionally, the attribute set information A = {a} for each historical interaction item is also provided. i1 ,…,a ik}, each historical interactive item v i The global category vector representation (embedding) is t i =W(t) i1 ,t i2 ,t i3 ), where W is a multilayer perceptron, which is used to compress three levels of dimensional vector representations (embedding) into a single hidden dimension.

[0041] For example, the project is used to control the program through the interaction between the user and the program (operations such as inputting data), so that the program runs according to the user's needs and obtains the results required by the user. The attribute set information is such as the category attribute of the interactive project, and the classification tree data is the data of different categories of the interactive project.

[0042] For example, in the financial field, a user's historical sequence of interactions with interactive items, such as item v i For querying projects, item v j To calculate the items, sort the query by time and find items that appear before the calculated items.

[0043] S102. Determine the user's intent vector through a dual intent network based on the historical interaction items arranged in chronological order and the classification tree data of each historical interaction item.

[0044] For example, dual-intent networks determine the user's intent vector by applying self-attention and soft attention mechanisms, making the recognition of the user's intent more accurate.

[0045] For example, based on the user's historical interaction records obtained from historical interaction items arranged in chronological order, after interacting with a query item, the user interacts with a calculation item. The types of query items and calculation items are determined by the classification tree data of historical interaction items. Thus, after the user completes the interaction with the query item, the user's intention is predicted to be calculation.

[0046] S103. Based on the historical interaction sequence and the basic data of the item set to be recommended, the target item to be recommended in the item set to be recommended is obtained through the inference network, and the target item vector of the target item to be recommended is obtained. The basic data includes the attribute set information of each item to be recommended and the classification tree data of each item to be recommended. The item set to be recommended includes items to be recommended. The items to be recommended are items that have not interacted with the user.

[0047] For example, based on the historical interaction sequence and the basic data of the item set to be recommended, the target items in the item set to be recommended are obtained through an inference network. The inference network can be a dynamic Bayesian network, which selects the target item vector from the item set to be recommended based on the user's historical interaction sequence and the basic data of the item set to be recommended.

[0048] S104. Calculate the recommendation score of the target item to be recommended based on the intent vector and the target item vector, and complete the item recommendation based on the recommendation score;

[0049] Based on the historical interaction items arranged in chronological order and the classification tree data of each historical interaction item, the user's intent vector is determined through a dual intent network. That is, the dual intent network uses a soft attention mechanism to enhance the user's current interest, making the recognition of the user's intent more accurate. Based on the historical interaction sequence and the basic data of the set of items to be recommended, the target items to be recommended in the set of items to be recommended are obtained through an inference network. That is, the item vector of the target item to be recommended is selected, breaking the closed loop of the historical interaction sequence. Based on the intent vector and the target item vector, the recommendation score of the target item to be recommended is calculated, and the item recommendation is completed according to the recommendation score. That is, the intent vector and the item vector are iterated to ensure that the items to be recommended can be recommended, and the user can obtain the recommended items in a timely manner, thereby improving the user experience.

[0050] In one possible implementation, prior to the step of determining the user's intent vector through a dual intent network based on chronologically ordered historical interaction items and classification tree data of each historical interaction item, the method further includes:

[0051] Construct a historical sequence diagram based on historical interaction items arranged in chronological order;

[0052] The user's intent vector is determined through a dual intent network based on the historical sequence diagram and the classification tree data of each historical interaction item.

[0053] For example, a historical sequence is transformed into a historical sequence graph, which represents that after a user interacts with one item, they immediately interact with another item. Each edge of the historical sequence graph consists of two parts: in-degree and out-degree. Then, an adjacency matrix is ​​obtained. The adjacency matrix is ​​used to represent the relationship between each vertex (historical interaction item) in the graph. A relationship is 1, and no relationship is 0.

[0054] For example, a user's intent vector can be determined using a dual intent network based on the adjacency matrix and classification tree data of each historical interaction item.

[0055] By representing the process of user interaction items through historical sequence diagrams, the relationships between different items can be revealed, thereby more accurately identifying user intent, such as learning the patterns of user interaction items.

[0056] In one possible implementation, the step of determining the user's intent vector through a dual intent network based on the historical sequence graph and the classification tree data of each historical interaction item includes:

[0057] The historical sequence diagram and the classification tree data of each historical interaction item are imported into a graph neural network to obtain the historical item vector and the category vector of each historical interaction item.

[0058] The user's intent vector is determined through a dual intent network based on the historical item vector and the category vector.

[0059] For example, the graph domain information of the historical sequence graph is processed by a graph neural network, that is, the relationship between each historical interaction item and each vertex (historical interaction item). The user's intent vector is determined by using the relationship between each historical interaction item and each vertex (historical interaction item) and the category vector of each historical interaction item as input to the dual intent network.

[0060] For example, graph neural networks (GNNs) are a type of deep learning-based method for processing graph information. They offer good performance and interpretability.

[0061] In one possible implementation, the step of determining the user's intent vector through a dual intent network based on the historical item vector and the category vector includes:

[0062] The first intent vector is obtained by using the logistic function in the dual intent network based on the historical item vector and the category vector.

[0063] The second intent vector is obtained by using the distribution function in the dual intent network based on the historical item vector and the category vector.

[0064] The user's intent vector is obtained by combining the first intent vector and the second intent vector using preset weight parameters.

[0065] For example, in a dual-intent network, the first is the user's α-intent network. Items the user has most recently interacted with reflect their current intent better than earlier interactions. For a user's sequence graph, the user's intent I... α It can be obtained from the following formula:

[0066] Embedding, W1 and W2 are two intent filtering matrices, v n and v i Given any two items in the historical interaction sequence, we can obtain the user's first intent vector I by applying a self-attention mechanism. α .

[0067] In the user beta intent network, the user's most recent interaction with the item reflects their current intent better than earlier interactions. Therefore, a soft attention mechanism is used in the second intent vector to enhance the user's current intent.

[0068]

[0069] In the formula v′ i It is vi The sequence arranged by importance, i.e., user interest, b i To make v i Substituting the distribution value into the β distribution function, b n For v n Substituting the distribution value into the β distribution function, due to the reduction of v i The probability distribution needs to be corrected using the β distribution function, as shown below:

[0070]

[0071] In the formula, avg(b) is the averaging function, and W3 is a multilayer perceptron with reduced dimensionality, yielding the result of the user's β intent (second intent vector):

[0072]

[0073] The first intent vector and the second intent vector are then combined to obtain the user's total intent vector:

[0074] I=λI α +(1-λ)I β

[0075] In the formula, λ∈[0,1] is the weight parameter of the intention.

[0076] By employing a soft attention mechanism to enhance the user's current intent, the user's intent is amplified, and the first intent vector and the second intent vector are combined with weights to make the obtained user intent more accurate.

[0077] In one possible implementation, the step of obtaining target items to be recommended from the target item set through an inference network based on the historical interaction sequence and the basic data of the item set to be recommended, and obtaining the target item vector of the target item to be recommended, wherein the basic data includes attribute set information of each item to be recommended and classification tree data of each item to be recommended, and the item set to be recommended includes items to be recommended that have not interacted with the user, includes:

[0078] The classification tree data of each item to be recommended and the items to be recommended are imported into a graph neural network to obtain the item vector of each item to be recommended and the category vector of each item to be recommended.

[0079] Based on the proposed item vector, the attribute set information of each proposed item, the category vector of each proposed item, the historical item vector of each historical interaction item, the attribute set information of each historical interaction item, and the category vector of each historical interaction item, the target proposed item in the proposed item set is obtained through the inference network, and the target item vector of the target proposed item is obtained.

[0080] For example, the vector of the item to be recommended, the attribute set information of each item to be recommended, the category vector of each item to be recommended, the historical item vector of each historical interaction item, the attribute set information of each historical interaction item, and the category vector of each historical interaction item are used as input data and input into the inference network. The similarity and correlation between each historical interaction item and each item to be recommended can be obtained. Then, the target item to be recommended in the set of items to be recommended can be obtained based on the similarity or correlation, and the target item vector of the target item to be recommended can be obtained.

[0081] For example, historical interaction items are query items, and the set of items to be recommended includes retrieval items, query items of the new algorithm, and maintenance items. The similarity (relevance) between query items and retrieval items, query items of the new algorithm, and maintenance items is calculated through the inference network. When the similarity (relevance) is higher than the threshold, the retrieval items and query items of the new algorithm are selected as target items to be recommended.

[0082] For example, the similarity (relevance) can be calculated by calculating the distance between the features of historical interaction items and the item to be recommended. If the distance is small, the similarity is large; if the distance is large, the similarity is small.

[0083] In one possible implementation, the step of obtaining the target item to be recommended from the set of items to be recommended through an inference network based on the item vector to be recommended, the attribute set information of each item to be recommended, the category vector of each item to be recommended, the historical item vector of each historical interaction item, the attribute set information of each historical interaction item, and the category vector of each historical interaction item, and obtaining the target item vector of the target item to be recommended, includes:

[0084] The attribute set information of each item to be recommended and the attribute set information of each historical interaction item are encoded separately to obtain the first encoded data corresponding to the attribute set information of each item to be recommended and the second encoded data corresponding to the attribute set information of each historical interaction item.

[0085] Based on the proposed item vector, the first encoded data, the category vectors of each proposed item, the historical item vectors of each historical interaction item, the second encoded data, and the category vectors of each historical interaction item, the target proposed item in the proposed item set is obtained through an inference network, and the target item vector of the target proposed item is obtained.

[0086] For example, the process of encoding the attribute set information of each item to be recommended and the attribute set information of each historical interaction item to obtain the first encoded data corresponding to the attribute set information of each item to be recommended and the second encoded data corresponding to the attribute set information of each historical interaction item is as follows:

[0087] The item and the category tree are encoded, and the encoding process is as follows:

[0088]

[0089]

[0090]

[0091]

[0092]

[0093] In the formula It is a list of item vectors, and t is the training step size. It is the i-th row of the matrix. and These are the weights and bias parameters, and These are the reset and update gates, σ(.) is the sigmoid function, ⊙ is the element-wise multiplication, tanh() is the hyperbolic tangent function, U is the user's vector set, and a i This represents the encoded item vector, where z is the z-th user in the vector set, r is the r-th user in the vector set, o is the o-th user in the vector set, and W is the user weight. and The encoding results for classification trees at different classification levels, for Intermediate parameters.

[0094] In one possible implementation, the step of calculating a recommendation score for the target item to be recommended based on the intent vector and the target item vector, and completing the item recommendation based on the recommendation score, includes:

[0095] Based on the intent vector and the target item vector, the recommendation score of the target item to be recommended is calculated to obtain the recommendation score of each target item to be recommended.

[0096] The recommended items are arranged in descending order of their scores to form a recommendation queue.

[0097] Complete the project recommendation according to the recommendation queue.

[0098] For example, through softmax(I T *c i The score of each target item to be recommended is obtained, and then sorted from highest to lowest score to obtain a recommendation queue. A predetermined number of top-ranked target items are then selected for recommendation. Where I T Let c be the intent vector. i Let be the target project vector.

[0099] In one possible implementation, such as Figure 3 As shown, this application provides an artificial intelligence-based project recommendation device, the device comprising:

[0100] Data acquisition module 201 is used to acquire the user's historical interaction sequence, the historical interaction sequence including historical interaction items arranged in chronological order, attribute set information of each historical interaction item, and classification tree data of each historical interaction item;

[0101] The intent determination module 202 is used to determine the user's intent vector through a dual intent network based on historical interaction items arranged in chronological order and classification tree data of each historical interaction item.

[0102] The selection module 203 is used to obtain the target items to be recommended in the set of items to be recommended through an inference network based on the historical interaction sequence and the basic data of the set of items to be recommended, and to obtain the target item vector of the target items to be recommended. The basic data includes the attribute set information of each item to be recommended and the classification tree data of each item to be recommended. The set of items to be recommended includes items to be recommended, and the items to be recommended are items that have not interacted with the user.

[0103] The recommendation module 204 is used to calculate the recommendation score of the target item to be recommended based on the intent vector and the target item vector, and to complete the item recommendation based on the recommendation score.

[0104] In one possible implementation, the data acquisition module 201 is used for:

[0105] Construct a historical sequence diagram based on historical interaction items arranged in chronological order;

[0106] The user's intent vector is determined through a dual intent network based on the historical sequence diagram and the classification tree data of each historical interaction item.

[0107] In one possible implementation, the determining module 202 is intended to:

[0108] The historical sequence diagram and the classification tree data of each historical interaction item are imported into a graph neural network to obtain the historical item vector and the category vector of each historical interaction item.

[0109] The user's intent vector is determined through a dual intent network based on the historical item vector and the category vector.

[0110] In one possible implementation, the determining module 202 is intended to:

[0111] The first intent vector is obtained by using the logistic function in the dual intent network based on the historical item vector and the category vector.

[0112] The second intent vector is obtained by using the distribution function in the dual intent network based on the historical item vector and the category vector.

[0113] The user's intent vector is obtained by combining the first intent vector and the second intent vector using preset weight parameters.

[0114] In one possible implementation, the selection module 203 is used for:

[0115] The classification tree data of each item to be recommended and the items to be recommended are imported into a graph neural network to obtain the item vector of each item to be recommended and the category vector of each item to be recommended.

[0116] Based on the proposed item vector, the attribute set information of each proposed item, the category vector of each proposed item, the historical item vector of each historical interaction item, the attribute set information of each historical interaction item, and the category vector of each historical interaction item, the target proposed item in the proposed item set is obtained through the inference network, and the target item vector of the target proposed item is obtained.

[0117] In one possible implementation, the selection module 203 is used for:

[0118] The attribute set information of each item to be recommended and the attribute set information of each historical interaction item are encoded separately to obtain the first encoded data corresponding to the attribute set information of each item to be recommended and the second encoded data corresponding to the attribute set information of each historical interaction item.

[0119] Based on the proposed item vector, the first encoded data, the category vectors of each proposed item, the historical item vectors of each historical interaction item, the second encoded data, and the category vectors of each historical interaction item, the target proposed item in the proposed item set is obtained through an inference network, and the target item vector of the target proposed item is obtained.

[0120] In one possible implementation, the recommendation module 204 is used for:

[0121] Based on the intent vector and the target item vector, the recommendation score of the target item to be recommended is calculated to obtain the recommendation score of each target item to be recommended.

[0122] The recommended items are arranged in descending order of their scores to form a recommendation queue.

[0123] Complete the project recommendation according to the recommendation queue.

[0124] This invention provides a project recommendation device. It determines the user's intent vector through a dual-intent network based on historical interaction items arranged chronologically and classification tree data of each historical interaction item. Specifically, the dual-intent network employs a soft attention mechanism to enhance the user's current interest, making the recognition of the user's intent more accurate. Based on the historical interaction sequence and the basic data of the project set to be recommended, a reasoning network is used to obtain the target project to be recommended from the project set. This involves selecting the project vector of the target project to be recommended, breaking the closed loop of the historical interaction sequence. A recommendation score for the target project to be recommended is calculated based on the intent vector and the target project vector, and the project recommendation is completed based on the recommendation score. In other words, the intent vector and project vector are iterated to ensure that projects to be recommended are recommended, allowing users to access recommended projects promptly and improving the user experience.

[0125] Specific limitations regarding the project recommendation device can be found in the limitations of the project recommendation method described above, and will not be repeated here. Each module in the aforementioned project recommendation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0126] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a server-side project recommendation method based on artificial intelligence.

[0127] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements client-side functions or steps of an artificial intelligence-based project recommendation method.

[0128] In one possible implementation, such as Figure 6 As shown, this application embodiment provides an electronic device 300, including: a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it implements: acquiring the user's historical interaction sequence, the historical interaction sequence including historical interaction items arranged in chronological order, attribute set information of each historical interaction item, and classification tree data of each historical interaction item;

[0129] The user's intent vector is determined through a dual intent network based on historical interaction items arranged in chronological order and classification tree data of each historical interaction item.

[0130] Based on the historical interaction sequence and the basic data of the item set to be recommended, the target item to be recommended in the item set to be recommended is obtained through the inference network, and the target item vector of the target item to be recommended is obtained. The basic data includes the attribute set information of each item to be recommended and the classification tree data of each item to be recommended. The item set to be recommended includes items to be recommended, and the items to be recommended are items that have not interacted with the user.

[0131] The recommendation score for the target item to be recommended is calculated based on the intent vector and the target item vector, and the item recommendation is completed based on the recommendation score.

[0132] Based on the historical interaction items arranged in chronological order and the classification tree data of each historical interaction item, the user's intent vector is determined through a dual intent network. That is, the dual intent network uses a soft attention mechanism to enhance the user's current interest, making the recognition of the user's intent more accurate. Based on the historical interaction sequence and the basic data of the set of items to be recommended, the target items to be recommended in the set of items to be recommended are obtained through an inference network. That is, the item vector of the target item to be recommended is selected, breaking the closed loop of the historical interaction sequence. Based on the intent vector and the target item vector, the recommendation score of the target item to be recommended is calculated, and the item recommendation is completed according to the recommendation score. That is, the intent vector and the item vector are iterated to ensure that the items to be recommended can be recommended, and the user can obtain the recommended items in a timely manner, thereby improving the user experience.

[0133] In one possible implementation, such as Figure 7 As shown, this application embodiment provides a computer-readable storage medium 400, on which a computer program 411 is stored. When the computer program 411 is executed by a processor, it implements: acquiring a user's historical interaction sequence, the historical interaction sequence including historical interaction items arranged in chronological order, attribute set information of each historical interaction item, and classification tree data of each historical interaction item;

[0134] The user's intent vector is determined through a dual intent network based on historical interaction items arranged in chronological order and classification tree data of each historical interaction item.

[0135] Based on the historical interaction sequence and the basic data of the item set to be recommended, the target item to be recommended in the item set to be recommended is obtained through the inference network, and the target item vector of the target item to be recommended is obtained. The basic data includes the attribute set information of each item to be recommended and the classification tree data of each item to be recommended. The item set to be recommended includes items to be recommended, and the items to be recommended are items that have not interacted with the user.

[0136] The recommendation score for the target item to be recommended is calculated based on the intent vector and the target item vector, and the item recommendation is completed based on the recommendation score.

[0137] Based on the historical interaction items arranged in chronological order and the classification tree data of each historical interaction item, the user's intent vector is determined through a dual intent network. That is, the dual intent network uses a soft attention mechanism to enhance the user's current interest, making the recognition of the user's intent more accurate. Based on the historical interaction sequence and the basic data of the set of items to be recommended, the target items to be recommended in the set of items to be recommended are obtained through an inference network. That is, the item vector of the target item to be recommended is selected, breaking the closed loop of the historical interaction sequence. Based on the intent vector and the target item vector, the recommendation score of the target item to be recommended is calculated, and the item recommendation is completed according to the recommendation score. That is, the intent vector and the item vector are iterated to ensure that the items to be recommended can be recommended, and the user can obtain the recommended items in a timely manner, thereby improving the user experience.

[0138] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0139] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0140] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0141] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0142] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0143] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

[0144] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A project recommendation method based on artificial intelligence, characterized in that, include: Obtain the user's historical interaction sequence, which includes historical interaction items arranged in chronological order, attribute set information of each historical interaction item, and classification tree data of each historical interaction item; The user's intent vector is determined through a dual intent network based on historical interaction items arranged in chronological order and classification tree data of each historical interaction item. Based on the historical interaction sequence and the basic data of the item set to be recommended, the target item to be recommended in the item set to be recommended is obtained through the inference network, and the target item vector of the target item to be recommended is obtained. The basic data includes the attribute set information of each item to be recommended and the classification tree data of each item to be recommended. The item set to be recommended includes items to be recommended, and the items to be recommended are items that have not interacted with the user. Calculate the recommendation score of the target item to be recommended based on the intent vector and the target item vector, and complete the item recommendation based on the recommendation score; Prior to the step of determining the user's intent vector through a dual intent network based on chronologically ordered historical interaction items and classification tree data of each historical interaction item, the method further includes: Construct a historical sequence diagram based on historical interaction items arranged in chronological order; Based on the historical sequence diagram and the classification tree data of each historical interaction item, the user's intent vector is determined through a dual intent network. The step of determining the user's intent vector through a dual intent network based on the historical sequence diagram and the classification tree data of each historical interaction item includes: The historical sequence diagram and the classification tree data of each historical interaction item are imported into a graph neural network to obtain the historical item vector and the category vector of each historical interaction item. The user's intent vector is determined using a dual intent network based on the historical item vector and the category vector. The step of determining the user's intent vector through a dual intent network based on the historical item vector and the category vector includes: The first intent vector is obtained by using the logistic function in the dual intent network based on the historical item vector and the category vector. The second intent vector is obtained by using the distribution function in the dual intent network based on the historical item vector and the category vector. The user's intent vector is obtained by combining the first intent vector and the second intent vector using preset weight parameters. The first intention vector is represented by the following formula: In the formula It is the sigmoid function. It's the concat function for concatenating characters. It is an item Global category embedding and These are two intent filtering matrices. and For any two items in the historical interaction sequence, the first intent vector .

2. The project recommendation method based on artificial intelligence as described in claim 1, characterized in that, The step of obtaining target items to be recommended from the target item set through an inference network based on the historical interaction sequence and the basic data of the item set to be recommended, and obtaining the target item vector of the target item to be recommended, wherein the basic data includes attribute set information of each item to be recommended and classification tree data of each item to be recommended, and the item set to be recommended includes items to be recommended, wherein the items to be recommended are items that have not interacted with the user, includes: The classification tree data of each item to be recommended and the items to be recommended are imported into a graph neural network to obtain the item vector of each item to be recommended and the category vector of each item to be recommended. Based on the proposed item vector, the attribute set information of each proposed item, the category vector of each proposed item, the historical item vector of each historical interaction item, the attribute set information of each historical interaction item, and the category vector of each historical interaction item, the target proposed item in the proposed item set is obtained through the inference network, and the target item vector of the target proposed item is obtained.

3. The project recommendation method based on artificial intelligence as described in claim 2, characterized in that, The step of obtaining the target project in the set of projects to be recommended through an inference network based on the vector of projects to be recommended, the attribute set information of each project to be recommended, the category vector of each project to be recommended, the historical project vector of each historical interaction project, the attribute set information of each historical interaction project, and the category vector of each historical interaction project, and obtaining the target project vector of the target project to be recommended, includes: The attribute set information of each item to be recommended and the attribute set information of each historical interaction item are encoded separately to obtain the first encoded data corresponding to the attribute set information of each item to be recommended and the second encoded data corresponding to the attribute set information of each historical interaction item. Based on the proposed item vector, the first encoded data, the category vectors of each proposed item, the historical item vectors of each historical interaction item, the second encoded data, and the category vectors of each historical interaction item, the target proposed item in the proposed item set is obtained through an inference network, and the target item vector of the target proposed item is obtained.

4. The project recommendation method based on artificial intelligence as described in claim 1, characterized in that, The step of calculating the recommendation score of the target item to be recommended based on the intent vector and the target item vector, and completing the item recommendation based on the recommendation score, includes: Based on the intent vector and the target item vector, the recommendation score of the target item to be recommended is calculated to obtain the recommendation score of each target item to be recommended. The recommended items are arranged in descending order of their scores to form a recommendation queue. Complete the project recommendation according to the recommendation queue.

5. An artificial intelligence-based project recommendation device, characterized in that, The device includes: The data acquisition module is used to acquire the user's historical interaction sequence, which includes historical interaction items arranged in chronological order, attribute set information of each historical interaction item, and classification tree data of each historical interaction item. The intent determination module is used to determine the user's intent vector through a dual intent network based on historical interaction items arranged in chronological order and classification tree data of each historical interaction item. The selection module is used to obtain the target recommended items in the recommended item set through an inference network based on the historical interaction sequence and the basic data of the recommended item set, and to obtain the target item vector of the target recommended item. The basic data includes the attribute set information of each recommended item and the classification tree data of each recommended item. The recommended item set includes recommended items, and the recommended items are items that have not interacted with the user. The recommendation module is used to calculate the recommendation score of the target item to be recommended based on the intent vector and the target item vector, and to complete the item recommendation based on the recommendation score; Prior to the step of determining the user's intent vector through a dual intent network based on chronologically ordered historical interaction items and classification tree data of each historical interaction item, the apparatus is further configured to: Construct a historical sequence diagram based on historical interaction items arranged in chronological order; Based on the historical sequence diagram and the classification tree data of each historical interaction item, the user's intent vector is determined through a dual intent network. The step of determining the user's intent vector through a dual intent network based on the historical sequence diagram and the classification tree data of each historical interaction item includes: The historical sequence diagram and the classification tree data of each historical interaction item are imported into a graph neural network to obtain the historical item vector and the category vector of each historical interaction item. The user's intent vector is determined using a dual intent network based on the historical item vector and the category vector. The step of determining the user's intent vector through a dual intent network based on the historical item vector and the category vector includes: The first intent vector is obtained by using the logistic function in the dual intent network based on the historical item vector and the category vector. The second intent vector is obtained by using the distribution function in the dual intent network based on the historical item vector and the category vector. The user's intent vector is obtained by combining the first intent vector and the second intent vector using preset weight parameters. The first intention vector is represented by the following formula: In the formula It is the sigmoid function. It's the concat function for concatenating characters. It is an item Global category embedding and These are two intent filtering matrices. and For any two items in the historical interaction sequence, the first intent vector .

6. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the AI-based project recommendation method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the artificial intelligence-based project recommendation method as described in any one of claims 1 to 4.

Images