Information recommendation method for government affair service platform

Abstract

The invention relates to an information recommendation method for a government affair service platform, which comprises the following steps of: according to a historical record of accessing a page of the government affair service platform by a user, constructing a relationship representation vector of linking the user to the page; mapping the relation representation vector to a tangent plane of a hyperbolic space to obtain hyperbolic features; the hyperbolic features serve as neurons of a BP neural network, the BP neural network is trained, and an information recommendation model is obtained and used for recommending needed pages for the user based on historical records of the user. The objective of the invention is to learn the historical behavior characteristics of the user more comprehensively, so that the subsequent prediction recommendation is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing, and in particular to an information recommendation method for a government service platform. Background Art

[0002] With the development of information technology, government service work is being realized in electronic form as much as possible, which improves the processing efficiency of government service work for users and government staff, and also has the traceability of processing, which greatly improves the efficiency of users' use of government services.

[0003] However, existing government service pages often focus on providing standardized service options, ignoring users' personalized needs and behavioral habits, resulting in a lack of pertinence and accuracy in service push. Nowadays, machine learning is an important means of big data processing. It accurately predicts user preferences by analyzing user historical behaviors (clicks, browsing, etc.), provides customized content, and significantly improves user satisfaction. However, machine learning is not accurate enough when extracting user behavior features, resulting in a reduction in the accuracy of subsequent predictions and recommendations. Summary of the invention

[0004] The purpose of the present invention is to be more comprehensive when learning the historical behavioral characteristics of users so that subsequent prediction and recommendation can be more accurate, and to provide an information recommendation method for a government service platform.

[0005] In order to achieve the above-mentioned object of the invention, the embodiment of the present invention provides the following technical solutions:

[0006] An information recommendation method for a government service platform comprises the following steps:

[0007] Step 1: Based on the user's history of visiting the government service platform page, a relationship representation vector linking the user to the page is constructed;

[0008] Step 2, mapping the relationship representation vector to the tangent plane of the hyperbolic space to obtain the hyperbolic feature;

[0009] Step 3: Use the hyperbolic features as neurons of the BP neural network, train the BP neural network, and obtain an information recommendation model for recommending required pages to the user based on the user's historical records.

[0010] The step 1 specifically comprises the following steps:

[0011] Step 1-1, let U represent the user set U={u 1 ,...u m ,u M}, where u m represents the mth user, M represents the number of users; user Z represents the page set Z={z 1,...z n ,...,z N}, where z n represents the nth page, N represents the number of pages; if user u m Visited pages n , then construct user u m Visit page n The link matrix ;

[0012] Step 1-2, based on the link matrix , calculate user u m For page z n The relationship representation index S mn ;

[0013] Step 1-3: Relationship representation index S mn After being processed by the Embedding layer, it is converted into a relation representation vector .

[0014] The step 1-1 specifically includes the following steps:

[0015] If user u m Visited pages n , then the link matrix , construct user u m Visit page n The link matrix :

[0016] in, is the number of interactions; is the number of sessions; is the number of entries; The duration of the most recent stay; is the maximum length of stay; For reference duration; is the page type feature vector; Rate the page; The frequency of entry.

[0017] The steps 1-2 specifically include the following steps:

[0018] The first relational indicator :

[0019] Among them, v=1,2,3; Indicates browsing time coefficient, used for balancing The size of ;

[0020] Second relationship characterization index :

[0021] in, Indicates page z n The type feature vector, r represents page z n Type;

[0022] The third relationship representation index :

[0023] User m For page z n The relationship representation index S mn for:

[0024] in, They are The weight coefficient is used to balance the importance of the first relationship representation index, the second relationship representation index and the third relationship representation index.

[0025] The step 2 specifically includes the following steps:

[0026] Let H d Represents a d-dimensional hyperbolic space with curvature c=-1 and constant k=-1 / c. Set the reference point to ; Represent the relationship vector Mapping to the tangent plane of hyperbolic space:

[0027] in, Representing the relationship vector Hyperbolic characteristics of ;||.|| L represents the L-order norm; [u m ,z n ] indicates user u m Link to page z n access operation; cosh represents the hyperbolic cosine function; sinh represents the hyperbolic sine function.

[0028] The step 2 further comprises the following steps:

[0029] The hyperbolic features are calculated for the pages visited by M users, thus obtaining the hyperbolic feature matrix :

[0030] Hyperbolic characteristic matrix The mth row in represents user um For the hyperbolic features of all pages, m=1,2,...,M; if user u m If some pages have not been visited, the corresponding hyperbolic feature is 0.

[0031] The step 3 specifically includes the following steps:

[0032] The non-zero hyperbolic features are used as neurons in the input layer of the BP neural network. If there are L non-zero hyperbolic features, the number of neurons in the input layer is L. The input data {x 1 ,x 2 ,...,x L} is mapped to a higher dimensional feature space, where x l represents the lth input data, l=1,2,...,L; through back propagation, the weights of the hidden layer neurons can be gradually adjusted to minimize the overall loss of the network and finally complete the learning task;

[0033] Construct the loss function Loss of the BP neural network:

[0034] Among them, x l represents the lth input data of the input layer, l=1,2,...,L; y n Indicates the page recommended for the user z n The probability that x l represents the hyperbolic feature of the mth user for the nth page, then y n represents the probability of recommending the nth page to the mth user; represents the weight coefficient, ;||.|| 2 represents the two-norm; represents the Hadamard product.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The present invention obtains the behavioral data of users accessing pages through the historical access records of users to the government service platform, constructs a link matrix, and calculates the relationship representation index of users to pages based on the elements in the link matrix. The index is calculated through three indicators and can fully represent the behavioral characteristics of users. After the relationship representation index is converted into a vector form, it is mapped to the tangent plane of the hyperbolic space. By utilizing the geometric characteristics of the hyperbolic space, the data of the hierarchical structure or tree-like relationship between the user's pages is more efficiently represented. Finally, the hyperbolic features output by the hyperbolic space are used as neurons of the BP neural network to train the BP neural network. During the training process, the loss function constructed enables the network to fit the complex data distribution of the page. The present invention combines the hyperbolic features mapped in the hyperbolic space with the neurons of the BP neural network, so that machine learning can fully learn the behavioral characteristics of users to web pages, as well as the hierarchical structure between these web pages. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0040] It should be noted that similar numbers and letters represent similar items in the following figures, so once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance, or implying any such actual relationship or order between these entities or operations. In addition, the terms "connected", "connected", etc. can be directly connected between elements, or indirectly connected via other elements.

[0041] Embodiment 1:

[0042] The present invention is achieved through the following technical solutions: Figure 1 As shown, a method for recommending information on a government service platform includes the following steps:

[0043] Step 1: Based on the user’s history of visiting government service platform pages, a relationship representation vector linking the user to the page is constructed.

[0044] Let U represent the user set U={u 1 ,...u m ,u M}, where u m represents the mth user, M represents the number of users. User Z represents the page set Z={z 1 ,...z n ,...,z N}, where z n Indicates the nth page, and N indicates the number of pages. The government service platform includes various types of pages, such as advertisements, news, notifications, and reporting. Reporting refers to filling in the user's personal information through the government service platform to achieve government service processing. Users can query some real-time events through the local government service platform, such as traffic operation conditions, weather conditions, housing sales conditions, etc.

[0045] If user u m Visited pages n , then the link matrix , construct user u m Visit page n The link matrix :

[0046] in, is the number of interactions, indicating that user u m Click on page z n The number of any links in ;

[0047] is the number of sessions, indicating that user u m On page z n Number of comments posted in

[0048] is the number of entries, indicating that user u m Enter page z n The number of times;

[0049] is the length of stay, indicating that user u m Last seen on page z n Duration of stay (seconds);

[0050] is the maximum stay time, indicating that user u m On page z n The longest duration of a stay (seconds);

[0051] is the reference time, indicating that the page has been browsed n Reference duration (seconds);

[0052] is the feature vector of page type, and page types include advertisement, news, notice, and report (if the information to be reported needs to go through multiple pages, then multiple pages of reporting one item are regarded as one page), etc.

[0053] Score the page, indicating that user u m For page z n Rating (with four rating levels: excellent 0.8, good 0.6, fair 0.4, poor 0.2);

[0054] is the entry frequency, indicating that user u m Enter page z n frequency.

[0055] Based on the link matrix , calculate user u m For page z n Relationship representation indicators:

[0056] The first relational indicator :

[0057] Among them, v=1,2,3; Indicates browsing time coefficient, used for balancing The size of .

[0058] Second relationship characterization index :

[0059] in, Indicates page z n The type feature vector, r represents page z n Type.

[0060] The third relationship representation index :

[0061] User m For page z n The relationship representation index S mn for:

[0062] in, They are The weight coefficient is used to balance the importance of the first relationship representation index, the second relationship representation index and the third relationship representation index.

[0063] The relationship representation index S mn After being processed by the Embedding layer, it is converted into a relation representation vector By calculating the above indicators, the relationship representation vector can fully represent the user u m For page z n Various behavioral characteristics of user u, while this scheme only calculates m Pages visited, for example, among N pages, only a portion of them are visited by user u m If the pages have been visited, only the relationship representation index corresponding to these pages is calculated, and for the pages that have not been visited, the link matrix is ​​expressed as .

[0064] Step 2: Map the relationship representation vector to the tangent plane of the hyperbolic space to obtain the hyperbolic feature.

[0065] Let H d Represents a d-dimensional hyperbolic space with curvature c=-1 and constant k=-1 / c. Set the reference point to . Represent the relationship vector Mapping to the tangent plane of hyperbolic space:

[0066] in, Representing the relationship vector Hyperbolic characteristics of ;||.|| L represents the L-order norm; [u m ,z n] indicates user u m Link to page z n access operation; cosh represents the hyperbolic cosine function; sinh represents the hyperbolic sine function.

[0067] Since there are M users and N pages, the hyperbolic features are calculated for each page visited by each user, thus obtaining the hyperbolic feature matrix :

[0068] Hyperbolic characteristic matrix The mth row in represents user u m For the hyperbolic features of all pages, m=1,2,...,M. If user u m If certain pages have not been visited, the corresponding hyperbolic feature is 0. For example, user u m Page not visited n-1 ,but .

[0069] This solution uses hyperbolic space mapping to embed the user's behavioral feature data on pages into a hyperbolic geometric space with negative curvature. By utilizing the geometric properties of the hyperbolic space, the user's data on the hierarchical structure or tree-like relationship between pages can be more efficiently represented.

[0070] Step 3: Use the hyperbolic features as neurons of the BP neural network, train the BP neural network, and obtain an information recommendation model for recommending required pages to the user based on the user's historical records.

[0071] Hyperbolic characteristic matrix The hyperbolic features of each user's page are included. The non-zero hyperbolic features are used as neurons in the input layer of the BP neural network. If there are L non-zero hyperbolic features, the number of neurons in the input layer is L. Then, the input data {x 1 ,x 2 ,...,x L} is mapped to a higher dimensional feature space (where x l Represents the lth input data, l=1,2,...,L), gradually extracts deep features, adjusts weights and biases, and enables the network to fit complex data distributions. Through back propagation, the weights of hidden layer neurons can be gradually adjusted to minimize the overall loss of the network and ultimately complete the learning task.

[0072] Construct the loss function Loss of the BP neural network:

[0073] Among them, x lrepresents the lth input data of the input layer, l=1,2,...,L; y n Indicates the page recommended for the user z n The probability that x l represents the hyperbolic feature of the mth user for the nth page, then y n represents the probability of recommending the nth page to the mth user; represents the weight coefficient, ;||.|| 2 represents the two-norm; represents the Hadamard product.

[0074] This scheme maps the records of users' historical visits to the government service platform pages through hyperbolic space, learns the users' behavioral characteristics of the pages, including browsing time, number of views, and type of the page, maps the users' behavioral characteristics of the pages into high-dimensional features that can be learned by machines, and trains the BP neural network through the loss function to obtain an information recommendation model, so that when users log in to the government service platform in the future, they can recommend the required pages to users.

[0075] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An information recommendation method for a government service platform, characterized in that: The following steps are involved: Step 1: Based on the user's history of visiting the government service platform page, a relationship representation vector linking the user to the page is constructed; Step 2, mapping the relationship representation vector to the tangent plane of the hyperbolic space to obtain the hyperbolic feature; Step 3: Use the hyperbolic features as neurons of the BP neural network, train the BP neural network, and obtain an information recommendation model for recommending required pages to the user based on the user's historical records.

2. The information recommendation method of the government service platform according to claim 1, characterized in that: The step 1 specifically comprises the following steps: Step 1-1, let U represent the user set U={u1,...u m ,u M }, where u m represents the mth user, M represents the number of users; user Z represents the page set Z={z1,...z n ,...,z N }, where z n represents the nth page, N represents the number of pages; if user u m Visited pages n , then construct user u m Visit page n The link matrix ; Step 1-2, based on the link matrix , calculate user u m For page z n The relationship representation index S mn ; Step 1-3: Relationship representation index S mn After being processed by the Embedding layer, it is converted into a relation representation vector .

3. The information recommendation method of the government service platform according to claim 2 is characterized by: The step 1-1 specifically includes the following steps: If user u m Visited pages n , then the link matrix , construct user u m Visit page n The link matrix : in, is the number of interactions; is the number of sessions; is the number of entries; The duration of the most recent stay; is the maximum length of stay; For reference duration; is the page type feature vector; Rate the page; The frequency of entry.

4. The information recommendation method of the government service platform according to claim 3 is characterized by: The steps 1-2 specifically include the following steps: The first relational indicator : Among them, v=1,2,3; Indicates browsing time coefficient, used for balancing The size of ; Second relationship characterization index : in, Indicates page z n The type feature vector, r represents page z n Type; The third relationship representation index : User m For page z n The relationship representation index S mn for: in, They are The weight coefficient is used to balance the importance of the first relationship representation index, the second relationship representation index and the third relationship representation index.

5. The information recommendation method of the government service platform according to claim 4, characterized in that: The step 2 specifically includes the following steps: Let H d Represents a d-dimensional hyperbolic space with curvature c=-1 and constant k=-1 / c. Set the reference point to ; Represent the relationship vector Mapping to the tangent plane of hyperbolic space: in, Representing the relationship vector Hyperbolic characteristics of ;||.|| L represents the L-order norm; [u m ,z n ] indicates user u m Link to page z n access operation; cosh represents the hyperbolic cosine function; sinh represents the hyperbolic sine function.

Images