Website registration method and apparatus, electronic device, and storage medium

Abstract

The present application relates to the technical field of communications, and provides a website registration method and apparatus, an electronic device, and a storage medium. The method comprises: during website registration, first, on the basis of respective types of a plurality of information input boxes on a registration webpage of a website, adding registration information respectively into the plurality of information input boxes, and sending the registration information to a website server; receiving registration prompt information sent by the website server; when the registration prompt information represents that at least one piece of registration information is inconsistent with a corresponding registration specification, inputting the registration prompt information into an attribute fusion deep learning model, so as to obtain a target prompt type and a target registration specification; and then on the basis of the target prompt type and the target registration specification, correcting target registration information, and performing website registration on the basis of registration information consistent with the registration specification and the corrected target registration information. By using the technical solution provided by the present application, website registration efficiency can be effectively improved.

Description

Website registration method and device, electronic equipment and storage medium

[0001] Cross-reference to Related Applications

[0002] This application is based on and claims priority to Chinese Patent Application No. 202411247008.8, filed on September 6, 2024, the entire contents of which are incorporated herein by reference. TECHNICAL FIELD

[0003] The present application relates to the field of communication technology, and in particular to a website registration method and device, electronic equipment and storage medium. BACKGROUND

[0004] The Internet is a vast network that covers hundreds of millions of users worldwide, connecting people together and enabling information sharing, exchange and collaboration. Websites are the most important part of the Internet, providing various information and services to users through web pages.

[0005] With the continuous development of Internet technology, the content and form of websites have become increasingly rich and diverse. From early static web pages to current dynamic interactive web pages, whether it is to obtain news, shopping, socializing or working, websites are indispensable. In order to use the services provided by these websites, users need to register, which not only facilitates users to obtain personalized services, but also helps websites to manage and maintain user information.

[0006] Traditional website registration methods usually require users to manually fill in a large amount of registration information, which results in low registration efficiency. SUMMARY

[0007] The present application provides a website registration method, device, electronic equipment and storage medium to solve the problem of low registration efficiency caused by the need for users to manually fill in a large amount of registration information in related technologies.

[0008] The present application provides a website registration method applied to an electronic device, the method comprising:

[0009] Based on the types of the plurality of information input boxes of the registration webpage of the website, adding registration information in the plurality of information input boxes respectively, and sending each registration information to a website server;

[0010] Receiving registration prompt information sent by the website server, the registration prompt information being determined based on each registration information and the corresponding registration specification of each information input box;

[0011] In a case where the registration prompt information is used to represent that there is at least one registration information inconsistent with the corresponding registration specification, the registration prompt information is input into an attribute fusion deep learning model to obtain a target prompt type and a target registration specification, the target prompt type is a type corresponding to target registration information inconsistent with the corresponding registration specification, and the target registration specification is the corresponding registration specification of the target registration information.

[0012] Based on the target prompt type and the target registration specification, the target registration information is corrected, and website registration is performed based on registration information consistent with the corresponding registration specification and the corrected target registration information.

[0013] The application also provides a website registration device applied to an electronic device, the device comprising:

[0014] A first processing unit is configured to add registration information in a plurality of information input boxes of a registration webpage of a website based on a type of each information input box.

[0015] A sending unit is configured to send each registration information to a website server.

[0016] A receiving unit is configured to receive registration prompt information sent by the website server, the registration prompt information being determined based on each registration information and a corresponding registration specification of each information input box.

[0017] A second processing unit is configured to input the registration prompt information into an attribute fusion deep learning model to obtain a target prompt type and a target registration specification in a case where the registration prompt information is used to represent that there is at least one registration information inconsistent with the corresponding registration specification, the target prompt type being a type corresponding to target registration information inconsistent with the corresponding registration specification, and the target registration specification being the corresponding registration specification of the target registration information.

[0018] A registration unit is configured to correct the target registration information based on the target prompt type and the target registration specification, and perform website registration based on registration information consistent with the corresponding registration specification and the corrected target registration information.

[0019] The application also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the website registration method of any of the above.

[0020] The application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the website registration method of any of the above.

[0021] The application further provides a computer program product comprising a computer program which, when executed by a processor, implements the website registration method according to any one of the above. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the application or the related art, the drawings needed to be used in the embodiments or the related art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0023] Fig. 1 is a flowchart of a website registration method provided by an embodiment of the application.

[0024] Fig. 2 is a processing framework diagram of a modal dialog box appearing when a click operation is performed on an information input box according to an embodiment of the application.

[0025] Fig. 3 is a flowchart of obtaining a target prompt type and a target registration specification according to an embodiment of the application.

[0026] Fig. 4 is a structural diagram of an attribute fusion deep learning model according to an embodiment of the application.

[0027] Fig. 5 is a structural diagram of a website registration device according to an embodiment of the application.

[0028] Fig. 6 is a structural diagram of an electronic device according to an embodiment of the application. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of the application more clear, the technical solutions in the application will be described clearly and completely in the following with reference to the drawings in the application. Obviously, the described embodiments are some embodiments of the application, but not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0030] In the embodiments of the application, “at least one” means one or more, and “multiple” means two or more. “And / or” describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the following three cases: A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. In the textual description of the application, the character “ / ” generally represents an “or” relationship between the associated objects before and after it.

[0031] The technical scheme provided in the embodiments of the present application can be applied to a website registration scenario. In order to use the services provided by a website, a user needs to register the website, which not only facilitates the user to obtain personalized services, but also helps the website to manage and maintain user information.

[0032] The traditional website registration method usually requires a user to manually fill in a large amount of registration information, which results in low registration efficiency.

[0033] In order to realize automatic registration of a website and improve the registration efficiency, the embodiments of the present application provide a website registration method. In the following, the website registration method provided in the present application will be described in detail through the following embodiments. It can be understood that the following embodiments can be combined with each other, and the same or similar concepts or processes can not be described in some embodiments.

[0034] FIG. 1 is a flowchart of a website registration method provided in the embodiments of the present application, which is applied to an electronic device. For example, as shown in FIG. 1, the website registration method can include the following steps.

[0035] In S101, based on the types of the plurality of information input boxes of the registration webpage of the website, registration information is added in the plurality of information input boxes respectively, and each registration information is sent to the website server.

[0036] For example, the plurality of information input boxes can include a username input box, a password input box, a verification code input box, or other input boxes, etc. Correspondingly, the types of the information input boxes can include a username prompt type, a password prompt type, a verification code prompt type, or other prompt types, etc., which can be set according to actual needs.

[0037] In order to realize automatic registration of a website, the electronic device can add registration information in the plurality of information input boxes respectively based on the types of the plurality of information input boxes. For example, when the type of the information input box is the username prompt type, the user can input the username to be registered in the information input box; when the type of the information input box is the password prompt type, the user can input the password in the information input box; and when the type of the information input box is the verification code prompt type, the user can input the verification code in the information input box.

[0038] For example, in the embodiments of the present application, when the username and the password to be registered are input, the username and the password can be automatically generated based on a pre-constructed user information library. The user information library is usually constructed by analyzing the registration prompts of a plurality of websites to meet the registration requirements of most websites.

[0039] After adding the registration information in the plurality of information input boxes respectively, the added registration information can be sent to the registration server, so that the registration server determines whether the received registration information conforms to the corresponding registration specification based on the received registration information, and sends registration prompt information to the electronic device based on the determination result.

[0040] In S102, the registration prompt information sent by the website server is received, and the registration prompt information is determined based on each registration information and the registration specification corresponding to each information input box.

[0041] For example, the registration prompt information can be used to represent that the registration is successful, that is, each registration information is consistent with the corresponding registration specification and meets the registration requirements; or the registration prompt information can also be used to represent that among the plurality of registration information, at least one registration information is inconsistent with the corresponding registration specification, in which case, the electronic device can determine the target prompt type and the target registration specification by means of the attribute fusion deep learning model, that is, perform S103 as follows:

[0042] In S103, in the case where the registration prompt information is used to represent that there is at least one registration information inconsistent with the corresponding registration specification, the registration prompt information is input into the attribute fusion deep learning model to obtain the target prompt type and the target registration specification.

[0043] The target prompt type is the type corresponding to the target registration information inconsistent with the registration specification, and the target registration specification is the registration specification corresponding to the target registration information.

[0044] For example, in the embodiments of the present application, the target prompt type can be a username prompt type, a password prompt type, a verification code prompt type or other prompt types, etc., which can be set according to actual needs.

[0045] For example, the attribute fusion deep learning model can be a deep learning-based model, which can be set according to actual needs.

[0046] For example, if the target prompt type output by the attribute fusion deep learning model is a username prompt type, it indicates that the target registration information inconsistent with the registration specification is a username, and the corresponding target registration specification is the registration specification corresponding to the username information input box, for example, a string composed of letters and numbers; if the target prompt type output by the attribute fusion deep learning model is a password prompt type, it indicates that the target registration information inconsistent with the registration specification is a password, and the corresponding target registration specification is the registration specification corresponding to the password information input box, for example, a string composed of uppercase letters, lowercase letters, and numbers with a length of no less than 8; if the target prompt type output by the attribute fusion deep learning model is a verification code prompt type, it indicates that the target registration information inconsistent with the registration specification is a verification code, and the corresponding target registration specification is the registration specification corresponding to the verification code information input box, for example, a string composed of 6 digits.

[0047] After determining the target prompt type and the target registration specification by means of the attribute fusion deep learning model, the target registration information inconsistent with the registration specification can be corrected based on the target prompt type and the target registration specification, that is, S104 is executed to realize automatic registration of the website, thereby solving the problem of low registration efficiency caused by the need for manual filling of a large amount of registration information in the related art, and effectively improving the registration efficiency.

[0048] S104, correcting the target registration information based on the target prompt type and the target registration specification, and performing website registration based on the registration information consistent with the registration specification and the corrected target registration information.

[0049] For example, if the target prompt type output by the attribute fusion deep learning model is a username prompt type, the target registration information inconsistent with the registration specification can be determined to be a username based on the target prompt type, and the added username can be corrected based on the corresponding username registration specification, for example, a string composed of letters and numbers, to obtain a corrected username, and then the website can be registered together with the registration information consistent with the registration specification, for example, the password and the verification code.

[0050] For example, if the target prompt type output by the attribute fusion deep learning model is a password prompt type, the target registration information inconsistent with the registration specification can be determined to be a password based on the target prompt type, and the added password can be corrected based on the registration specification corresponding to the password information input box, for example, a string composed of uppercase letters, lowercase letters, and numbers with a length of no less than 8, to obtain a corrected password, and then the website can be registered together with the registration information consistent with the registration specification, for example, the username and the verification code.

[0051] It can be seen that in the embodiments of the application, when registering a website, the registration information can be added in the plurality of information input boxes based on the types of the plurality of information input boxes of the registration webpage of the website respectively, and each registration information is sent to the website server; and the registration prompt information sent by the website server is received; when the registration prompt information indicates that there is at least one registration information inconsistent with the corresponding registration specification, the registration prompt information is input into the attribute fusion deep learning model to obtain the target prompt type and the target registration specification, and the target registration information is corrected based on the target prompt type and the target registration specification, and the website registration is performed based on the registration information consistent with the registration specification and the corrected target registration information. In this way, with the help of the attribute fusion deep learning model, the target registration information can be automatically corrected based on the target prompt type and the target registration specification when there is at least one registration information inconsistent with the corresponding registration specification, not only solving the problem of low registration efficiency caused by the need for manual filling of a large amount of registration information in related technologies, but also solving the problem of process interruption caused by unsuccessful website registration, thereby effectively improving the registration efficiency.

[0052] Based on the embodiment shown in FIG. 1, in the above S101, when adding the registration information in the plurality of information input boxes based on the types of the plurality of information input boxes of the registration webpage of the website, considering that many websites (including illegal websites) usually display activities and preferential information on the webpage to guide new users to register and encourage existing users to continue to participate, these information generally appears in the form of a modal dialog box, and in view of the characteristics of this modal dialog forced interaction, in the embodiments of the application, a modal exception automatic correction branch process can be added, including: for each information input box, in response to a click operation on the information input box, determining a target solution in the case of outputting a modal dialog box; processing the modal dialog box based on the target solution, and adding the registration information in the information input box in the case that the modal dialog box has been processed, so that it can be started automatically without human intervention, and the influence of the exception on the website automation process can be minimized.

[0053] For example, the target solution can be an exception capture processing solution of an element outside the information input box, which can be set according to actual needs.

[0054] For example, see Figure 2, which is a schematic diagram of a processing framework for displaying a modal dialog box when a click operation is performed on an information input box according to an embodiment of this application. When a click operation is performed on the information input box, it can first be determined whether the click operation is successful. If the click is successful, the Document Object Model (DOM) structure before and after the click operation is compared to determine whether the registration page has changed, thus determining whether the click operation is effective. The DOM displays the HTML document in a tree structure, including elements, attributes, and text nodes, reflecting all page changes. A depth-first traversal is used to parse the HTML text and save all nodes in the form of [node tag, node object, [list of child nodes]]. A breadth-first traversal is used to compare and identify changed nodes level by level, checking for additions or deletions. If nodes are added or deleted, it is determined that the registration page has changed. If the registration page changes, the click operation is confirmed. If the click fails, a modal dialog box interrupts the automated registration process, and feedback is received. Further analysis of the feedback and automatic determination of the target solution are performed. If the target solution successfully handles the modal dialog box, the click operation on the information input box is performed again, and the click operation is completed. If the target solution fails to handle the modal dialog box, the entire page is analyzed, the target solution is redefined, and the click operation on the information input box is performed again, and the click operation is completed. If the registration page remains unchanged, a modal dialog box interrupts the automated registration process, and there is no feedback. Further analysis of the entire page and redefined of the target solution are performed, and the click operation on the information input box is performed again, and the click operation is completed. This automatic correction of branch processes by adding modal exceptions solves the problem of process interruption caused by certain mandatory user responses that prevent other operations from being performed before the response. Furthermore, it can automatically start without manual intervention, minimizing the impact of exceptions on the automated website process.

[0055] For example, in this application embodiment, the click operation identifier used to trigger a specific function on the registration page can be defined as a "button," and the buttons described below all belong to this type of identifier. Buttons can be divided into three main categories: image buttons, whose function is triggered by clicking on an image element in the webpage; non-image buttons, whose working method is similar to the former; and submit buttons, which have various presentation methods, but all implement the form submission function, and differ from the aforementioned two types of buttons.

[0056] For example, the image button retrieval logic follows a three-stage process: First, execute... The first step is to filter the tag elements; the second step is to extract the source file path and image class name and filter based on keywords; the third step is to parse and output the corresponding path of the determined element.

[0057] Based on the embodiment shown in FIG. 1, the registration prompt information may, for example, include HyperText Markup Language (HTML) text, style attributes of nodes in a website, and a hierarchy of the nodes.

[0058] The style attributes of the nodes may, for example, include a location of the nodes, a graphic size, a pop-up display manner, and the like; and the hierarchy of the nodes may include labels and attribute values of all child nodes in the nodes, and a tree structure of the entire nodes.

[0059] In a case where the registration prompt information includes the HTML text, the style attributes of the nodes in the website, and the hierarchy of the nodes, in S103, the registration prompt information is input to the attribute fusion deep learning model to obtain the target prompt type and the target registration specification. For details, refer to an embodiment shown in FIG. 2.

[0060] FIG. 3 is a flowchart of a method for obtaining a target prompt type and a target registration specification according to an embodiment of the present application. The method may include the following steps.

[0061] In S301, the HTML text is input to a first embedding layer in the attribute fusion deep learning model to obtain a first text feature vector; and the first text feature vector is input to an activation layer to obtain the target registration specification.

[0062] For example, in an embodiment of the present application, referring to FIG. 4, which is a structural diagram of an attribute fusion deep learning model according to an embodiment of the present application, the attribute fusion deep learning model may include a first embedding layer, a second embedding layer, a third embedding layer, an activation layer, and a post-processing module.

[0063] In a case where the target registration specification is obtained from the HTML text, the HTML text is input to the first embedding layer in the attribute fusion deep learning model, the HTML text is converted into the first text feature vector by the first embedding layer, and the converted first text feature vector is input to the activation layer to obtain the target registration specification. In this way, the target registration specification corresponding to the target registration information inconsistent with the registration specification can be obtained based on the HTML text.

[0064] Next, in combination with S302-S304, the target prompt type corresponding to the target registration information inconsistent with the registration specification can be obtained, so that the target registration information can be automatically corrected based on the target prompt type and the target registration specification in the future. This not only solves the problem that a user needs to manually fill in a large amount of registration information in the related art, resulting in a low registration efficiency, but also solves the problem that the process is interrupted due to unsuccessful registration prompted by a website, thereby effectively improving the registration efficiency.

[0065] S302, input the style attribute of the node into a second embedding layer in the attribute fusion deep learning model to obtain a string feature vector.

[0066] For example, in the embodiment of the present application, the style attribute of the node includes numerical and categorical types. When the style attribute of the node is input into the second embedding layer in the attribute fusion deep learning model to obtain a string feature vector, the style attribute of the node can be first input into the second embedding layer in the attribute fusion deep learning model. For the numerical type of the style attribute, the numerical type of the style attribute is normalized to obtain the normalized style attribute. For the categorical type of the style attribute, the categorical type of the style attribute is encoded to obtain the encoded style attribute. The normalized style attribute and the encoded style attribute are spliced to obtain the string feature vector.

[0067] For example, for the numerical type of the style attribute, such as node positioning or size, the numerical type of the style attribute is normalized. For example, the following formula 1 can be referred to:

[0068] Where y represents the normalized style attribute, x represents the numerical type of the style attribute, min(x) represents the smallest style attribute, and max(x) represents the largest style attribute.

[0069] For example, for the categorical type of the style attribute, such as display mode and element transformation attribute, the categorical type of the style attribute can be one-hot encoded to obtain the encoded style attribute. The normalized style attribute and the encoded style attribute are spliced to obtain the string feature vector.

[0070] Suppose the numerical type of the style attribute is denoted as classA, and the categorical type of the style attribute is denoted as classB, where, and Then the normalized style attribute and the encoded style attribute are spliced to obtain the string feature vector: [attri a1,attri b j ]->[1,0,0,…,0,…,0,1] 1×(i+j) . attri a i represents the a i th numerical type of the style attribute, and attri b j represents the b j th categorical type of the style attribute.

[0071] S303, input the hierarchy of the node into a third embedding layer in the attribute fusion deep learning model to obtain a string feature matrix.

[0072] For example, in the embodiment of the present application, when the hierarchical structure of the nodes is input into the third embedding layer in the attribute fusion deep learning model to obtain the string feature matrix, the hierarchical structure of the nodes can be first input into the third embedding layer in the attribute fusion deep learning model to determine the string corresponding to each node, the string corresponding to the node including the label and the attribute value of the node; and based on the string set composed of the string corresponding to each node, the character-level feature matrix corresponding to the hierarchical structure of the nodes is determined; and based on the string set, the word-level feature matrix corresponding to the hierarchical structure of the nodes is determined; and then the character-level feature matrix and the word-level feature matrix are weighted and spliced to obtain the string feature matrix.

[0073] Generally, the hierarchical structure of the nodes will allocate a unique identifier to each node of the tree structure thereof, and a vector is created for each node, the vector being a string composed of the label and the attribute value of the node, therefore, when the hierarchical structure of the nodes is input into the third embedding layer in the attribute fusion deep learning model to determine the string corresponding to each node, it is assumed that the attribute value of node 1 can be represented by strings stringA and stringB, then the string corresponding to node 1 can be recorded as <tag1: attriA = stringA, attriB = stringB>; it is assumed that the attribute value of node 2 can be represented by string stringC, then the string corresponding to node 2 can be recorded as <tag2: attriC = stringC>. <tag2:attric>Assuming that the attribute values of node 3 can be represented by strings stringD and stringE, the string corresponding to node 3 can be denoted as <tag3: attriD = stringD, attriE = stringE>.

[0074] After determining the string corresponding to each node respectively, the string corresponding to each node can be converted into a corresponding string set based on the hierarchical structure of the nodes. In an embodiment of the present application, the string set can be denoted as U.

[0075] For example, in an embodiment of the present application, when determining the character-level feature matrix corresponding to the hierarchical structure of the nodes based on the string set composed of the strings corresponding to the nodes, the tokenization operation can be performed on each string in the string set composed of the strings corresponding to the nodes, to obtain a tokenization result set. The character-level vector corresponding to the string set can be generated according to the frequency of the character elements in the tokenization result set. The dimensions of all vectors in the character-level vector are adjusted to obtain a vector set, in which the dimensions of all vectors are the same. All vectors in the vector set are encoded, and the matrix obtained by the encoding is processed by dimension reduction to obtain a dense matrix. The character-level feature matrix is obtained by feature extraction on the dense matrix.

[0076] For example, when determining the character-level feature matrix based on the string set composed of the strings corresponding to the nodes, the tokenization operation can be performed on all tags in the string set U by characters. The tokenization result set is denoted as V, representing the set of all characters appearing in the tag set. The frequency of the character elements in the tokenization result set V is assigned a number 1, 2, …, n, to generate the character-level vector corresponding to the string set. The value of n can be set according to actual needs. Secondly, padding operation is performed on all vectors in the character-level vector to fix the dimensions of all tag vectors. For vectors with insufficient length, 0 is added in front of the vectors. For vectors with excessive length, the part exceeding the length of the vector is truncated. Finally, the vector set W is formed. As shown in the following formula 2, the dimensions of all vectors in the vector set W are the same. W = {w1, w2, w3, …, wi} Formula 2

[0077] wherein w1 represents the first vector, and wi represents the i-th vector.

[0078] In addition, one-hot encoding is performed on all vectors in the vector set to form a one-hot matrix G. In order to construct a dense low-dimensional feature matrix, the matrix G is processed by dimension reduction to obtain a dense matrix S, as shown in the following formula 3.

[0079] Wherein, V1 is a dimension reduction matrix, used for dimension reduction processing of the n2-dimensional matrix G, to reduce its dimension to v2, to obtain the dense matrix S.

[0080] The dense matrix S is subjected to feature extraction by TextCNN one-dimensional convolution to obtain a character-level feature matrix CV, which can be seen from the following formula 4: CV={cv1, cv2, cv3,..., cvi} Formula 4

[0081] Wherein, cv1 represents the first character-level feature, and cvi represents the i-th character-level feature.

[0082] For example, in the embodiment of the application, when determining the word-level feature matrix corresponding to the hierarchical structure of the nodes based on the string set, the strings in the string set formed by the strings corresponding to the nodes can be segmented first to obtain a segmentation matrix, and the lengths of all segmentation results in the segmentation matrix are the same; the website word vector matrix corresponding to the segmentation matrix is subjected to dimension reduction processing to obtain a low-dimensional word vector matrix; and the low-dimensional word vector matrix is subjected to feature extraction to obtain a word-level feature matrix.

[0083] For example, when determining the word-level feature matrix corresponding to the hierarchical structure of the nodes based on the string set, the strings in the string set U can be segmented according to the corpus of the BERT pre-training model, a maximum segmentation length is set, the part exceeding the maximum length is directly deleted, and the part less than the maximum length is automatically supplemented with 0 in subsequent input to the BERT model to obtain a segmentation matrix T. T={t1, t2, t3,..., ti} Formula 5

[0084] Wherein, t1 represents the first segmentation, and ti represents the i-th segmentation.

[0085] Secondly, each row ti in the segmentation matrix T is converted into idSequence to obtain an ids matrix, and then 12 hidden layers are obtained through the pre-trained BERT model, and a 768-dimensional website word vector matrix Word facing prompt information is obtained by summing the last four rows, which can be seen from the following formula 6: Word={word1, word2, word3,..., wordi} Formula 6

[0086] Wherein, word1 represents the first website word vector, and wordi represents the i-th website word vector.

[0087] Then, the high-dimensional website word vector matrix Word is subjected to dimension reduction processing to obtain a low-dimensional word vector matrix R, which can be seen from the following formula 7:

[0088] wherein, W represents a high-dimensional website word vector matrix Word, V2 is a dimension reduction matrix, used for dimension reduction processing of the high-dimensional website word vector matrix Word in L B dimension reduction processing of the high-dimensional website word vector matrix Word in L

[0089] The low-dimensional word vector matrix R is subjected to feature extraction through a one-dimensional convolution of TextCNN to obtain a word-level feature matrix WV, which can be seen from the following formula 8: WV={wv1, wv2, wv3, …, wvi} Formula 8

[0090] wherein, wv1 represents the first word-level feature, and wvi represents the i-th word-level feature.

[0091] The character-level feature matrix CV and the word-level feature matrix WV are then weighted and spliced to obtain a string feature matrix.

[0092] In combination with the above description, after obtaining the first text feature vector, the string feature vector and the string feature matrix respectively, the following S304 can be executed:

[0093] S304, inputting the first text feature vector, the string feature vector and the string feature matrix into a post-processing module in the attribute fusion deep learning model to obtain a target prompt type.

[0094] It should be noted that, in the embodiments of the present application, the operation of inputting the first text feature vector into the activation layer to obtain the target registration specification in the above S301 has no sequence with the operation of obtaining the target prompt type in the above S302-S304. In this embodiment of the present application, only the operation of inputting the first text feature vector into the activation layer to obtain the target registration specification, and then executing the operation of obtaining the target prompt type in the above S202-S304 is taken as an example for description, but it does not mean that the embodiments of the present application are limited to this.

[0095] As shown in the above FIG. 4, for example, in the embodiments of the present application, the post-processing module can include a convolution layer, a max-pooling layer, a regularization layer and an activation layer connected in sequence, which can be set according to actual needs.

[0096] When the first text feature vector, the string feature vector, and the string feature matrix are input into the post-processing module in the attribute fusion deep learning model, the first text feature vector, the string feature vector, and the string feature matrix can be subjected to global pooling and a full connection layer for each feature first; secondly, a connection activation function is used to generate feature weight scores, so as to achieve the effect of weighting for each feature; thirdly, a global self-attention module is used to obtain better global information. Then, a convolution layer is used to capture local patterns in the vector, a pooling layer is used to reduce the dimension of the feature vector, and a dropout layer is used to avoid model overfitting; finally, the target prompt type is output.

[0097] In the case where at least one registration information is inconsistent with the corresponding registration specification, after the target prompt type and the target registration specification are determined by means of the attribute fusion deep learning model, the target registration information can be corrected based on the target prompt type and the target registration specification, that is, S104 described above is executed.

[0098] For example, in S104 described above, when the target registration information is corrected based on the target prompt type and the target registration specification, the label sequence of all nodes in the website can be first converted into a corresponding index vector, and the index vector and the HTML text can be fused and spliced to obtain a fusion sequence; the fusion sequence can be input into a large language model to obtain a correction strategy for the target registration information; and the target registration information can be corrected based on the target prompt type, the target registration specification, and the correction strategy. In this way, not only does the problem of low registration efficiency caused by the need for users to manually fill in a large amount of registration information in related technologies be solved, but also the problem of process interruption caused by unsuccessful website registration prompts is solved, thereby effectively improving the registration efficiency.

[0099] For example, when the label sequence of all nodes in the website is converted into a corresponding index vector, a label dictionary for registration information prompts can be constructed in advance, wherein all information prompt related labels in the current HTML specification and five special identifiers [SEP], [LEFT], [RIGHT], [UNK], and [PAD] are screened, a total of 64. In this way, the label dictionary can be combined to convert and map the label sequence of all nodes into a corresponding index vector, and the index vector and the HTML text can be fused and spliced to obtain a fusion sequence, which is ready to be input into a large language model to output a correction strategy for the target registration information by the large language model.

[0100] For example, in the embodiments of the present application, in order to enable the large language model to fully understand the vertical scene of website registration, the prompt paradigm enhancement method of multi-concept combination can be used to replace the prompt type with a set of concept words or phrases with the same meaning, such as "character length" replaced with "password length", "username length", "within a certain number of characters", etc., to enrich the semantic meaning of the target, so that the large language model obtained by pre-training can learn more related knowledge. Given the prompt template T(·) and the type word set V(character length, special character, uppercase letter, number, occupied, space, etc.), for each node fusion sequence x={x1, x2,...,x n} and the concept word combination {concept1, concept2,..., conceptk}, where k represents the kth concept word. First, the fusion sequence x of the node and each concept word are combined using the given prompt template to map to the prompt input x concepti =T(x, concepti)={[CLS], x1, x2,...,x n .concepti, [MASK], [SEP]}; then each prompt input is fed into a separate large language model M to obtain the hidden layer representation h [MASK] of [MASK], and the probability of the registration prompt type at [MASK] is:

[0101] where p i ([MASK]=v|x concepti ) represents the probability of v under x concepti , and v is the encoding vector of v in the pre-trained large language model M. After calculating the mask prediction probability, the probability is normalized using the softmax function.

[0102] Finally, the activation layer output of the large language model obtains the correction strategy of the target registration information. The label word probability learned by each concept word through prompt learning is different due to the different concept words and learned knowledge, so they need to be fused to obtain the overall type probability:

[0103] Through the above large language model, the correction strategy of the target registration information can be determined, that is, the adjustment direction of the target registration information, so as to jointly correct the target registration information based on the target prompt type and the target registration specification, and reattempt registration until successful registration. This not only solves the problem of low registration efficiency caused by manual filling of a large amount of registration information in related technologies, but also solves the problem of process interruption caused by unsuccessful website prompt registration, thereby effectively improving the registration efficiency.

[0104] The website registration apparatus provided in the present application is described below, and the website registration apparatus described below can be referred to in correspondence with the website registration method described above.

[0105] FIG. 5 is a structural schematic diagram of a website registration apparatus provided in an embodiment of the present application, which is applied to an electronic device. For example, as shown in FIG. 5, the website registration apparatus 50 can include:

[0106] The first processing unit 501 is configured to add registration information in the plurality of information input boxes respectively based on the types of the plurality of information input boxes of the registration webpage of the website;

[0107] The sending unit 502 is configured to send the registration information to the website server;

[0108] The receiving unit 503 is configured to receive registration prompt information sent by the website server, the registration prompt information being determined based on the registration information and the registration specification corresponding to the information input box;

[0109] The second processing unit 504 is configured to, when the registration prompt information indicates that at least one of the registration information is inconsistent with the corresponding registration specification, input the registration prompt information into an attribute fusion deep learning model to obtain a target prompt type and a target registration specification, the target prompt type being a type corresponding to target registration information that is inconsistent with the registration specification, and the target registration specification being a registration specification corresponding to the target registration information;

[0110] The registration unit 505 is configured to correct the target registration information based on the target prompt type and the target registration specification, and perform website registration based on the registration information that is consistent with the registration specification and the corrected target registration information.

[0111] For example, in the embodiment of the present application, the registration prompt information includes HTML text, style attributes of a node in the website, and a hierarchical structure of the node.

[0112] When the registration prompt information is input into the attribute fusion deep learning model to obtain the target prompt type corresponding to the target registration information that is inconsistent with the registration specification and the target registration specification, the second processing unit 504 is further configured to:

[0113] input the HTML text into a first embedding layer in the attribute fusion deep learning model to obtain a first text feature vector, and input the first text feature vector into an activation layer to obtain the target registration specification;

[0114] input the style attribute of the node into a second embedding layer in the attribute fusion deep learning model to obtain a string feature vector;

[0115] input the hierarchical structure of the node into a third embedding layer in the attribute fusion deep learning model to obtain a string feature matrix;

[0116] input the first text feature vector, the string feature vector and the string feature matrix into a post-processing module in the attribute fusion deep learning model to obtain the target prompt type.

[0117] For example, in the embodiments of the present application, the style attribute of the node includes numerical type and category type;

[0118] In the process of inputting the style attribute of the node into the second embedding layer in the attribute fusion deep learning model to obtain the string feature vector, the second processing unit 504 is further configured to:

[0119] input the style attribute of the node into the second embedding layer in the attribute fusion deep learning model, and for the numerical type of the style attribute, perform normalization processing on the numerical type of the style attribute to obtain the normalized style attribute;

[0120] for the category type of the style attribute, encode the category type of the style attribute to obtain the encoded style attribute;

[0121] splice the normalized style attribute and the encoded style attribute to obtain the string feature vector.

[0122] For example, in the embodiments of the present application, in the process of inputting the hierarchical structure of the node into the third embedding layer in the attribute fusion deep learning model to obtain the string feature matrix, the second processing unit 504 is further configured to:

[0123] input the hierarchical structure of the node into the third embedding layer in the attribute fusion deep learning model to determine the string corresponding to each node;

[0124] determine the character-level feature matrix corresponding to the hierarchical structure of the node based on the string set composed of the strings corresponding to the nodes, and determine the word-level feature matrix corresponding to the hierarchical structure of the node based on the string set;

[0125] weight splice the character-level feature matrix and the word-level feature matrix to obtain the string feature matrix.

[0126] For example, in the embodiment of the present application, when the character-level feature matrix corresponding to the hierarchical structure of the nodes is determined based on the string set composed of the strings corresponding to the nodes, the second processing unit 504 is further configured to:

[0127] performing word segmentation on each string in the string set composed of the strings corresponding to the nodes to obtain a word segmentation result set;

[0128] generating a character-level vector corresponding to the string set according to the frequency of occurrence of the character elements in the word segmentation result set;

[0129] adjusting the dimensions of all vectors in the character-level vector to obtain a vector set, wherein the dimensions of all vectors in the vector set are the same;

[0130] encoding all vectors in the vector set and performing dimensionality reduction processing on the matrix obtained by encoding to obtain a dense matrix;

[0131] performing feature extraction on the dense matrix to obtain the character-level feature matrix.

[0132] For example, in the embodiment of the present application, when the word-level feature matrix corresponding to the hierarchical structure of the nodes is determined based on the string set, the second processing unit 504 is further configured to:

[0133] performing word segmentation on each string in the string set composed of the strings corresponding to the nodes to obtain a word segmentation matrix, wherein the lengths of all word segmentation results in the word segmentation matrix are the same;

[0134] performing dimensionality reduction processing on a website word vector matrix corresponding to the word segmentation matrix to obtain a low-dimensional word vector matrix;

[0135] performing feature extraction on the low-dimensional word vector matrix to obtain the word-level feature matrix.

[0136] For example, in the embodiment of the present application, when the target registration information is corrected based on the target prompt type and the target registration specification, the registration unit 505 is further configured to:

[0137] convert the label sequence of all nodes in the website into a corresponding index vector, and fuse and splice the index vector with the HTML text to obtain a fusion sequence;

[0138] inputting the fusion sequence into a large language model to obtain a correction strategy of the target registration information;

[0139] correcting the target registration information based on the target prompt type, the target registration specification, and the correction strategy.

[0140] For example, in the embodiments of the present application, when the types of the plurality of information input boxes of the website-based registration web page are executed, the first processing unit 501 is further configured to:

[0141] For each of the information input boxes, in response to a click operation on the information input box, the target solution is determined in the case of an output modal dialog box;

[0142] The modal dialog box is processed based on the target solution, and the registration information is added in the information input box in the case that the modal dialog box has been processed.

[0143] The website registration device 50 provided by the embodiments of the present application can execute the technical solutions of the website registration method in any of the above embodiments, and the implementation principles and beneficial effects thereof are similar to those of the website registration method. For details, refer to the implementation principles and beneficial effects of the website registration method, which will not be described here.

[0144] FIG. 6 is a schematic diagram of the physical structure of an electronic device according to an embodiment of the present application. As shown in FIG. 6, the electronic device can include a processor 610, a communications interface 620, a memory 630, and a communications bus 640. The processor 610, the communications interface 620, and the memory 630 can communicate with each other through the communications bus 640. The processor 610 can invoke the logical instructions in the memory 630 to execute the website registration method, which includes: based on the types of the plurality of information input boxes of the website-based registration web page, adding registration information in the plurality of information input boxes respectively, and sending each of the registration information to a website server; receiving registration prompt information sent by the website server, the registration prompt information being determined based on each of the registration information and the corresponding registration specification of each of the information input boxes; in the case that the registration prompt information indicates that there is at least one registration information inconsistent with the corresponding registration specification, inputting the registration prompt information into an attribute fusion deep learning model to obtain a target prompt type and a target registration specification, the target prompt type being a type corresponding to the target registration information inconsistent with the registration specification, and the target registration specification being a registration specification corresponding to the target registration information; based on the target prompt type and the target registration specification, correcting the target registration information, and based on the registration information consistent with the registration specification and the corrected target registration information, performing website registration.

[0145] In addition, the logic instructions in the memory 630 described above can be implemented in the form of software functional units and sold or used as independent products, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that make contributions to the related art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the operations of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0146] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to perform the website registration method provided by the above-mentioned methods. The method comprises: based on the types of a plurality of information input boxes of a registration webpage of a website, adding registration information in the plurality of information input boxes respectively, and sending each registration information to a website server; receiving registration prompt information sent by the website server, the registration prompt information being determined based on each registration information and a corresponding registration specification; in the case that the registration prompt information is used to represent that there is at least one registration information inconsistent with the corresponding registration specification, inputting the registration prompt information into an attribute fusion deep learning model to obtain a target prompt type and a target registration specification, the target prompt type being a type corresponding to a target registration information inconsistent with the registration specification, and the target registration specification being a registration specification corresponding to the target registration information; based on the target prompt type and the target registration specification, correcting the target registration information, and performing website registration based on registration information consistent with the registration specification and the corrected target registration information.

[0147] In yet another aspect, the present application also provides a non-transitory computer-readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the website registration method provided by each of the above methods, and the method comprises: based on the types of a plurality of information input boxes of a registration webpage of a website, adding registration information in the plurality of information input boxes respectively, and sending each of the registration information to a website server; receiving registration prompt information sent by the website server, the registration prompt information being determined based on each of the registration information and a corresponding registration specification of each of the information input boxes; in a case where the registration prompt information is used to represent that there is at least one registration information inconsistent with the corresponding registration specification, inputting the registration prompt information into an attribute fusion deep learning model to obtain a target prompt type and a target registration specification, the target prompt type being a type corresponding to target registration information inconsistent with the registration specification, and the target registration specification being a registration specification corresponding to the target registration information; based on the target prompt type and the target registration specification, correcting the target registration information, and performing website registration based on registration information consistent with the registration specification and the corrected target registration information.

[0148] The device embodiments described above are merely illustrative, wherein the units illustrated as separate components can or can not be physically separated, and the components illustrated as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0149] From the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software plus a necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions, essentially or in terms of related technology, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0150] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the same; although the present application has been described in detail with reference to the foregoing examples, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A website registration method applied to an electronic device, the method comprising: adding registration information in a plurality of information input boxes of a registration webpage of a website based on a type of each of the plurality of information input boxes, and sending each of the registration information to a website server; receiving registration prompt information sent by the website server, the registration prompt information being determined based on each of the registration information and a corresponding registration specification of each of the information input boxes; in a case where the registration prompt information indicates that at least one of the registration information is inconsistent with the corresponding registration specification, inputting the registration prompt information into an attribute fusion deep learning model to obtain a target prompt type and a target registration specification, the target prompt type being a type corresponding to a target registration information that is inconsistent with the corresponding registration specification, and the target registration specification being the corresponding registration specification of the target registration information; based on the target prompt type and the target registration specification, correcting the target registration information, and performing website registration based on registration information consistent with the corresponding registration specification and the corrected target registration information.

2. The website registration method of claim 1, wherein, the registration prompt information comprises HTML text, style attributes of nodes in the website, and a hierarchical structure of the nodes; the inputting of the registration prompt information into the attribute fusion deep learning model to obtain the target prompt type and the target registration specification comprises: inputting the HTML text into a first embedding layer in the attribute fusion deep learning model to obtain a first text feature vector, and inputting the first text feature vector into an activation layer to obtain the target registration specification; inputting the style attributes of the nodes into a second embedding layer in the attribute fusion deep learning model to obtain a string feature vector; inputting the hierarchical structure of the nodes into a third embedding layer in the attribute fusion deep learning model to obtain a string feature matrix; inputting the first text feature vector, the string feature vector, and the string feature matrix into a post-processing module in the attribute fusion deep learning model to obtain the target prompt type.

3. The website registration method of claim 2, wherein, the style attributes of the nodes comprise numerical types and categorical types; the inputting of the style attributes of the nodes into the second embedding layer in the attribute fusion deep learning model to obtain the string feature vector comprises: inputting the style attributes of the nodes into the second embedding layer in the attribute fusion deep learning model, performing normalization processing on the numerical type of the style attributes to obtain normalized style attributes for the numerical type of the style attributes; performing encoding on the categorical type of the style attributes to obtain encoded style attributes for the categorical type of the style attributes; splicing the normalized style attributes and the encoded style attributes to obtain the string feature vector.

4. The website registration method of claim 2, wherein, the inputting of the hierarchical structure of the nodes into the third embedding layer in the attribute fusion deep learning model to obtain the string feature matrix comprises: inputting the hierarchical structure of the nodes into the third embedding layer in the attribute fusion deep learning model to determine a string corresponding to each node. determine a character-level feature matrix corresponding to the hierarchical structure of the nodes based on a string set composed of the strings corresponding to the nodes; and determine a word-level feature matrix corresponding to the hierarchical structure of the nodes based on the string set; perform weighted splicing on the character-level feature matrix and the word-level feature matrix to obtain the string feature matrix.

5. The website registration method of claim 4, wherein, The method of determining the character-level feature matrix corresponding to the hierarchical structure of the nodes based on the string set composed of the strings corresponding to the nodes includes: perform a word segmentation operation on each string in the string set composed of the strings corresponding to the nodes to obtain a word segmentation result set; generate a character-level vector corresponding to the string set according to the frequency of occurrence of character elements in the word segmentation result set; adjust the dimensions of all vectors in the character-level vector to obtain a vector set, and the dimensions of all vectors in the vector set are the same; encode all vectors in the vector set and perform dimension reduction processing on the matrix obtained by encoding to obtain a dense matrix; perform feature extraction on the dense matrix to obtain the character-level feature matrix.

6. The website registration method of claim 4, wherein, The method of determining the word-level feature matrix corresponding to the hierarchical structure of the nodes based on the string set includes: perform a word segmentation operation on each string in the string set composed of the strings corresponding to the nodes to obtain a word segmentation matrix, and the lengths of all word segmentation results in the word segmentation matrix are the same; perform dimension reduction processing on a website word vector matrix corresponding to the word segmentation matrix to obtain a low-dimensional word vector matrix; perform feature extraction on the low-dimensional word vector matrix to obtain the word-level feature matrix.

7. The website registration method of any of claims 2-6, wherein, The method of correcting the target registration information based on the target prompt type and the target registration specification includes: convert the label sequence of all nodes in the website into a corresponding index vector, and fuse and splice the index vector with the HTML text to obtain a fusion sequence; input the fusion sequence into a large language model to obtain a correction strategy for the target registration information; correct the target registration information based on the target prompt type, the target registration specification, and the correction strategy.

8. The website registration method of any of claims 1-6, wherein, The method of adding registration information in the plurality of information input boxes of the website-based registration webpage based on the type of each information input box includes: for each information input box, in response to a click operation on the information input box, determine a target solution in the case of an output modal dialog box; based on the target solution, process the modal dialog box, and in the case that the modal dialog box has been processed, add the registration information in the information input box.

9. The website registration method of claim 2, wherein, The method of inputting the first text feature vector, the string feature vector, and the string feature matrix into a post-processing module in the attribute fusion deep learning model to obtain the target prompt type includes: perform global pooling processing on the first text feature vector, the string feature vector, and the string feature matrix respectively, and input the pooling results into a fully connected layer; processing the output of the fully connected layer using an activation function to generate feature weight scores, and weighting the first text feature vector, the string feature vector, and the string feature matrix based on the feature weight scores; inputting the weighted features into a global self-attention module to output global information; inputting the global information into a convolution layer to capture local patterns to output convolution features; inputting the convolution features into a pooling layer to reduce dimensionality to output pooled features; inputting the pooled features into a dropout layer to prevent overfitting; outputting the target prompt type based on the output features of the dropout layer.

10. A website registration apparatus applied to an electronic device, the apparatus comprising: a first processing unit configured to add registration information in a plurality of information input boxes of a registration webpage of a website based on a type of each of the plurality of information input boxes; a sending unit configured to send each of the registration information to a website server; a receiving unit configured to receive registration prompt information sent by the website server, the registration prompt information being determined based on each of the registration information and a corresponding registration specification of each of the information input boxes; a second processing unit configured to, when the registration prompt information indicates that at least one of the registration information is inconsistent with the corresponding registration specification, input the registration prompt information into an attribute fusion deep learning model to obtain a target prompt type and a target registration specification, the target prompt type being a type corresponding to a target registration information that is inconsistent with the corresponding registration specification, and the target registration specification being the corresponding registration specification of the target registration information; a registration unit configured to correct the target registration information based on the target prompt type and the target registration specification, and perform website registration based on registration information that is consistent with the corresponding registration specification and the corrected target registration information.

11. The website registration apparatus of claim 10, wherein, The registration prompt information comprises HTML text, style attributes of a node in the website, and a hierarchical structure of the node. In the step of inputting the registration prompt information into the attribute fusion deep learning model to obtain the target prompt type and the target registration specification, the second processing unit is further configured to: input the HTML text into a first embedding layer in the attribute fusion deep learning model to obtain a first text feature vector, and input the first text feature vector into an activation layer to obtain the target registration specification; input the style attributes of the node into a second embedding layer in the attribute fusion deep learning model to obtain a string feature vector; input the hierarchical structure of the node into a third embedding layer in the attribute fusion deep learning model to obtain a string feature matrix; input the first text feature vector, the string feature vector, and the string feature matrix into a post-processing module in the attribute fusion deep learning model to obtain the target prompt type.

12. The website registration apparatus of claim 11, wherein, The style attributes of the node comprise numerical and categorical types. In the execution of the inputting the style attribute of the node into the second embedding layer in the attribute fusion deep learning model to obtain a string feature vector, the second processing unit is further configured to: input the style attribute of the node into the second embedding layer in the attribute fusion deep learning model, normalize the numerical style attribute to obtain a normalized style attribute; encode the category type style attribute to obtain an encoded style attribute; concatenate the normalized style attribute and the encoded style attribute to obtain the string feature vector.

13. The website registration apparatus of claim 11, wherein, In the execution of the inputting the hierarchy of the node into the third embedding layer in the attribute fusion deep learning model to obtain a string feature matrix, the second processing unit is further configured to: input the hierarchy of the node into the third embedding layer in the attribute fusion deep learning model to determine the string corresponding to each node; determine the character-level feature matrix corresponding to the hierarchy of the node based on the string set composed of the strings corresponding to the nodes, and determine the word-level feature matrix corresponding to the hierarchy of the node based on the string set; weight and concatenate the character-level feature matrix and the word-level feature matrix to obtain the string feature matrix.

14. The website registration apparatus of claim 13, wherein, In the execution of determining the character-level feature matrix corresponding to the hierarchy of the node based on the string set composed of the strings corresponding to the nodes, the second processing unit is further configured to: perform a word segmentation operation on each string in the string set composed of the strings corresponding to the nodes to obtain a word segmentation result set; generate a character-level vector corresponding to the string set according to the frequency of the character elements in the word segmentation result set; adjust the dimensions of all vectors in the character-level vector to obtain a vector set, and the dimensions of all vectors in the vector set are the same; encode all vectors in the vector set, and perform dimension reduction processing on the matrix obtained by encoding to obtain a dense matrix; perform feature extraction on the dense matrix to obtain the character-level feature matrix.

15. The website registration apparatus of claim 13, wherein, In the execution of determining the word-level feature matrix corresponding to the hierarchy of the node based on the string set, the second processing unit is further configured to: perform word segmentation processing on each string in the string set composed of the strings corresponding to the nodes to obtain a word segmentation matrix, and the lengths of all word segmentation results in the word segmentation matrix are the same; perform dimension reduction processing on the website word vector matrix corresponding to the word segmentation matrix to obtain a low-dimensional word vector matrix; perform feature extraction on the low-dimensional word vector matrix to obtain the word-level feature matrix.

16. The website registration apparatus of any of claims 11-15, wherein, In the execution of correcting the target registration information based on the target prompt type and the target registration specification, the registration unit is further configured to: convert the label sequence of all nodes in the website into a corresponding index vector, and fuse and concatenate the index vector with the HTML text to obtain a fusion sequence; The fusion sequence is input into a large language model to obtain a correction strategy of the target registration information; Based on the target prompt type, the target registration specification and the correction strategy, the target registration information is corrected.

17. The website registration apparatus of any of claims 10-15, wherein, When the types of the plurality of information input boxes of the website-based registration webpage are executed, and the registration information is added in the plurality of information input boxes respectively, the first processing unit is further configured to: For each information input box, in response to a click operation on the information input box, determine a target solution in the case of an output modal dialog box; Based on the target solution, the modal dialog box is processed, and in the case that the modal dialog box has been processed, the registration information is added in the information input box.

18. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the website registration method of any one of claims 1 to 9.

19. A non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the website registration method of any one of claims 1 to 9.

20. A computer program product, comprising a computer program, wherein the computer program is executed by a processor to implement the website registration method of any one of claims 1 to 9.

Images