Method, system and equipment for identifying fraud-related website and medium
By combining pre-trained LXMERT network and CNN neural network with fraud-related website recognition model, the cross-modal fusion encoder and object detection algorithm is used to solve the problem of low accuracy in fraud-related website recognition in the prior art, and the diversity and Highly accurate identification of varied and multi-element fraud-related websites.
Patent Information
- Application Number
- CN202411085478.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-08
- Publication Date
- 2024-12-03
AI Technical Summary
The existing fraud-related website identification model has low accuracy when identifying diverse, variable and multi-element fraud-related websites, and cannot adapt to frequent website changes or changing web pages by changing headlines.
The pre-trained LXMERT network and CNN neural network are used to build a fraud-related website recognition model. Through the web page category identification module, the first fraud-related identification module and the second fraud-related identification module, single-element and non-single-element web pages are processed separately, and cross-modal Fusion encoder and object detection algorithm extract feature information to improve recognition accuracy.
It improves the accuracy of identification of fraud-related websites, can effectively identify diverse, variable and multi-element fraud-related websites, adapt to frequent website changes or page changes, reduce unnecessary information interference, and improve the accuracy of identification.
Smart Images

Figure CN119071030A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of network computing technology, and in particular to a method, system, device and medium for identifying fraudulent websites. Background Art
[0002] The identification of fraudulent websites generally adopts the similarity comparison method. The text, image and other information of the fraudulent websites are obtained through crawlers and other means to build a fraudulent website library. The constructed web page similarity intelligent recognition model is then used to identify the similarity between the URL to be detected and the fraudulent websites in the fraudulent website library, thereby determining the fraud risk of the URL to be detected.
[0003] Currently, the crawler data of known fraudulent URLs are mostly two-modal data of text and image content. For this reason, existing fraudulent URL identification solutions usually adopt web page similarity detection technology that combines text and image recognition to extract the web page text content and web page image content corresponding to the URL, calculate the web page similarity based on NLP text semantic similarity, semantic understanding algorithm, etc., and determine the possibility that the URL data belongs to a fraudulent website based on the similarity between the web page information of the URL data and known fraudulent websites.
[0004] Existing fraudulent website identification solutions usually use web page similarity detection technology for text and images, which results in reduced accuracy of trained fraudulent website identification models when identifying single-element web pages and web pages with more elements, and they are also unable to adapt to situations where the URL is frequently changed or the web page is changed by means of changing the header. Therefore, the existing fraudulent website identification models have low accuracy when identifying diverse, changeable and multi-element fraudulent websites.
[0005] It can be seen that how to solve the low accuracy of the existing fraudulent website identification model in identifying diverse, changeable and multi-element fraudulent websites has become a technical problem that technical personnel in this field need to solve urgently. Summary of the invention
[0006] The present invention provides a method, system, device and medium for identifying fraudulent websites to solve the technical problem that the existing fraudulent website identification model has low accuracy when identifying diverse, changeable and multi-element fraudulent websites, realizes effective identification of diverse, changeable and multi-element fraudulent websites, and improves the accuracy of fraudulent website identification.
[0007] In a first aspect, the present invention provides a method for identifying fraudulent websites, the method comprising: collecting a fraudulent website dataset and a benign website dataset, the fraudulent website dataset and the benign website dataset both comprising: URL data, web page information and attribute information;
[0008] Constructing a fraudulent website identification model, the fraudulent website identification model comprising a web page category identification module, a first fraudulent identification module and a second fraudulent identification module; the web page category identification module is used to identify the web page category of the web page information, the web page category includes a single-element web page and a non-single-element web page; the first fraudulent identification module is used to identify the non-single-element web page; the second fraudulent identification module is used to identify the single-element web page; the elements include: text, images, hyperlinks and navigation bars;
[0009] Training the fraudulent website identification model based on the fraudulent website dataset and the benign website dataset;
[0010] In the process of detecting fraudulent websites, the webpage information of the website to be detected is input into the trained fraudulent website identification model to obtain corresponding attribute information;
[0011] Determine whether the website to be detected is a fraudulent website based on the attribute information.
[0012] Preferably, the first fraud identification module includes a pre-trained LXMERT network;
[0013] The pre-trained LXMERT network includes an embedding layer, an encoding layer, an output layer and a fully connected layer;
[0014] The encoding layer includes a text encoder, an image encoder, a hyperlink encoder, a navigation bar encoder and a cross-modal fusion encoder;
[0015] Each layer of the text encoder, the image encoder, the hyperlink encoder and the navigation bar encoder includes a multi-head attention sublayer and a first feedforward neural network sublayer;
[0016] Each layer of the cross-modal fusion encoder includes a bidirectional cross-attention sub-layer, two self-attention sub-layers and two second feed-forward neural network sub-layers.
[0017] Preferably, the image encoder is used to extract features from the image in the non-single-element web page;
[0018] The step of extracting features from the image in the non-pure image webpage includes:
[0019] Perform feature recognition on the image based on the marked feature library to obtain fraud-related marks and corresponding mark categories contained in the image;
[0020] Using a target detection algorithm to partition each of the fraud-related marks, obtain the position information corresponding to each of the fraud-related marks, and the position relationship between the fraud-related marks of different mark categories;
[0021] The fraud-related mark, and the mark category, position information and position relationship corresponding to the fraud-related mark are used as image feature vectors of the image.
[0022] Preferably, the tag categories include: obfuscation tags, functional tags and business tags.
[0023] Preferably, the first fraud identification module further includes a text feature extraction module, which is connected to the text encoder through the embedding layer and is used to extract text information from the web page information;
[0024] The text feature extraction module is also used to extract the text description of the image, and input the text information and the text description into the text encoder.
[0025] Preferably, the method further comprises:
[0026] Based on the judgment result, an identification log is generated;
[0027] The identification log is sent to the registration entity corresponding to the website, the third-party fraud monitoring terminal and the cloud database respectively.
[0028] Preferably, before inputting the web page information corresponding to the URL data to be detected into the trained fraud-related website identification model, the method further includes:
[0029] Compare the URL data to be detected with a pre-built blacklist database, and deny access to the URL data to be detected that matches the blacklist database;
[0030] Compare the URL data to be detected with a pre-built whitelist database, and allow access to the URL data to be detected that matches the whitelist database;
[0031] The web page information corresponding to the URL data to be detected that does not match the blacklist database and the whitelist database is input into the trained fraud-related website identification model.
[0032] In a second aspect, the present invention further provides a fraudulent website identification system, the system comprising: a data set acquisition unit, a model construction unit, a model training unit, a model detection unit and a fraudulent website judgment unit;
[0033] The data set collection unit is used to collect a fraudulent website data set and a benign website data set, wherein the fraudulent website data set and the benign website data set both include: URL data, web page information and attribute information;
[0034] The model building unit is used to build a fraudulent website identification model, and the fraudulent website identification model includes a web page category identification module, a first fraudulent identification module, and a second fraudulent identification module; the web page category identification module is used to identify the web page category of the web page information, and the web page category includes a single-element web page and a non-single-element web page; the first fraudulent identification module is used to identify the non-single-element web page; the second fraudulent identification module is used to identify the single-element web page; the elements include: text, images, hyperlinks, and navigation bars;
[0035] The model training unit is used to train the fraudulent website identification model based on the fraudulent website data set and the benign website data set;
[0036] The model detection unit is used to input the webpage information of the website to be detected into the trained fraud-related website identification model during the fraud-related website detection process to obtain corresponding attribute information;
[0037] The fraudulent website determination unit is used to determine whether the website to be detected is a fraudulent website according to the attribute information.
[0038] In a third aspect, the present invention also provides a computer device, comprising a memory, a processor and a transceiver, which are connected via a bus; the memory is used to store a set of computer program instructions and data, and transmit the stored data to the processor, and the processor executes the program instructions stored in the memory to execute the above-mentioned fraudulent website identification method.
[0039] In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium. When the computer program is executed, the above-mentioned method for identifying fraudulent websites is implemented.
[0040] The present invention provides a method, system, device and medium for identifying fraudulent websites. Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following:
[0041] (1) Adapt to diverse, changeable, and multi-element fraudulent websites. The pre-trained LXMERT network and CNN neural network are combined to construct a fraudulent website recognition model to identify web page information of different types of fraudulent websites, to cope with the diversification of fraudulent websites, and the frequent changes of URLs or page changes by means of changing headers.
[0042] (2) The accuracy of identifying fraudulent websites is high. Dedicated modules are used to identify different types of fraudulent websites, reducing the interference of unnecessary information and improving the accuracy of identifying fraudulent websites. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 This is a schematic diagram of the steps of a method for identifying fraudulent websites provided by a preferred embodiment of the present invention;
[0044] Figure 2 It is a schematic diagram of the steps of a method for extracting features of image data in a non-single-element web page provided by a preferred embodiment of the present invention;
[0045] Figure 3 This is an example of image data of a fraudulent website provided by a preferred embodiment of the present invention;
[0046] Figure 4 It is a schematic diagram of the method steps for comparing black and white lists provided by a preferred embodiment of the present invention;
[0047] Figure 5 This is a schematic diagram of the structure of a fraudulent website identification system provided by a preferred embodiment of the present invention;
[0048] Figure 6 It is a schematic diagram of a computer device provided by a preferred embodiment of the present invention. DETAILED DESCRIPTION
[0049] The following is a detailed explanation of the implementation mode of the present invention in conjunction with the accompanying drawings. The embodiments are provided only for illustrative purposes and cannot be understood as limiting the present invention. The accompanying drawings are only for reference and illustration purposes and do not constitute a limitation on the scope of patent protection of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In the description of the present invention, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, the meaning of "multiple" is two or more.
[0050] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used herein are only for illustrative purposes, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0051] In the description of the present invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood by specific circumstances.
[0052] See also Figure 1 In an embodiment of the present invention, a method for identifying fraudulent websites is provided, the method comprising:
[0053] S1. Collect a fraudulent website data set and a benign website data set, wherein the fraudulent website data set and the benign website data set both include: URL data, web page information and attribute information.
[0054] S2. Construct a fraudulent website identification model, the fraudulent website identification model includes a web page category identification module, a first fraudulent identification module and a second fraudulent identification module; the web page category identification module is used to identify the web page category of the web page information, the web page category includes a single-element web page and a non-single-element web page; the first fraudulent identification module is used to identify the non-single-element web page; the second fraudulent identification module is used to identify the single-element web page; the elements include: text, images, hyperlinks and navigation bars;.
[0055] S3. Training the fraudulent website identification model based on the fraudulent website dataset and the benign website dataset.
[0056] S4. During the fraud-related website detection process, the web page information of the website to be detected is input into the trained fraud-related website recognition model to obtain corresponding attribute information.
[0057] S5. Determine whether the website to be detected is a fraud-related website based on the attribute information.
[0058] In a preferred embodiment of the present invention, a website data set is obtained from the public data of websites of social network security institutions such as the operator security interception database, the National Network Security and the University of New Brunswick, and the website data set includes a fraudulent website data set and a benign website data set. In the fraudulent website data set and the benign website data set, each piece of data includes: URL data, web page information and attribute information. In a preferred embodiment of the present invention, the attribute information includes virtual winning, search engine, refund fraud, pornography fraud, illegal gambling, benign URLs and other URLs. Specifically, based on the web page information in the fraudulent website data set, the URL data is marked as virtual winning, search engine, refund fraud, pornography fraud, illegal gambling and other URLs, and the URL data in the benign website data set are all marked as benign URLs.
[0059] In a preferred embodiment of the present invention, the fraudulent website identification model includes a web page category identification module, a first fraudulent identification module and a second fraudulent identification module. The first fraudulent identification module is constructed based on a pre-trained LXMERT network, and the second fraudulent identification module is constructed based on a CNN neural network. The web page category identification module is used to identify web page information, mark web page information containing at least one of text, images, hyperlinks and navigation bars as non-single-element web pages, and mark web page information containing only images and no text information on the images as single-element web pages. For non-single-element web pages, the pre-trained LXMERT network is used for training and identification, and for single-element web pages, the CNN neural network is used for training and identification. After the web page category identification module identifies the web page information, the non-single-element web page is input into the first fraudulent identification module, and the single-element web page is input into the second fraudulent identification module to achieve classification processing of web page information, so as to adapt to diverse and changeable fraudulent websites and improve the accuracy of fraudulent website identification.
[0060] The pre-trained LXMERT network takes images, texts, hyperlinks and navigation bars as input elements respectively. The image input, text input, hyperlink input and navigation bar input in the URL data stream are processed separately through Transformer and cross-modal fusion is performed to obtain the single-modal output results and cross-modal output results.
[0061] The pre-trained LXMERT network includes an embedding layer, an encoding layer, an output layer, and a fully connected layer. The encoding layer includes a text encoder, an image encoder, a hyperlink encoder, a navigation bar encoder, and a cross-modal fusion encoder.
[0062] In a preferred embodiment of the present invention, the first fraud identification module also includes a text feature extraction module, which is connected to the text encoder through an embedding layer and is used to extract text information from web page information. The text feature extraction module is also used to extract text descriptions of images, combine the text information and text descriptions into text data, and input the text data into the text encoder.
[0063] The text description is obtained through the machine vision algorithm, specifically: the input image is converted into a pixel matrix, and the pixel value is normalized to between 0 and 1. The SIFT algorithm is used to extract the feature points in the pixel matrix. The feature points are relatively prominent points in the pixel matrix, such as points that constitute the contour, bright spots in the darker area, and dark spots in the brighter area. For the extracted feature points, the BRIEF algorithm is used to obtain the associated points near the feature points, and the pixel matrix composed of the feature points and the associated points is used as the descriptor of the feature point. The number of feature points and the matrix dimension of the descriptor are mapped to the object through the pre-stored semantic information table, so as to obtain the semantic information such as the name, relationship, and attributes of the object, such as "dice", "registration box", "color", etc. The mapped semantic information is converted into text, such as "red six-sided dice". Further, the text contained in the image is recognized, such as "Dear QQ user: Congratulations, your QQ number has been selected by the system as the second prize. You will receive a bonus of 12,000 yuan and an Apple laptop! Please remember your verification code: 5168" on the image. The above information is the text description of the current image.
[0064] The first fraud-related identification module also includes a hyperlink feature extraction module, which is connected to the hyperlink encoder through an embedding layer and is used to extract the hyperlink URL of the web page information and input the hyperlink URL into the text encoder.
[0065] The first fraud-related identification module also includes a navigation bar feature extraction module, which is connected to the navigation bar encoder through an embedding layer, and is used to extract navigation bar information from web page information and input the navigation bar information into the navigation bar encoder.
[0066] The embedding layer of the LXMERT network is used to convert the input into a sequence. The pre-trained LXMERT network includes four embedding layers, namely the text embedding layer, image embedding layer, hyperlink embedding layer, and navigation bar embedding layer. The text embedding layer converts the text data input into text embedding features, the image embedding layer converts the image input into image embedding features, the hyperlink embedding layer converts the hyperlink URL input into hyperlink embedding features, and the navigation bar embedding layer converts the navigation bar information input into navigation bar embedding features.
[0067] The text embedding features, image embedding features, hyperlink embedding features and navigation bar embedding features are further processed by the text encoder, image encoder, hyperlink encoder and navigation bar encoder of the encoding layer respectively.
[0068] Each layer of the text encoder, image encoder, hyperlink encoder and navigation bar encoder of the encoding layer contains a multi-head attention sublayer and a first feedforward neural network sublayer. The multi-head attention sublayer and the first feedforward neural network sublayer are connected to a residual connection and a normalization layer in sequence.
[0069] The text encoder of the pre-trained LXMERT network is used to extract features from text data to obtain a text feature vector. The image encoder is used to extract features from images in non-single-element web pages to obtain an image feature vector. The hyperlink encoder is used to extract features from hyperlink URLs to obtain a hyperlink feature vector. The navigation bar encoder is used to extract features from navigation bar information to obtain a navigation bar feature vector. The cross-modal fusion encoder is used to fuse text feature vectors, image feature vectors, hyperlink feature vectors, and navigation bar feature vectors to obtain a cross-modal output result.
[0070] In a preferred embodiment of the present invention, feature extraction is performed on images in non-single-element web pages, such as Figure 2 As shown, the following steps are included:
[0071] S01, performing feature recognition on the image based on a marked feature library to obtain fraud-related marks and corresponding mark categories contained in the image data;
[0072] S02, using a target detection algorithm to partition each of the fraud-related marks, and obtain position information corresponding to each of the fraud-related marks, as well as positional relationships between fraud-related marks of different mark categories;
[0073] S03: taking the fraud-related mark, and the mark category, position information and position relationship corresponding to the fraud-related mark as an image feature vector of the image.
[0074] In a preferred embodiment of the present invention, feature recognition is performed on an image based on a marked feature library to obtain the fraudulent marks and corresponding mark categories contained in the image, specifically: obtain the feature vector of each mark in the image, including the geometric feature vector and the topological feature vector; use the distance measurement method to calculate the distance between the feature vector of the image and the feature vector of each known category feature in the marked feature library; obtain the feature vectors of the first K features with the smallest distance based on the calculated distance, and obtain the mark categories corresponding to the K features; use the decision rule to count and analyze these K categories, if the number of one category exceeds half of the K, then the mark is judged to be the corresponding category; if the number of no category exceeds half of the K, then the category of the feature with the closest distance is selected as the category of the mark. If the category finally determined belongs to the mark category of the fraudulent mark, then the mark is a fraudulent mark, and the corresponding feature vector is used to represent the fraudulent mark.
[0075] In a preferred embodiment of the present invention, the tag types include: obfuscation tags, functional tags, and business tags. The obfuscation tag is a mark that misleads the user that the current URL is a benign URL, such as Figure 3 The example of the picture of the fraudulent website shown in the figure is that the penguin icon + Tencent + QQ.COM in the upper left corner is a confusing mark. Functional marks and business marks are marks that lead users into fraud traps, such as virtual winning and refund fraud winning / refund marks, and illegal gambling betting marks. Figure 3 The "Login to receive the prize" shown is the business tag. Since fraudulent websites often change URLs frequently, but often use a set of image features, the target detection algorithm is used to partition each fraudulent tag, obtain the location information corresponding to each fraudulent tag, and the location relationship between fraudulent tags of different tag categories. Using the fraudulent tags, the tag categories, location information and location relationships corresponding to the fraudulent tags as the image feature vectors of the image can effectively improve the success rate of fraudulent website identification.
[0076] Each layer of the cross-modal fusion encoder of the encoding layer includes a bidirectional cross-attention sublayer, two self-attention sublayers and two second feed-forward neural network sublayers. The bidirectional cross-attention sublayer, the self-attention sublayer and the second feed-forward neural network sublayer are connected to a residual connection and a normalization layer in sequence.
[0077] The bidirectional cross-attention sublayer is used to exchange and align information on the text feature vector output by the text encoder, the image feature vector output by the image encoder, the hyperlink feature vector output by the hyperlink encoder, and the navigation bar feature vector output by the navigation bar encoder to achieve joint cross-modal representation and obtain cross-modal output results.
[0078] The output layer includes three output channels: language, vision, and cross-modality. The language output and visual output are feature sequences generated by the Transformer cross-modal decoder based on the text feature vector, image feature vector, hyperlink feature vector, and navigation bar feature vector. The cross-modal output is a fused feature vector obtained by the Transformer cross-modal decoder by fusing the text feature vector, image feature vector, hyperlink feature vector, and navigation bar feature vector. The Transformer cross-modal decoder is used to introduce the text feature vector, hyperlink feature vector, and navigation bar feature vector into the mutual attention module of the Transformer cross-modal decoder. The image feature vector is used as the input of the Transformer cross-modal decoder to identify the correlation between each recognition area in the image feature vector. Through the interactive operation between the text feature features, hyperlink feature vector, navigation bar feature vector, and image feature vector, cross-modal information fusion is completed.
[0079] The fully connected layer takes language output, visual output and cross-modal output as input and outputs the corresponding attribute information. The language output, visual output and cross-modal output are input to the fully connected layer for nonlinear transformation and then mapped to the prediction result through the softmax layer.
[0080] In a preferred embodiment of the present invention, the second fraud identification module is used to identify a single-element web page. The second fraud identification module includes a CNN neural network. The CNN neural network mainly includes a convolution layer and a pooling layer. The number of convolution layers and pooling layers can be set as needed during the model building process. The CNN neural network uses the picture in the single-element web page as the input of the model, extracts the image feature vector from the picture, and maps the image feature vector to the prediction result.
[0081] The fraudulent website identification model is trained based on the fraudulent website dataset and the benign website dataset. The web page information corresponding to the URL data is used as the input of the fraudulent website identification model, and the attribute information corresponding to the URL data is used as the output to obtain a trained fraudulent website identification model.
[0082] In the process of detecting fraudulent websites, the web page information of the website to be detected is extracted and collected, and the web page category identification module identifies the web page information. If it is a non-single-element web page, it is input into the first fraud-related identification module. If it is a single-element web page, it is input into the second fraud-related identification module to obtain the attribute information of the website to be detected, such as virtual winning, search engine, refund fraud, pornographic fraud, illegal gambling, benign URLs and other URLs. According to the attribute information, it is judged whether the URL data to be detected is a fraudulent website. Among them, virtual winning, search engine, refund fraud, pornographic fraud, illegal gambling and other URLs are all fraudulent websites, and only benign URLs are non-fraudulent websites. The present invention combines the pre-trained LXMERT network and the CNN neural network to construct a fraudulent website identification model to identify the web page information of different fraudulent websites respectively, to cope with the diversification of fraudulent websites, and the frequent change of URLs or the change of pages by means of changing headers, etc., to improve the accuracy of fraudulent website identification.
[0083] In a preferred embodiment of the present invention, an identification log is generated based on the judgment result, and the identification log is sent to the filing entity corresponding to the website, the third-party fraud monitoring terminal and the cloud database respectively. The release of all websites needs to be reviewed by the filing entity, so sending the identification log to the filing entity corresponding to the website can fundamentally prevent the further spread of the fraudulent website. The third-party fraud monitoring terminal is generally a national information security regulatory department. Sending the identification log to the third-party fraud monitoring terminal will help improve the supervision work of the national information security regulatory department. Sending the identification log to the cloud can realize the sharing of identification data and further improve network security.
[0084] In a preferred embodiment of the present invention, before the webpage information of the website to be detected is input into the trained fraud-related website identification model, Figure 4 As shown, it also includes:
[0085] S001. Compare the URL data to be detected with a pre-built blacklist database, and deny access to the URL data to be detected that matches the blacklist database.
[0086] S002: Compare the URL data to be detected with a pre-built whitelist database, and allow access to the URL data to be detected that matches the whitelist database.
[0087] S003. Input the web page information corresponding to the URL data to be detected that does not match the blacklist database and the whitelist database into the trained fraud-related website identification model.
[0088] A blacklist database is constructed with the URL data of the fraudulent website data set in the website data set, and a whitelist database is constructed with the URL data of the benign website data set in the website data set. Before model detection is performed on the URL data to be detected, it is compared with the blacklist database and the whitelist database in turn, and access is denied to the URL data to be detected that matches the blacklist database, and access is allowed to the URL data to be detected that matches the whitelist database. The URL data to be detected that does not match both the blacklist database and the whitelist database is filtered out. The web page information corresponding to the URL data to be detected that does not match both the blacklist database and the whitelist database is input into the trained fraud-related website identification model for model identification. Black and white list filtering can greatly improve the efficiency of fraud-related website identification and meet the needs of large-scale network data identification.
[0089] In a preferred embodiment of the present invention, a fraudulent website data set and a benign website data set are collected, and both the fraudulent website data set and the benign website data set include: URL data, web page information and attribute information; a fraudulent website identification model is constructed, and the fraudulent website identification model includes a web page category identification module, a first fraudulent identification module and a second fraudulent identification module; the web page category identification module is used to identify the web page category of the web page information, and the web page category includes a single-element web page and a non-single-element web page; the first fraudulent identification module is used to identify non-single-element web pages; the second fraudulent identification module is used to identify single-element web pages; the fraudulent website identification model is trained based on the fraudulent website data set and the benign website data set; in the process of fraudulent website detection, the web page information of the website to be detected is input into the trained fraudulent website identification model to obtain the attribute information of the website to be detected; and whether the website to be detected is a fraudulent website is determined according to the attribute information. The fraudulent website identification method provided by the present invention combines the pre-trained LXMERT network and the CNN neural network to construct a fraudulent website identification model, so as to respectively identify the web page information of different fraudulent websites, to cope with the diversification of fraudulent websites, and the situation of frequently changing the URL or changing the page by means of changing the header, etc., to improve the accuracy of fraudulent website identification.
[0090] Accordingly, if Figure 5 As shown, based on a fraudulent website identification method, an embodiment of the present invention further provides a fraudulent website identification system, including: a data set acquisition unit 1, a model construction unit 2, a model training unit 3, a model detection unit 4 and a fraudulent website judgment unit 5;
[0091] The data set collection unit 1 is used to collect a fraudulent website data set and a benign website data set, and both the fraudulent website data set and the benign website data set include: URL data, web page information and attribute information.
[0092] The model building unit 2 is used to build a fraud-related website identification model, and the fraud-related website identification model includes a web page category identification module, a first fraud-related identification module and a second fraud-related identification module; the web page category identification module is used to identify the web page category of the web page information, and the web page category includes a single-element web page and a non-single-element web page; the first fraud-related identification module is used to identify the non-single-element web page; the second fraud-related identification module is used to identify the single-element web page, and the elements include: text, images, hyperlinks and navigation bars.
[0093] The model training unit 3 is used to train the fraudulent website identification model based on the fraudulent website data set and the benign website data set.
[0094] The model detection unit 4 is used to input the web page information of the website to be detected into the trained fraud-related website identification model during the fraud-related website detection process to obtain corresponding attribute information.
[0095] The fraudulent website determination unit 5 is used to determine whether the website to be detected is a fraudulent website according to the attribute information.
[0096] For the specific definition of a fraudulent website identification system, please refer to the above-mentioned definition of a fraudulent website identification method, which will not be repeated here. A person of ordinary skill in the art can realize that the various modules and steps described in conjunction with the embodiments disclosed in the present invention can be implemented in hardware, software, or a combination of both. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0097] like Figure 6 As shown, a computer device provided in an embodiment of the present invention includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps in the above-mentioned fraudulent website identification method embodiment are implemented, for example Figure 1 Steps S1 to S5 described in .
[0098] Those skilled in the art will understand that the Figure 6 It is only an example of a computer device and does not constitute a limitation of the computer device. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, etc.
[0099] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the computer device, and uses various interfaces and lines to connect various parts of the entire computer device.
[0100] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the computer device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.
[0101] Wherein, if the module integrated in the computer device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.
[0102] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.
[0103] Accordingly, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to perform the steps in the fraudulent website identification method of the above embodiment, for example Figure 1 Steps S1 to S5 described in .
[0104] A fraudulent website identification method, system, computer device and storage medium provided in this embodiment are aimed at the technical problem that the existing fraudulent website identification model cannot adapt to the changeable and diverse fraudulent websites and has low identification accuracy. The present invention collects a fraudulent website data set and a benign website data set, both of which include: URL data, web page information and attribute information; constructs a fraudulent website identification model, which includes a web page category identification module, a first fraudulent identification module and a second fraudulent identification module; the web page category identification module is used to identify the web page category of web page information, and the web page category includes a single-element web page and a non-single-element web page; the first fraudulent identification module is used to identify non-single-element web pages; the second fraudulent identification module is used to identify single-element web pages; the fraudulent website identification model is trained based on the fraudulent website data set and the benign website data set; in the process of fraudulent website detection, the web page information of the website to be detected is input into the trained fraudulent website identification model to obtain the corresponding attribute information; according to the attribute information, it is judged whether the website to be detected is a fraudulent website. The fraudulent website identification method provided by the present invention combines the pre-trained LXMERT network and the CNN neural network to construct a fraudulent website identification model, so as to respectively identify the web page information of different fraudulent websites, cope with the diversification of fraudulent websites, and the frequent changes of URLs or changes of pages by means of changing headers, etc., and improve the accuracy of fraudulent website identification.
[0105] The above-mentioned embodiments only express several preferred implementation modes of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principle of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the protection scope of the claims.
Claims
1. A method for identifying fraudulent websites, characterized in that: The method comprises: Collecting a fraudulent website data set and a benign website data set, wherein the fraudulent website data set and the benign website data set both include: URL data, web page information, and attribute information; Constructing a fraudulent website identification model, the fraudulent website identification model comprising a web page category identification module, a first fraudulent identification module and a second fraudulent identification module; the web page category identification module is used to identify the web page category of the web page information, the web page category includes a single-element web page and a non-single-element web page; the first fraudulent identification module is used to identify the non-single-element web page; the second fraudulent identification module is used to identify the single-element web page; the elements include: text, images, hyperlinks and navigation bars; Training the fraudulent website identification model based on the fraudulent website dataset and the benign website dataset; In the process of detecting fraudulent websites, the webpage information of the website to be detected is input into the trained fraudulent website identification model to obtain corresponding attribute information; Determine whether the website to be detected is a fraudulent website based on the attribute information.
2. The method for identifying fraudulent websites according to claim 1, characterized in that: The first fraud identification module includes a pre-trained LXMERT network; The pre-trained LXMERT network includes an embedding layer, an encoding layer, an output layer and a fully connected layer; The encoding layer includes a text encoder, an image encoder, a hyperlink encoder, a navigation bar encoder and a cross-modal fusion encoder; Each layer of the text encoder, the image encoder, the hyperlink encoder and the navigation bar encoder includes a multi-head attention sublayer and a first feedforward neural network sublayer; Each layer of the cross-modal fusion encoder includes a bidirectional cross-attention sub-layer, two self-attention sub-layers and two second feed-forward neural network sub-layers.
3. The method for identifying fraudulent websites according to claim 2, characterized in that: The image encoder is used to extract features from the image in the non-single-element web page; The step of extracting features from the image in the non-pure image webpage includes: Perform feature recognition on the image based on the marked feature library to obtain fraud-related marks and corresponding mark categories contained in the image; Using a target detection algorithm to partition each of the fraud-related marks, obtain the position information corresponding to each of the fraud-related marks, and the position relationship between the fraud-related marks of different mark categories; The fraud-related mark, and the mark category, position information and position relationship corresponding to the fraud-related mark are used as image feature vectors of the image.
4. The method for identifying fraudulent websites according to claim 3, characterized in that: The tag categories include: obfuscation tags, functional tags and business tags.
5. The method for identifying fraudulent websites according to claim 2, characterized in that: The first fraud identification module further includes a text feature extraction module, which is connected to the text encoder through the embedding layer and is used to extract text information from the web page information; The text feature extraction module is also used to extract the text description of the image, and input the text information and the text description into the text encoder.
6. The method for identifying fraudulent websites according to claim 1, characterized in that: The method further comprises: Based on the judgment result, an identification log is generated; The identification log is sent to the registration entity corresponding to the website, the third-party fraud monitoring terminal and the cloud database respectively.
7. The method for identifying fraudulent websites according to claim 1, characterized in that: Before inputting the web page information corresponding to the URL data to be detected into the trained fraud-related website identification model, the method further includes: Compare the URL data to be detected with a pre-built blacklist database, and deny access to the URL data to be detected that matches the blacklist database; Compare the URL data to be detected with a pre-built whitelist database, and allow access to the URL data to be detected that matches the whitelist database; The web page information corresponding to the URL data to be detected that does not match the blacklist database and the whitelist database is input into the trained fraud-related website identification model.
8. A fraudulent website identification system, characterized in that: The system includes: a data set acquisition unit, a model construction unit, a model training unit, a model detection unit and a fraudulent website judgment unit; The data set collection unit is used to collect a fraudulent website data set and a benign website data set, wherein the fraudulent website data set and the benign website data set both include: URL data, web page information and attribute information; The model building unit is used to build a fraudulent website identification model, and the fraudulent website identification model includes a web page category identification module, a first fraudulent identification module, and a second fraudulent identification module; the web page category identification module is used to identify the web page category of the web page information, and the web page category includes a single-element web page and a non-single-element web page; the first fraudulent identification module is used to identify the non-single-element web page; the second fraudulent identification module is used to identify the single-element web page; the elements include: text, images, hyperlinks, and navigation bars; The model training unit is used to train the fraudulent website identification model based on the fraudulent website data set and the benign website data set; The model detection unit is used to input the webpage information of the website to be detected into the trained fraudulent website identification model during the fraudulent website detection process to obtain corresponding attribute information; The fraudulent website determination unit is used to determine whether the website to be detected is a fraudulent website according to the attribute information.
9. A computer device, characterized in that: The computer device includes a memory, a processor and a transceiver, which are connected via a bus; the memory is used to store a set of computer program instructions and data, and transmit the stored data to the processor, and the processor executes the program instructions stored in the memory to execute the fraudulent website identification method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the method for identifying fraudulent websites as described in any one of claims 1 to 7 is implemented.
Citation Information
Cited By
Counterfeit and fraud website detection method and computer equipment
CN119583218A
