GAN-based website traffic authenticity measurement method, system and storage medium

Abstract

A GAN-based website traffic authenticity measurement method, system, and storage medium, the method includes collecting website click event data; using the website click event data to construct a behavior sequence; training a generative adversarial network model, using the generator in the generative adversarial network model to continuously generate simulated behavior sequences, and the discriminator to continuously learn how to judge whether the sequence is real or fake, until the loss of the generator and the discriminator no longer decreases, terminating the training to obtain a trained generative adversarial network model; using the discriminator of the trained generative adversarial network model to perform a discriminant measurement on the authenticity of the constructed behavior sequence. The system includes a data collection module, a behavior sequence construction module, a network model training module, and a authenticity discrimination module. The present invention can achieve more accurate recognition, use machine learning methods to learn user behavior patterns, can discover some patterns that cannot be discovered manually, and can adapt to different websites.

Description

Technical Field

[0001] The present invention belongs to the technical field of advertising display effect monitoring, and specifically relates to a GAN-based website traffic authenticity measurement method, system, and storage medium. Background Art

[0002] Traffic fraud has become an open secret in the advertising industry, with a significant amount of non-human traffic involved. A recent report from Virginia-based Distil Networks, a bot monitoring and API security expert, shows that over one-third of web traffic in 2018 was driven by bots. Of this, over 20% was driven by "bad bots" (as opposed to benign bots like search engine crawlers), a 6.4% increase from 2017. Bots are automated software scripts / applications used to perform repetitive tasks online. Common emulators in the domestic market include NetEase MUMU, Thunder, Xiaoyao, Nox, and BlueStacks. These emulators can manipulate location and device parameters. Fraudsters use these emulators to simulate users visiting websites and clicking on ads, defrauding advertisers of their marketing fees. Webmasters often mix real and fake traffic to increase their revenue and avoid detection by detection agencies. Therefore, it is necessary to verify the authenticity of website users, specifically whether there is fraudulent activity and the proportion of fraudulent activity.

[0003] Traditional detection solutions are basically based on manual rules, such as user access frequency, whether the IP address is from the computer room, IP address dispersion, etc. These detection methods also have the following drawbacks:

[0004] 1) Once the rules are established, they are rarely modified and cannot adapt to changes in circumstances.

[0005] 2) Rules rely on people’s experience and subjective judgment and cannot be objective.

[0006] 3) Different websites have different situations, and a one-size-fits-all rule is difficult to meet the identification requirements. Summary of the Invention

[0007] The purpose of the present invention is to address the above-mentioned problems in the prior art and provide a GAN-based website traffic authenticity measurement method, system and storage medium, which avoids the rigidity and subjectivity of manual rules and can achieve more accurate identification.

[0008] In order to achieve the above object, the present invention has the following technical solutions:

[0009] A GAN-based website traffic authenticity measurement method, comprising:

[0010] Collect website click event data;

[0011] Use website click event data to construct behavior sequences;

[0012] Train the generative adversarial network model. The generator in the generative adversarial network model continuously generates simulated behavior sequences, and the discriminator continuously learns how to distinguish whether the sequence is real or fake. Training is terminated when the loss of the generator and discriminator no longer decreases, and a trained generative adversarial network model is obtained.

[0013] The discriminator of the trained generative adversarial network model is used to perform discriminative measurement of the authenticity of the constructed behavior sequence.

[0014] As a preferred solution of the GAN-based website traffic authenticity measurement method of the present invention, the step of collecting website click event data uses JavaScript to collect device-related parameters and click behavior logs from the website front-end page.

[0015] As a preferred solution of the GAN-based website traffic authenticity measurement method of the present invention, the step of collecting website click event data specifically includes: deploying a JavaScript script on a media website, and when the current page is loaded, synchronously loading and executing the deployed JavaScript script, obtaining the document object model DOM and event information of the current page, and thereby capturing all information of the current page and the user's operation record on the current page.

[0016] As a preferred solution of the GAN-based website traffic authenticity measurement method of the present invention, in the step of collecting website click event data, the format of the collected original data is:<session_id,height,width,event,x,y,timestamp> ; In the formula, session_id represents an ID that marks a user's session on the website, height represents the screen height of the current device, width represents the screen width of the current device, event represents a behavioral event, x represents the x-coordinate of the screen when the current behavioral event occurs, y represents the y-coordinate of the screen when the current behavioral event occurs, and timestamp represents the timestamp of the current behavioral event. The behavioral event includes the following expressions: touchstart is triggered when the hand touches the screen, touchmove is triggered when the finger slides on the screen, and touchend is triggered when the finger moves away from the screen.

[0017] As a preferred solution of the GAN-based website traffic authenticity measurement method of the present invention, in the step of constructing a behavior sequence using website click event data, behaviors are defined as sliding and clicking. Sliding is expressed as a combination of touchstart, touchmove*N, and touchend, and clicking is expressed as a combination of touchstart and touchend.

[0018] As a preferred solution of the GAN-based website traffic authenticity measurement method of the present invention, in the step of constructing a behavior sequence using website click event data, the constructed behavior sequence format is:<wait_time,x_start,y_start,x_end,y_end,is_click> ;Wherein, wait_time represents the interval between two behaviors, which is equal to the time interval from the touchend time of the previous behavior to the touchstart time of this behavior. The first behavior defines wait_time as 0; x_start represents the x-coordinate of the start of a behavior after normalization, which is equal to the x-coordinate of touchstart / screen width width; y_start represents the y-coordinate of the start of a behavior after normalization, which is equal to the y-coordinate of touchstart / screen height height; x_end represents the x-coordinate of the end of a behavior after normalization, which is equal to the x-coordinate of touchend / screen width width; y_end represents the y-coordinate of the end of a behavior after normalization, which is equal to the y-coordinate of touchend / screen height height; is_click represents whether it is a click behavior based on whether touchmove is contained between touchstart and touchend. If no touchmove event is included, it is defined as a click, recorded as 1; if a touchmove event is included, it is defined as a sliding behavior, recorded as 0.

[0019] As a preferred solution of the GAN-based website traffic authenticity measurement method of the present invention, in the step of constructing a behavior sequence using website click event data, user behaviors with the same session_id are aggregated into a sequence, and all events collected by a session_id are sorted according to timestamp, indicating the sequence of all behavioral events of a user in the current session; when constructing the behavior sequence, a pair of touchstart and touchend is searched from the beginning, and whether it is a sliding or clicking behavior is determined based on whether there is a touchmove event in the middle, so that the event in the original data becomes the is_click mark in the behavior sequence; the x and y coordinates of the touchstart event of a behavior are scaled according to the width and height, expressed as the x / width of touchstart as the value of x_start in the sample, and the y / height of touch_start as the value of y_start in the sample; the x, y coordinates and width and height of the touchend event of a behavior are also generated by the same method to generate x_end and y_end in the sample; for click behavior,<x_start,y_start> and<x_end,y_end> The time interval between the two behaviors is equal to the time interval from the timestamp corresponding to the touchend event of the previous behavior to the timestamp corresponding to the touchstart event of this behavior, which is used as the value of wait_time; set the first wait_time of a sequence to 0.

[0020] As a preferred solution of the GAN-based website traffic authenticity measurement method of the present invention, in the step of using the discriminator of the trained generative adversarial network model to perform discriminant measurement on the authenticity of the constructed behavior sequence, website click event data is collected on other media websites, and the behavior sequence is constructed using the website click event data, which is input into the discriminator of the trained generative adversarial network model. The user's authenticity is scored according to the discrimination result, and the user's authenticity scores are averaged as the site traffic authenticity score; for media websites whose site traffic authenticity scores are lower than the set threshold, traffic purchase is not carried out, and the bidding for advertising display is guided according to the site traffic authenticity score.

[0021] The present invention also provides a GAN-based website traffic authenticity measurement system, comprising:

[0022] Data collection module, used to collect website click event data;

[0023] Behavior sequence construction module, used to construct behavior sequences using website click event data;

[0024] The network model training module is used to train the generative adversarial network model. The generator in the generative adversarial network model continuously generates simulated behavior sequences, and the discriminator continuously learns how to distinguish whether the sequence is real or fake. The training is terminated when the loss of the generator and discriminator no longer decreases, and a trained generative adversarial network model is obtained.

[0025] The authenticity discrimination module is used to use the discriminator of the trained generative adversarial network model to discriminate and measure the authenticity of the constructed behavior sequence.

[0026] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the GAN-based website traffic authenticity measurement method.

[0027] Compared with the prior art, the present invention has at least the following beneficial effects:

[0028] GAN-based learning of patterns in user scrolling and clicking behavior sequences is essentially learning common user habits. This makes it significantly more difficult for fraudsters to simulate user behavior compared to simulating IP addresses and visit frequency metrics. Furthermore, user behavior patterns on website visits are extremely complex. Manual judgment can only rely on statistically relevant metrics, such as the average number of clicks and scrolls per session, the number of clicks between scrolls, the approximate start and end coordinates of scrolls, and the approximate click coordinates. However, it's impossible to determine whether a behavior sequence is real or simulated based on a single metric. A combination of multiple factors is required, making manual judgment virtually impossible. However, machine learning, leveraging the large number of parameters in neural networks, can learn patterns and uncover patterns that are difficult for humans to discover. Furthermore, network-based models can evaluate each user's behavior to determine the authenticity of the site's users, avoiding the rigidity and subjectivity of manual rules and enabling more accurate identification. The behavioral patterns are also invariant and adaptable to different websites. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0030] Figure 1 Flowchart of the GAN-based website traffic authenticity measurement method according to an embodiment of the present invention;

[0031] Figure 2 Schematic diagram of the user's real behavior sequence obtained by an embodiment of the present invention;

[0032] Figure 3 An example diagram showing that each row of the raw data in the embodiment of the present invention represents an event;

[0033] Figure 4 Each array of the behavior sequence constructed in the embodiment of the present invention represents an example graph of user behavior for a session;

[0034] Figure 5(a) Heat map of real user sliding;

[0035] Figure 5(b) Heat map of simulator user sliding;

[0036] Figure 6 (a) Heat map of real user clicks;

[0037] Figure 6(b) Heat map of simulator user clicks. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0039] Based on the embodiments of the present invention, ordinary technicians in this field can make some simple modifications and improvements without making any creative work. All other embodiments obtained are within the scope of protection of the present invention.

[0040] References to "embodiments" in this disclosure mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the disclosure. The appearance of such phrases in various locations in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0041] See also Figure 1 In an embodiment of the present invention, a GAN-based website traffic authenticity measurement method is proposed. Taking users visiting mobile websites as an example, the sliding and clicking behaviors on the website are recorded to generate training samples. The model is trained using a generative adversarial network to learn the generator and discriminator, and then the discriminator is used to score the authenticity of users on other websites.

[0042] The specific steps include:

[0043] Step 1: Data collection;

[0044] During the data collection process, select a few sites with stable conversion effects and good reputation. You can use JavaScript technology to collect device-related parameters and user click behavior logs from the front-end page of the website.

[0045] The format of the collected raw data is:<session_id,height,width,event,x,y,timestamp> ;

[0046] In the formula, session_id represents an ID that marks a user's session on the website;

[0047] Height indicates the screen height of the current device;

[0048] Width indicates the screen width of the current device;

[0049] Events represent behavioral events, including touchstart (triggered by a touch of the screen), touchmove (triggered by a finger sliding across the screen), and touchend (triggered by a finger moving away from the screen). Because different browsers trigger touchmove events at varying frequencies, it's impossible to determine sliding speed based on this. A touchstart event followed by a touchend event is considered a click, while a touchstart event followed by one or more touchmove events and finally a touchend is considered a slide.

[0050] x is the x-coordinate of the screen, that is, the x-coordinate where the current behavior event occurs;

[0051] y is the y coordinate of the screen, that is, the y coordinate of the current behavior event;

[0052] timestamp is the timestamp, that is, the timestamp of the current behavior event.

[0053] Step 2: Construct the sample;

[0054] Use the collected raw data to construct a complete user behavior trajectory. Aggregate the user behaviors of the same session, sort them by the timestamp of the behavior, and find<touchstart,touchend> Pairs of events are combined into behaviors.

[0055] The embodiment of the present invention defines two types of behaviors: sliding and clicking;

[0056] Sliding is expressed as a combination of touchstart, touchmove*N, and touchend.

[0057] Click is expressed as a combination of touchstart and touchend.

[0058] The format of the constructed behavior sequence is:<wait_time,x_start,y_start,x_end,y_end,is_click> ;

[0059] Where wait_time represents the interval between two actions, that is, the time interval from the touchend time of the previous action to the touchstart time of this action. The first wait_time of the session is 0.

[0060] x_start represents the normalized x-coordinate of the start of a behavior, which is equal to the x-coordinate of touchstart / screen width;

[0061] y_start represents the normalized y coordinate of the start of an action, which is equal to the y coordinate of touchstart / screen height;

[0062] x_end represents the normalized x-coordinate of the end of a behavior, which is equal to the x-coordinate of touchend / screen width;

[0063] y_end indicates the y coordinate of the end of a behavior after normalization, which is equal to the y coordinate of touchend / screen height;

[0064] is_click indicates whether it is a click behavior based on whether touchmove is included between touchstart and touchend. If no touchmove event is included, it is defined as a click and recorded as 1. If a touchmove event is included, it is defined as a slide behavior and recorded as 0.

[0065] In an optional implementation, the format of the collected raw data is:<session_id,height,width,event,x,y,timestamp> , the generated sample format is:<wait_time,x_start,y_start,x_end,y_end,is_click> ,like Figure 3 as well as Figure 4As shown, the user behaviors of the same session_id are aggregated into a sequence, and all events collected by a session_id are sorted by timestamp. This is the order of all behavioral events of a user in the current session. The process of sample construction is to convert the collected events into user behaviors (sliding behaviors / clicking behaviors). According to the definition of sliding behaviors and clicking behaviors, the pairs of touchstart and touchend are searched from the beginning. Whether it is a sliding or clicking behavior is determined based on whether there is a touchmove event in the middle. Thus, the event in the original data becomes the is_click mark in the sample. The x and y coordinates of the touchstart event of a behavior are scaled according to the width and height of the screen, that is, the x / width of touchstart is used as the value of x_start in the sample, and the y / height of touch_start is used as the value of y_start in the sample; similarly, the x, y coordinates and width, height of the touchend event of a behavior are also used in the same way to generate x_end and y_end in the sample. For click behaviors,<x_start,y_start> and<x_end,y_end> The time interval between two behaviors, that is, the time interval between the timestamp corresponding to the touchend event of the previous behavior and the timestamp corresponding to the touchstart event of this behavior, is used as the value of wait_time. The first wait_time of a sequence is set to 0.

[0066] Step 3: Model training;

[0067] This step aims to learn the differences between normal user behavior patterns and those of simulator users. However, since the collected data is only normal user behavior patterns, a generative adversarial network (GAN) model is used: one network learns to generate samples, and another network distinguishes between normal and simulator behavior. Because the GAN model in this invention needs to generate and distinguish sequences, the DoppelGANger model was chosen to simultaneously learn sequence generation and distinguish between real behavior sequences.

[0068] Step 4: Authenticity assessment;

[0069] Using the trained discriminator of the generative adversarial network model, extract the trained discriminator function. Using the same approach as steps 1 and 2, collect and construct samples from other media websites. These samples are fed into the discriminator to determine user authenticity and assign a score. The user authenticity scores of the sampled websites are averaged to form the site traffic authenticity score. Traffic purchases are not made for websites with low traffic authenticity, and ad placement bids are guided by traffic authenticity.

[0070] The GAN of the present invention trains two models simultaneously. A generator continuously generates simulated user behavior sequences, and a discriminator continuously learns how to distinguish whether the sequence is real or constructed. During the training process, the two models progress together. During the training process, it can be observed that the losses of the two models decrease alternately, that is, the model effect is constantly getting better, that is, the generation ability of the generator is constantly improving, and the discrimination ability of the discriminator is also constantly improving. Finally, when neither of them rises, the model training can be stopped. It shows that the generation ability and the discrimination ability have reached the best effect generated by the current network. In the actual operation process, a heat map of the behavior sequence can be drawn to compare the heat map differences of the sliding start coordinates and the sliding end coordinates of the real behavior sequence and the simulated behavior sequence, as well as the heat map differences of the click coordinates.

[0071] Figure 2 The figure shows the actual behavior sequence of the user obtained by the embodiment of the present invention. In the figure, the first column is wait_time, the second column is x_start, the third column is y_start, the fourth column is x_end, the fifth column is y_end, and the last column is is_click.

[0072] Comparing Figures 5(a) and 5(b) and Figures 6(a) and 6(b), we can use heatmaps to evaluate the difference between the user behavior generated by the GAN generator and the real user behavior. If the difference is large, it means that the user behavior simulation is not good. If the heatmaps are close, it means that the generator training has met the requirements and the corresponding discriminator has also been trained well.

[0073] Compared to rules such as IP / frequency, the GAN-based website traffic authenticity measurement method of the present invention is more difficult to simulate sliding and clicking behavior and has higher recognition. Because this solution uses GAN to learn the patterns of user sliding and clicking behavior sequences, it is learning common user operating habits. For cheaters, the difficulty will be greatly increased compared to simulating IP / visit frequency indicators. The present invention uses machine learning methods to learn user behavior patterns, which can discover some patterns that cannot be discovered manually. User behavior patterns in website visits are extremely complex. Manual experience can only use some statistically related indicators to judge, such as: the average number of clicks and slides in a session, the number of slides with a click, the approximate start / end coordinates of the slide, the approximate coordinates of the click, etc. However, many of these cannot be determined based on a single indicator to determine whether the behavior sequence is real or simulated. It requires a combination of multiple situations to judge, which is basically impossible to do manually. However, through machine learning, the neural network has a large number of parameters to learn these patterns, and can unearth patterns that cannot be discovered manually. Finally, the behavioral patterns are invariant, and the present invention can adapt to the situations of different websites. The present invention directly models the user's operating habits and learns the user's behavioral habits on the mobile phone, that is, where the user likes to click, how to slide from where to where, how many times to slide to click, etc. Therefore, it can adapt to the recognition needs of various websites.

[0074] Another embodiment of the present invention further provides a GAN-based website traffic authenticity measurement system, comprising:

[0075] Data collection module, used to collect website click event data;

[0076] Behavior sequence construction module, used to construct behavior sequences using website click event data;

[0077] The network model training module is used to train the generative adversarial network model. The generator in the generative adversarial network model continuously generates simulated behavior sequences, and the discriminator continuously learns how to distinguish whether the sequence is real or fake. The training is terminated when the loss of the generator and discriminator no longer decreases, and a trained generative adversarial network model is obtained.

[0078] The authenticity discrimination module is used to use the discriminator of the trained generative adversarial network model to discriminate and measure the authenticity of the constructed behavior sequence.

[0079] Another embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the GAN-based website traffic authenticity measurement method are implemented.

[0080] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in a computer-readable storage medium and executed by the processor to implement the GAN-based website traffic authenticity measurement method of the present invention. The one or more modules / units may be a series of computer-readable instruction segments capable of performing specific functions, which are used to describe the execution process of the computer program on the server.

[0081] The server can be a computing device such as a smartphone, laptop, PDA, or cloud server. The server can include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the server can include more or fewer components, or a combination of certain components, or different components. For example, the server can also include input and output devices, network access devices, buses, and the like.

[0082] 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. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0083] The memory may be an internal storage unit of the server, such as a hard disk or memory of the server. The memory may also be an external storage device of the server, such as a plug-in hard disk equipped on the server, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. Furthermore, the memory may include both an internal storage unit of the server and an external storage device. The memory is used to store the computer-readable instructions and other programs and data required by the server. The memory may also be used to temporarily store data that has been output or is about to be output.

[0084] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0085] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0086] If the integrated unit 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 this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method 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 method embodiments when executed by a processor. Among them, the computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the camera / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunications signal and a software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk.

[0087] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0088] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A GAN-based website traffic authenticity measurement method, characterized by: include: Collect website click event data; Use website click event data to construct behavior sequences; Train the generative adversarial network model. The generator in the generative adversarial network model continuously generates simulated behavior sequences, and the discriminator continuously learns how to distinguish whether the sequence is real or fake. Training is terminated when the loss of the generator and discriminator no longer decreases, and a trained generative adversarial network model is obtained. Use the discriminator of the trained generative adversarial network model to measure the authenticity of the constructed behavior sequence; The step of collecting website click event data uses JavaScript to collect device-related parameters and click behavior logs from the website front-end page; In the step of collecting website click event data, the format of the collected original data is:<session_id,height,width,event,x,y,timestamp> ; In the formula, session_id represents an ID that marks a user's session on the website, height represents the screen height of the current device, width represents the screen width of the current device, event represents a behavioral event, x represents the x-coordinate of the screen when the current behavioral event occurs, y represents the y-coordinate of the screen when the current behavioral event occurs, and timestamp represents the timestamp of the current behavioral event. The behavioral event includes the following expressions: touchstart is triggered when the hand touches the screen, touchmove is triggered when the finger slides on the screen, and touchend is triggered when the finger moves away from the screen.

2. The GAN-based website traffic authenticity measurement method according to claim 1 is characterized in that: The steps of collecting website click event data specifically include: deploying a JavaScript script on a media website, synchronously loading and executing the deployed JavaScript script when the current page is loaded, obtaining the document object model DOM of the current page and event Event information, and thereby capturing all information of the current page and the user's operation records on the current page.

3. The GAN-based website traffic authenticity measurement method according to claim 1 is characterized in that: In the step of constructing a behavior sequence using website click event data, behaviors are defined as sliding and clicking. Sliding is expressed as a combination of touchstart, touchmove*N, and touchend, and clicking is expressed as a combination of touchstart and touchend.

4. The GAN-based website traffic authenticity measurement method according to claim 1 is characterized in that: In the step of constructing a behavior sequence using website click event data, the format of the constructed behavior sequence is:<wait_time,x_start,y_start,x_end,y_end,is_click> ;Wherein, wait_time represents the interval between two behaviors, which is equal to the time interval from the touchend time of the previous behavior to the touchstart time of this behavior. The first behavior defines wait_time as 0; x_start represents the x-coordinate of the start of a behavior after normalization, which is equal to the x-coordinate of touchstart / screen width width; y_start represents the y-coordinate of the start of a behavior after normalization, which is equal to the y-coordinate of touchstart / screen height height; x_end represents the x-coordinate of the end of a behavior after normalization, which is equal to the x-coordinate of touchend / screen width width; y_end represents the y-coordinate of the end of a behavior after normalization, which is equal to the y-coordinate of touchend / screen height height; is_click represents whether it is a click behavior based on whether touchmove is contained between touchstart and touchend. If no touchmove event is included, it is defined as a click, recorded as 1; if a touchmove event is included, it is defined as a sliding behavior, recorded as 0.

5. The GAN-based website traffic authenticity measurement method according to claim 4 is characterized in that: In the step of constructing a behavior sequence using website click event data, user behaviors with the same session_id are aggregated into a sequence, and all events collected by a session_id are sorted by timestamp to represent the sequence of all behavior events of a user in the current session; when constructing a behavior sequence, a pair of touchstart and touchend is searched from the beginning, and whether it is a sliding or clicking behavior is determined based on whether there is a touchmove event in the middle, so that the event in the original data becomes the is_click mark in the behavior sequence; the x and y coordinates of the touchstart event of a behavior are scaled according to the width and height, and expressed as the x / width of touchstart as the value of x_start in the sample, and the y / height of touch_start as the value of y_start in the sample; the x, y coordinates and width and height of the touchend event of a behavior are also generated by the same method to generate x_end and y_end in the sample; for click behavior,<x_start,y_start> and<x_end,y_end> The time interval between the two behaviors is equal to the time interval from the timestamp corresponding to the touchend event of the previous behavior to the timestamp corresponding to the touchstart event of this behavior, which is used as the value of wait_time; set the first wait_time of a sequence to 0.

6. The GAN-based website traffic authenticity measurement method according to claim 1 is characterized in that: In the step of using the discriminator of the trained generative adversarial network model to perform a discriminant measurement on the authenticity of the constructed behavior sequence, website click event data is collected from other media websites, the website click event data is used to construct a behavior sequence, and the sequence is input into the discriminator of the trained generative adversarial network model. The authenticity of the user is scored according to the discrimination result, and the average of the user's authenticity scores is calculated as the score of the site traffic authenticity; for media websites whose site traffic authenticity scores are lower than the set threshold, traffic purchase is not carried out, and the bid for advertising display is guided according to the site traffic authenticity score.

7. A website traffic authenticity measurement system based on GAN, characterized by: include: Data collection module, used to collect website click event data; In the process of collecting website click event data, JavaScript is used to collect device-related parameters and click behavior logs from the website front-end page; When collecting website click event data, the format of the collected raw data is:<session_id,height,width,event,x,y,timestamp> ; In the formula, session_id represents an ID that marks a user's session on the website, height represents the screen height of the current device, width represents the screen width of the current device, event represents the behavior event, x represents the x-coordinate of the screen when the current behavior event occurs, y represents the y-coordinate of the screen when the current behavior event occurs, and timestamp represents the timestamp of the current behavior event. A behavior sequence construction module is used to construct a behavior sequence using website click event data; the behavior event includes the following expressions: touchstart is triggered when a hand touches the screen, touchmove is triggered when a finger slides on the screen, and touchend is triggered when a finger moves away from the screen; The network model training module is used to train the generative adversarial network model. The generator in the generative adversarial network model continuously generates simulated behavior sequences, and the discriminator continuously learns how to distinguish whether the sequence is real or fake. The training is terminated when the loss of the generator and discriminator no longer decreases, and a trained generative adversarial network model is obtained. The authenticity discrimination module is used to use the discriminator of the trained generative adversarial network model to discriminate and measure the authenticity of the constructed behavior sequence.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the GAN-based website traffic authenticity measurement method according to any one of claims 1 to 6 are implemented.

Images