CN112149404A

CN112149404A - Method, device and system for identifying risk content of user privacy data

Info

Publication number: CN112149404A
Application number: CN202010987464.1A
Authority: CN
Inventors: 陈永环; 侯辉超; 张正雄
Original assignee: Alipay Hangzhou Information Technology Co Ltd
Current assignee: Alipay Hangzhou Information Technology Co Ltd
Priority date: 2020-09-18
Filing date: 2020-09-18
Publication date: 2020-12-29

Abstract

One or more embodiments of the present specification provide a method, an apparatus, and a system for identifying risk content of user privacy data, where the method is applied to a client, and the method includes: user privacy data generated by a user using a target application in a client is monitored. Performing semantic recognition on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is disposed on a client after being processed by a preset knowledge distillation method. And if the semantic representation information comprises a first semantic feature vector used for representing suspicious risk content sentences, sending the first semantic feature vector to the server, so that the server identifies the risk content of the user privacy data based on the first semantic feature vector.

Description

Method, device and system for identifying risk content of user privacy data

Technical Field

The present disclosure relates to the field of internet technologies, and in particular, to a method, an apparatus, and a system for identifying risk content of user privacy data.

Background

At present, with the coming of the internet era, the internet is widely applied to daily study, work and life of people. Various daily transactions can be processed and presented through the internet. Meanwhile, with the rapid development of the mobile internet, each internet service provider provides corresponding business services for users by developing respective application programs, and the users can install corresponding application programs, such as information applications, video applications, chat applications, shopping applications, payment applications and the like, in the smart phones according to respective actual requirements.

However, the application program may add risky content to the page browsing information presented by the user, especially adding a small amount of risk information in a large amount of normal browsing information, resulting in the risk information being hidden; and considering that the page browsing information of the user may belong to user privacy data, the client directly uploads the user privacy data (such as browsing content of the user on a certain webpage) to the server under the condition that the user is not aware of the user, so that the server identifies whether risk information exists in the user privacy data, thereby causing the problem of invading the user privacy, and therefore, the user privacy protection and the content risk content identification cannot be considered at the same time.

Therefore, it is necessary to provide a technical solution for identifying the risk content of the user privacy data while ensuring the user privacy protection.

Disclosure of Invention

An object of one or more embodiments of the present specification is to provide a method of risk content identification of user privacy data. The method for identifying the risk content of the user privacy data is applied to the client and comprises the following steps:

monitoring user privacy data generated by a user using a target application in the client. Performing semantic recognition on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at the client. And if the semantic representation information comprises a first semantic feature vector used for representing suspicious risk content sentences, sending the first semantic feature vector to a server, so that the server identifies the risk content of the user privacy data based on the first semantic feature vector.

An object of one or more embodiments of the present specification is to provide a risky content identification apparatus of user privacy data. Set up in the customer end, this risk content recognition device of user's privacy data includes:

a user privacy data monitoring module that monitors user privacy data generated by a user using a target application in the client. The privacy data semantic identification module is used for carrying out semantic identification on the user privacy data by utilizing a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at the client. And the semantic feature vector uploading module is used for sending the first semantic feature vector to a server if the semantic representation information comprises the first semantic feature vector used for representing the suspicious risk content statement, so that the server identifies the risk content of the user privacy data based on the first semantic feature vector.

An object of one or more embodiments of the present specification is to provide a risky content identifying apparatus of user privacy data, including: a processor; and a memory arranged to store computer executable instructions.

The computer-executable instructions, when executed, cause the processor to monitor user privacy data generated by a user using a target application in the client. Performing semantic recognition on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at the client. And if the semantic representation information comprises a first semantic feature vector used for representing suspicious risk content sentences, sending the first semantic feature vector to a server, so that the server identifies the risk content of the user privacy data based on the first semantic feature vector.

It is an object of one or more embodiments of the present specification to provide a storage medium for storing computer-executable instructions. The executable instructions, when executed by a processor, monitor user privacy data generated by a user using a target application in the client. Performing semantic recognition on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at the client. And if the semantic representation information comprises a first semantic feature vector used for representing suspicious risk content sentences, sending the first semantic feature vector to a server, so that the server identifies the risk content of the user privacy data based on the first semantic feature vector.

Drawings

In order to more clearly illustrate one or more embodiments or prior art solutions of the present specification, the drawings that are needed in the description of the embodiments or prior art will be briefly described below, it is obvious that the drawings in the following description are only some of the embodiments described in one or more of the specification, and that other drawings can be obtained by those skilled in the art without inventive exercise.

Fig. 1 is a schematic application scenario of a system for risk content identification of user privacy data according to one or more embodiments of the present disclosure;

fig. 2 is a first flowchart of a method for risk content identification of user privacy data according to one or more embodiments of the present disclosure;

fig. 3 is a second flowchart of a method for risk content identification of user privacy data according to one or more embodiments of the present disclosure;

fig. 4 is a schematic flow chart of a risk content identification method for user privacy data according to one or more embodiments of the present disclosure;

fig. 5 is a fourth flowchart of a method for risk content identification of user privacy data according to one or more embodiments of the present disclosure;

fig. 6 is a schematic flow chart of a method for risk content identification of user privacy data according to one or more embodiments of the present disclosure;

fig. 7 is a sixth flowchart of a method for risk content identification of user privacy data according to one or more embodiments of the present disclosure;

fig. 8 is a seventh flowchart of a method for risk content identification of user privacy data according to one or more embodiments of the present disclosure;

fig. 9 is a schematic block diagram illustrating a risk content identification apparatus for user privacy data according to one or more embodiments of the present disclosure;

fig. 10 is a schematic system structure diagram of a risk content identification device for user privacy data according to one or more embodiments of the present specification;

fig. 11 is a schematic structural diagram of a risk content identification device for user privacy data according to one or more embodiments of the present specification.

Detailed Description

In order to make the technical solutions in one or more embodiments of the present disclosure better understood, the technical solutions in one or more embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in one or more embodiments of the present disclosure, and it is obvious that the described embodiments are only a part of one or more embodiments of the present disclosure, but not all embodiments. All other embodiments that can be derived by a person skilled in the art from the embodiments given in one or more of the present specification without inventive step shall fall within the scope of protection of this document.

It should be noted that one or more embodiments and features of the embodiments in the present description may be combined with each other without conflict. Reference will now be made in detail to one or more embodiments of the disclosure, examples of which are illustrated in the accompanying drawings.

One or more embodiments of the present disclosure provide a method, an apparatus, and a system for identifying risk content of user privacy data, which monitor user privacy data generated by a user using a target application in real time at a client side, perform semantic feature vector conversion and preliminary suspicious risk content sentence identification on the user privacy data by using a preset semantic representation model, and send the generated semantic feature vector for representing the suspicious risk content sentence to a server side, that is, the client side uploads the preliminarily selected suspicious privacy data to the server side in the form of the semantic feature vector, so that the server side performs final risk content identification on the user privacy data based on the semantic feature vector, thereby avoiding the problem of user privacy disclosure caused by directly uploading the user privacy data to the server side, and achieving the purpose of protecting user privacy, and identifying the risk content of the user privacy data, and then managing and controlling the risk content information or the carrier of the risk content information in time.

Fig. 1 is a schematic view of an application scenario of a system for identifying risky content of user privacy data according to one or more embodiments of the present specification, as shown in fig. 1, the system includes: the risk identification system comprises a client, a business processing server and a risk identification server, wherein the client can be a client such as a smart phone and a tablet personal computer, the client can also be terminal equipment such as a personal computer, the business processing server can be a server for providing corresponding business services for the client using target applications, and the risk identification server can be an independent server or a server cluster consisting of a plurality of servers.

The specific process of identifying the risk content of the user privacy data is as follows:

the client sends a service request to a service processing server based on the trigger operation of a user for the target application;

the service processing server, in response to a service request from a client, sends page rendering data of an application operation page corresponding to the service request to the client, where the page rendering data includes: operating page display information;

the client performs page rendering based on the received page rendering data, and determines operation page display information in the page rendering data as user privacy data generated by a user using a target application in the client;

the client performs semantic recognition on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at a client;

if the semantic representation information comprises a first semantic feature vector used for representing the suspicious risk content sentence, the client sends the first semantic feature vector to the risk identification server;

and the risk identification server receives a first semantic feature vector sent by the client, and carries out risk content identification on the user privacy data based on the first semantic feature vector.

In the application scenario, a client sends a service processing request to a service processing server based on the triggering operation of a user for a target application, and the service processing server returns corresponding page display information to the client, the client monitors user privacy data generated by the user using the target application in real time, performs semantic feature vector conversion and preliminary suspicious risk content statement identification on the user privacy data by using a preset semantic representation model, and sends the generated semantic feature vector for representing the suspicious risk content statement to the server, namely, the client uploads the preliminarily screened suspicious privacy data to the server in the form of the semantic feature vector, so that the server performs final risk content identification on the user privacy data based on the semantic feature vector, and thus, the problem of user privacy leakage caused by directly uploading the user privacy data to the server can be avoided, therefore, under the condition of protecting the privacy of the user, the risk content identification is carried out on the privacy data of the user, and then the risk content information or the carrier of the risk content information is managed and controlled in time.

Fig. 2 is a first flowchart of a method for risk content identification of user privacy data according to one or more embodiments of the present specification, where the method in fig. 2 can be executed by the client in fig. 1, as shown in fig. 2, and the method includes at least the following steps:

s202, monitoring user privacy data generated by a user using a target application in a client; the target application can be an independently installed application program, or an applet or Html5 webpage accessed to a specified host application; the user privacy data may be operation page display information in the page rendering data returned by the service processing server.

S204, performing semantic recognition on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at a client; the preset semantic representation model can be a BERT model, and can also be other semantic recognition models, such as a Word2vec model. Wherein the semantic representation information may include: semantic feature vectors of a plurality of text sentences in the operation page display information and risk weights for representing the probability that each text sentence is a risk sentence;

specifically, the preset semantic representation model is obtained by a server side through a preset knowledge distillation method and based on normal corpus sample training, knowledge distillation is carried out on the preset semantic identification model through the knowledge distillation method, the size of the preset semantic identification model is compressed to a preset value, data processing and model storage pressure of a client side are reduced, and feasibility of primary risk screening by the preset semantic representation model at the client side is guaranteed; specifically, a preset semantic representation model is obtained by training in advance at the server side by using a preset knowledge distillation method and based on a normal corpus sample, and the preset semantic representation model is deployed at the client.

S206, if the semantic representation information comprises a first semantic feature vector used for representing a suspicious risk content statement, sending the first semantic feature vector to a server, so that the server identifies the risk content of the user privacy data based on the first semantic feature vector; wherein, the server can be the risk identification server;

if the semantic representation information does not include the first semantic feature vector for representing the suspicious risk content statement, the step S202 is continuously executed to monitor the user privacy data generated by the user using the target application in the client.

In specific implementation, for the case that the user privacy data includes the suspicious risk content statements, the client side may only upload the first semantic feature vector for representing the suspicious risk content statements to the server side, so that the upload amount of the user privacy data may be reduced, and the first semantic feature vector for representing the suspicious risk content statements and the second semantic feature vector for representing the risk-free content statements may also be uploaded at the same time, which manner is specifically adopted may be set according to actual conditions.

Specifically, semantic recognition is carried out on user privacy data by utilizing a preset semantic representation model at a client side, suspicious content in the user privacy data is preliminarily screened based on a semantic recognition result, the screened suspicious content is uploaded to a server side in a feature vector mode, and final risk content recognition is carried out on the user privacy data by the server side, so that the data volume of the user privacy data uploaded to the server side can be reduced, the semantic privacy representation vector of the user privacy data is uploaded to the server side instead of plaintext user privacy data, and risk content recognition and user privacy protection are simultaneously achieved.

In one or more embodiments of the present description, user privacy data generated by a user using a target application is monitored in real time at a client side, and the user privacy data is represented by a preset semantic representation model, performing semantic feature vector conversion and suspicious risk content sentence preliminary identification on the user privacy data, and sending the generated semantic feature vector for characterizing the suspicious risk content sentence to the server, namely, the client uploads the preliminarily screened suspicious privacy data to the server in the form of semantic feature vectors, so that the server performs final risk content identification on the user privacy data based on the semantic feature vector, therefore, the problem of user privacy disclosure caused by directly uploading the user privacy data to the server side can be avoided, and under the condition of protecting the user privacy, and identifying the risk content of the user privacy data, and then managing and controlling the risk content information or the carrier of the risk content information in time.

As shown in fig. 3, in the step S202, monitoring user privacy data generated by the user using the target application in the client specifically includes:

s2022, after the triggering operation of the user for the target application in the client is monitored, acquiring the operation page display information of the user under the target application; the triggering operation can be a page access triggering operation, and correspondingly, the operation page display information is access page information which is requested to return to the service processing server based on the page access triggering operation;

wherein, the operation page display information comprises: and at least one item of multimedia information, text information and picture information displayed on the terminal interface.

And S2024, determining user privacy data generated by the user using the target application based on the acquired operation page display information.

Specifically, considering that the input information of the preset semantic representation model is text type information, the non-text type information in the operation page display information is converted into text type display information; for example, for voice information in multimedia information, text conversion can be performed on the voice information to obtain text information corresponding to the semantic information; as another example, for picture class information, text class information may be extracted from the picture class information.

In view of the fact that a plurality of hosted applications may be accessed under an independent host application installed in a client, where the hosted applications are from a third-party service platform, and the third-party service platform may have a situation where a small amount of risk page information is doped in a large amount of normal page information displayed to a user by a target application, so as to achieve the purpose of delivering risk content information to the user, therefore, in order to improve the security of page information browsed by the user using the host application, risk monitoring needs to be performed on the page information displayed to the user by the hosted application by the third-party service platform, and based on this, the target application includes: accessing an applet or Html5 web page of a specified host application;

correspondingly, in the step S2022, after the triggering operation of the user on the target application in the client is monitored, acquiring the operation page display information of the user under the target application, specifically including:

after monitoring the triggering operation of a user for a specified host application, acquiring page text information browsed by the user under an applet or an Html5 webpage;

and determining the acquired page text information as the operation page display information of the user under the target application.

For example, the text information of the page browsed by the user under the applet or Html5 webpage is "according to data published at the heart row, as soon as 6 months and end in 2019, the short-term consumption loan scale of a department of residents of deposit financial institutions is 9.11 trillion yuan, good looking videos of the women and the you please contact xxx to log in www. XxNx. tv, the net increase in 2019 in the last half year is 3293.19 billion yuan, and the increase in the last half year is not optimistic. "among them, the risk content sentence contained in the page text message is" good looking vjuxue video, please contact xxx, log in www. XxNx. tv ", i.e. a small amount of risk page information intermingled with a large amount of normal page information.

Wherein, the user privacy data comprises: the page text information browsed by the user under the target application;

correspondingly, as shown in fig. 4, in the step S202, monitoring user privacy data generated by the user using the target application in the client specifically includes:

s2026, after monitoring the trigger operation of the user for the target application in the client, acquiring the page text information browsed by the user under the target application;

correspondingly, in the step S204, performing semantic identification on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information, specifically including:

s2042, splitting page text information browsed by a user under a target application to obtain a plurality of text content sentences to be identified;

s2044, performing semantic recognition on each text content sentence to be recognized by using a preset semantic representation model to obtain a semantic feature vector and a risk weight of each text content sentence; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at a client;

the risk weight is used for representing the probability that the text content sentence is a suspicious risk content sentence, and is determined based on the correlation degree between each target participle and adjacent participles in the text content sentence obtained when semantic recognition is carried out on the text content sentence to generate a semantic feature vector;

the preset semantic representation model can be a BERT model, the BERT model is obtained by utilizing a knowledge distillation method and training based on a normal corpus sample, correspondingly, in the process of semantic recognition of text content sentences, Basic Tokenizer in the BERT model can be adopted to carry out segmentation, whether each segmentation is a normal content corpus or not is recognized, if not, the segmentation is determined to be a suspicious risk word, and the risk weight corresponding to the text content sentences is increased.

S2046, determining semantic representation information of the user privacy data according to the semantic feature vectors and the risk weights of the text content sentences obtained through splitting; wherein the semantic representation information includes: semantic feature vectors and risk weights for each text content statement.

In one or more embodiments of the present description, a page text message is split into a plurality of page text sentences, that is, a long sentence is split into a plurality of short sentences, and then a preset semantic representation model is used to perform semantic recognition on each page text sentence, where the preset semantic representation model may be a BERT model, and correspondingly, a Basic token in the BERT model may be used to perform word segmentation, so that adventure counterwords (if an out-of-set word is an OOV word) can be more easily recognized, thereby avoiding a situation that a small amount of risky contents are submerged by a large amount of non-risky contents, and meanwhile, performing semantic recognition on the page text message through the preset semantic representation model can also retain context semantic information, perform semantic privacy characterization of risk countermeasures as vectors, and protect user privacy disclosure.

In the above S2044, for the process of determining the semantic feature vector and the risk weight of each text content sentence, performing semantic recognition on each text content sentence to be recognized by using a preset semantic representation model to obtain the semantic feature vector and the risk weight of each text content sentence, specifically including:

inputting each text content sentence to be identified into a preset semantic representation model to obtain a semantic feature vector of each text content sentence; and the number of the first and second groups,

step two, determining the number of suspicious risk words contained in each text content sentence by using a preset semantic representation model according to a risk pre-evaluation value of each target participle obtained when semantic recognition is carried out on the text content sentence to generate a semantic feature vector for each text content sentence; wherein the suspicious risk word comprises: at least one of risk keywords and outliers; the risk pre-evaluation value of the target word segmentation comprises the following steps: the correlation degree of each target participle and adjacent participles and/or the matching degree of the target participle and normal content corpus;

specifically, the out-of-vocabulary word is an OOV word, out of vocabularies, i.e., words beyond the word list, for example, please contact xxx to log in www for the above page text statement "good looking for the video of usu. XxNx. tv' including OOV words "" in the page text statement. Specifically, the preset semantic representation model is obtained by training based on a normal corpus sample, and the participles in the text content sentence except for the normal content corpus can be determined as suspicious risk words, that is, the smaller the matching degree between the target participle and the normal content corpus is, the higher the probability that the target participle is a risk word is.

For another example, the page text statement "good looking you video, please contact xxx, log in www. XxNx. tv ", the page text statement also contains the risk keyword XxNx.

Determining the risk weight of the text content sentence according to the determined number of the suspicious risk words; wherein the risk weight is positively correlated with the number of suspicious risk words contained in the currently identified text content sentence; that is to say, the larger the number of suspicious risk words in the text content sentence is, the larger the risk weight of the text content sentence in the page text information is, that is, the higher the probability that the text content sentence is a risk content sentence is, so that the condition that the OOV words contained in the normal page text information are submerged can be avoided.

For example, when a model training sample of a preset semantic representation model is a normal corpus sample, semantic recognition is performed on each participle in each text content sentence to be recognized by using the preset semantic representation model, and the matching degree between each target participle in the text content sentence and the normal corpus sample is determined for each text content sentence; the smaller the matching degree is, the greater the probability that the target participle does not belong to the normal content corpus is, so that whether the corresponding participle is a risk word or not can be determined according to the matching degree of each participle in the text content sentence and the normal corpus sample, and specifically, if the matching degree is smaller than a preset threshold value, the participle is determined to be a risk word.

For another example, when semantic recognition is performed on each participle in each text content sentence to be recognized by using the preset semantic representation model, for each text content sentence, an attention weight (i.e., an attribute value) assigned to each target participle in the text content sentence is determined, where the attribute value can represent an importance degree of a currently recognized target participle to its adjacent participle, that is, the greater the attribute value is, the higher the importance degree of the target participle to its adjacent participle is, that is, the higher the correlation degree of the target participle to its adjacent participle is, therefore, the attribute value can be used as a basis for recognizing a risk word in the text content sentence, that is, according to the size of the attribute value of each participle in the text content sentence, it is determined whether the corresponding participle is a risk word, specifically, if the attribute value is smaller than a preset threshold, the participle is determined as a risk word.

In specific implementation, risk word labeling can be performed on semantic feature vectors corresponding to text content sentences based on position information of risk words in the text content sentences aiming at each text content sentence;

and determining the labeled semantic feature vector as the semantic feature vector corresponding to the text content sentence, so that the server side can quickly lock the position of the risk word in the semantic feature vector corresponding to the text content sentence.

Further, the semantic representation information further includes: a second semantic feature vector for characterizing risk-free content statements;

and the risk weight corresponding to the second semantic feature vector is smaller than the risk weight corresponding to the first semantic feature vector.

Specifically, the second semantic feature vector is a semantic feature vector corresponding to a text content sentence with the participles being normal content corpus; correspondingly, the first semantic feature vector is a semantic feature vector corresponding to a text content statement including an abnormal content corpus, that is, the text content statement includes the abnormal content corpus and a normal content corpus.

In the splitting process of the page text information, in step S2042, the page text information browsed by the user in the target application is split to obtain a plurality of text content sentences to be recognized, and the splitting process specifically includes:

judging whether the character length of the page text information browsed by the user under the target application is larger than a preset maximum character length or not;

if so, splitting the page text information browsed by the user under the target application according to the preset maximum character length to obtain a plurality of text content sentences to be identified.

In specific implementation, aiming at the splitting process of the page text information, the page text information browsed by a user under a target application can be split according to preset punctuation marks, so that a plurality of text content sentences to be identified are obtained.

Further, with respect to the process of determining the first semantic feature vector characterizing the suspicious risk content sentence based on the risk weight of each text content sentence, on the basis of fig. 4, as shown in fig. 5, in the above S2046, after determining the semantic characterization information of the user privacy data according to the semantic feature vector and the risk weight of each text content sentence obtained by splitting, the method further includes:

s2048, determining whether the obtained page text information contains the suspicious text content sentences or not according to the risk weight of each text content sentence;

if yes, executing S2050, and determining that the semantic representation information comprises a first semantic feature vector used for representing the suspicious risk content statement;

specifically, a semantic feature vector of a suspicious text content sentence identified from user privacy data is determined as a first semantic feature vector for characterizing a suspicious risk content sentence.

Aiming at the identification process of suspicious text content sentences, if the risk weight of the text content sentences with the suspicious risk words of zero number is set to be zero; correspondingly, in the step S2048, according to the risk weight of each text content sentence, it is determined whether the obtained page text information includes a suspicious text content sentence, which specifically includes:

and if at least one risk weight is not zero, determining that the page text information contains the suspicious text content sentence.

Further, according to the risk weight of each text content sentence, determining a first semantic feature vector for representing a suspicious risk content sentence;

specifically, according to the sequence of the risk weights from high to low, semantic feature vectors corresponding to a preset number of risk weights ranked in the top are determined as first semantic feature vectors for representing suspicious risk content sentences;

or determining the semantic feature vector corresponding to the risk weight which is greater than the preset risk threshold value in the plurality of risk weights as a first semantic feature vector for representing the suspicious risk content sentence.

Wherein, the target application comprises: accessing an applet or Html5 web page of a specified host application; correspondingly, for the condition that the server corresponding to the specified host application is different from the server corresponding to the target application, the risk identification server may be a server providing business services for the client using the specified host application, that is, the server corresponding to the specified host application performs risk monitoring on the applet or the Html5 webpage accessing the specified host application;

correspondingly, as shown in fig. 6, in the step S206, if the semantic representation information includes a first semantic feature vector for representing a suspicious risk content statement, sending the first semantic feature vector to the server, so that the server performs risk content identification on the user privacy data based on the first semantic feature vector, specifically including:

s2062, if the semantic representation information includes a first semantic feature vector used for representing a suspicious risk content statement, sending the first semantic feature vector to a server corresponding to a specified host application accessed by the target application, so that the server identifies the risk content of the user privacy data based on the first semantic feature vector.

Specifically, semantic recognition is carried out on user privacy data generated by using a target application by using a preset semantic representation model at a client side, and a first semantic feature vector which is preliminarily screened and used for representing suspicious risk content statements is sent to a server side corresponding to a specified host application accessed by the target application, so that risk control is carried out on an applet or an Html5 webpage accessed to the host application by a certain host application, and the security of the applet or the Html5 webpage accessed to the host application is further ensured.

Further, for a case that the server finally determines that the operation page display information under the target application includes risk content information, the target application needs to be managed and controlled, and as shown in fig. 7, in the step S206, if the semantic representation information includes a first semantic feature vector used for representing a suspicious risk content statement, the first semantic feature vector is sent to the server, so that the server performs risk content identification on the user privacy data based on the first semantic feature vector, which specifically includes:

s2064, if the semantic representation information comprises a first semantic feature vector used for representing a suspicious risk content sentence, uploading the determined first semantic feature vector and identification information of the target application to a server, so that the server carries out risk content identification on the suspicious risk content sentence in the page text information based on the first semantic feature vector, and if the suspicious risk content sentence is determined to be a risk content sentence, carrying out preset processing on the target application; wherein the preset process includes: any one of interception, control and off-shelf.

Specifically, considering that the client cannot access the target application due to the fact that the target application is directly preset at the server, in order to improve user experience, after the target application is preset at the server, corresponding abnormal prompt information is returned to the client, so that the user can know the reason why the client cannot normally access the target application, based on this, in S2064, the determined first semantic feature vector and the identification information of the target application are uploaded to the server, so that the server performs risk content recognition on a suspicious risk content statement in page text information based on the first semantic feature vector, and if the suspicious risk content statement is determined to be a risk content statement, after the target application is preset, the method further includes:

receiving a preset processing result which is returned by the server and aims at the target application, wherein the preset processing result is generated when the suspicious risk content sentence is determined to be a risk content sentence by the server;

and displaying corresponding abnormal prompt information to the user according to a preset processing result aiming at the target application.

Further, considering that generally trained preset semantic representation models are relatively large, if the trained preset semantic representation models are deployed at a client, certain pressure is brought to calculation and storage of the client, so that response speed of the client for a user is slowed down, and user experience is affected, as shown in fig. 8, before monitoring user privacy data generated by the user using a target application in the client, in S202, the method further includes:

s208, receiving file packet information of a preset semantic representation model issued by a server; the preset semantic representation model is obtained by training based on preset corpus samples by using a knowledge distillation method; the preset corpus sample comprises: normal corpus samples; in specific implementation, the preset corpus sample may further include: risk keywords, extravehicular words; the knowledge distillation method can be any one of ALBERT, Q8BERT, DistillBERT and TinyBERT so as to obtain a semantic representation model after knowledge distillation compression;

when the server side trains the preset semantic representation model based on the preset corpus samples, a knowledge distillation method is introduced, the size of the preset semantic representation model is compressed to a preset value by the knowledge distillation method, for example, the size of the preset semantic representation model can be compressed to about 10M, so that the preset semantic representation model is deployed at the client side in a light weight mode, namely, the knowledge distillation method is utilized in the training process of the preset semantic representation model, model deployment lightweight is embodied, edge calculation is facilitated, semantic recognition is conveniently carried out on user privacy data by the preset semantic representation model at the client side, risk content primary screening is supported on the user privacy data at the client side, and the user privacy data do not need to be uploaded to the cloud side in a full quantity.

Specifically, the pre-server extracts a normal content corpus from a specified data provider, and uses the normal content corpus as a training sample of a preset semantic representation model, for example, the specified data provider may be an official website without illegal content, such as a people's daily report, a people network, and the like; as another example, the specified profile provider may also be a third-party website that does not have illegal content, such as encyclopedia, news, interactive communities, and the like.

And S210, storing the trained preset semantic representation model locally based on the received file packet information.

In order to improve the accuracy of recognizing the risk words of the page text information by using the preset semantic representation model, in the process of training the preset semantic representation model, normal corpus samples (namely corpus samples without risk content) are used as a model training corpus sample set of the preset semantic representation model, so that the risk counterwords are accurately recognized in the page text information, the purpose of countermeasures of the risk content is achieved, and the risk information is prevented from being submerged by a large amount of normal information.

In the method for identifying risk content of user privacy data in one or more embodiments of the present specification, user privacy data generated by a user using a target application is monitored in real time at a client side, a preset semantic feature model is utilized to perform semantic feature vector conversion and preliminary identification of suspicious risk content statements on the user privacy data, and the generated semantic feature vectors for representing the suspicious risk content statements are sent to a server side, that is, the client side uploads the preliminarily screened suspicious privacy data to the server side in the form of semantic feature vectors, so that the server side performs final risk content identification on the user privacy data based on the semantic feature vectors, thereby avoiding the problem of user privacy leakage caused by directly uploading the user privacy data to the server side, and realizing risk content identification on the user privacy data under the condition of protecting the user privacy, and then the risk content information or the carrier of the risk content information is managed and controlled in time.

On the basis of the same technical concept, corresponding to the method for identifying the risk content of the user privacy data described in fig. 2 to 8, one or more embodiments of the present specification further provide an apparatus for identifying the risk content of the user privacy data, and fig. 9 is a schematic diagram of modules of the apparatus for identifying the risk content of the user privacy data provided in one or more embodiments of the present specification, where the apparatus is disposed at a client and configured to perform the method for identifying the risk content of the user privacy data described in fig. 2 to 8, and as shown in fig. 9, the apparatus includes:

a user privacy data monitoring module 902 that monitors user privacy data generated by a user using a target application in the client;

a privacy data semantic identification module 904, which performs semantic identification on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at the client;

a semantic feature vector uploading module 906, configured to send a first semantic feature vector to a server if the semantic representation information includes the first semantic feature vector used for representing a suspicious risk content statement, so that the server performs risk content identification on the user privacy data based on the first semantic feature vector.

Optionally, the user privacy data monitoring module 902:

after monitoring the trigger operation of a user for a target application in the client, acquiring the operation page display information of the user under the target application;

and determining user privacy data generated by the user by using the target application based on the operation page display information.

Optionally, the target application includes: accessing an applet or Html5 web page of a specified host application; the user privacy data monitoring module 902, which:

after monitoring the triggering operation of a user for the specified host application, acquiring page text information browsed by the user under the applet or the Html5 webpage;

and determining the page text information as the operation page display information of the user under the target application.

Optionally, the user privacy data comprises: page text information browsed by the user under the target application; the private data semantic identification module 904 that:

splitting the page text information to obtain a plurality of text content sentences to be identified;

performing semantic recognition on each text content statement by using a preset semantic representation model to obtain a semantic feature vector and a risk weight of each text content statement;

and determining corresponding semantic representation information according to the semantic feature vectors and the risk weights of the text content sentences.

Optionally, the private data semantic identifying module 904:

inputting each text content sentence into the preset semantic representation model to obtain a semantic feature vector of each text content sentence; and the number of the first and second groups,

determining the number of suspicious risk words contained in each text content sentence by using the preset semantic representation model aiming at each text content sentence;

determining the risk weight of the text content sentence according to the number of the suspicious risk words; wherein the risk weight is positively correlated with the number of suspicious risk words.

Optionally, the semantic representation information further includes: a second semantic feature vector for characterizing the risk-free content statement,

the risk weight corresponding to the second semantic feature vector is less than the risk weight corresponding to the first semantic feature vector.

Optionally, the private data semantic identifying module 904:

judging whether the character length of the page text information is larger than the maximum character length;

if so, splitting the page text information according to the maximum character length to obtain a plurality of text content sentences to be identified.

Optionally, the apparatus further comprises: a suspicious risk content determination module that:

determining whether the page text information contains suspicious text content sentences according to the risk weight of each text content sentence;

and if so, determining that the semantic representation information comprises a first semantic feature vector for representing the suspicious risk content sentence.

Optionally, if the risk weight corresponding to the text content sentence with the number of the suspicious risk words being zero is zero;

the suspicious risk content determination module is configured to:

if at least one risk weight is not zero, determining that the page text information contains suspicious text content sentences; and the number of the first and second groups,

determining semantic feature vectors corresponding to a preset number of risk weights ranked in the front as first semantic feature vectors for representing suspicious risk content sentences according to the sequence of the risk weights from high to low; or determining the semantic feature vector corresponding to the risk weight which is greater than the preset risk threshold value in the plurality of risk weights as a first semantic feature vector for representing the suspicious risk content sentence.

Optionally, the target application includes: accessing an applet or Html5 web page of a specified host application; the semantic feature vector upload module 906 that:

and sending the first semantic feature vector to a server corresponding to the specified host application.

Optionally, the semantic feature vector uploading module 906, which:

uploading the first semantic feature vector and the identification information of the target application to a server, so that the server performs risk content identification on the suspicious risk content sentences based on the first semantic feature vector, and performing preset processing on the target application if the suspicious risk content sentences are determined to be risk content sentences.

Optionally, the apparatus further comprises: an application processing result prompting module that:

receiving a preset processing result which is returned by the server and aims at the target application, wherein the preset processing result is generated when the suspicious risk content statement is determined to be a risk content statement by the server;

and displaying corresponding abnormal prompt information to the user according to the preset processing result aiming at the target application.

Optionally, the apparatus further comprises: a semantic representation model deployment module that:

receiving file packet information of a preset semantic representation model issued by a server; the preset semantic representation model is obtained by training based on preset corpus samples by using a knowledge distillation method; the preset corpus sample comprises: normal corpus samples, risk keywords and extracollected words;

and storing the preset semantic representation model locally based on the file packet information.

Optionally, the preset semantic representation model includes: BERT model.

In the risk content identification device for user privacy data in one or more embodiments of the present specification, user privacy data generated by a user using a target application is monitored in real time at a client side, a preset semantic feature model is utilized to perform semantic feature vector conversion and preliminary identification of suspicious risk content statements on the user privacy data, and the generated semantic feature vectors for representing the suspicious risk content statements are sent to a server side, that is, the client side uploads the preliminarily screened suspicious privacy data to the server side in the form of semantic feature vectors, so that the server side performs final risk content identification on the user privacy data based on the semantic feature vectors, thereby avoiding the problem of user privacy leakage caused by directly uploading the user privacy data to the server side, and realizing risk content identification on the user privacy data under the condition of protecting the user privacy, and then the risk content information or the carrier of the risk content information is managed and controlled in time.

It should be noted that, the embodiment of the device for identifying risk content related to user privacy data in this specification and the embodiment of the method for identifying risk content related to user privacy data in this specification are based on the same inventive concept, and therefore, for specific implementation of this embodiment, reference may be made to implementation of the method for identifying risk content related to user privacy data described above, and repeated details are not described again.

On the basis of the same technical concept, corresponding to the method for recognizing the risk content of the user privacy data described in fig. 2 to 8, one or more embodiments of the present specification further provide a system for recognizing the risk content of the user privacy data, and fig. 10 is a schematic structural composition diagram of the system for recognizing the risk content of the user privacy data provided in one or more embodiments of the present specification, the system is configured to execute the method for recognizing the risk content of the user privacy data described in fig. 2 to 8, as shown in fig. 10, the system includes: the system comprises a client and a risk identification server;

the client monitors user privacy data generated by a user by using a target application in the client; performing semantic recognition on the generated user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at a client; if the obtained semantic representation information comprises a first semantic feature vector used for representing the suspicious risk content sentence, the first semantic feature vector is sent to the risk identification server;

the risk identification server receives a first semantic feature vector which is uploaded by a client and used for representing suspicious risk content sentences; and performing risk content identification on the user privacy data based on the first semantic feature vector.

Specifically, the risk identification server identifies risk content of a suspicious risk content statement based on a first semantic feature vector uploaded by a client, and if the suspicious risk content statement is determined to be a risk content statement, performs preset processing on a target application; wherein, this includes: any one of interception, control and off-shelf.

Further, after determining that the suspicious risk content sentence is a risk content sentence, the risk identification server generates a preset processing result for the target application, and returns the preset processing result to the client, so that the client displays corresponding abnormal prompt information to the user according to the preset processing result for the target application.

In the system for identifying risk content of user privacy data in one or more embodiments of the present specification, user privacy data generated by a user using a target application is monitored in real time at a client side, a preset semantic feature model is utilized to perform semantic feature vector conversion and preliminary identification of suspicious risk content statements on the user privacy data, and the generated semantic feature vectors for representing the suspicious risk content statements are sent to a server side, that is, the client side uploads the preliminarily screened suspicious privacy data to the server side in the form of semantic feature vectors, so that the server side performs final risk content identification on the user privacy data based on the semantic feature vectors, thereby avoiding the problem of user privacy leakage caused by directly uploading the user privacy data to the server side, and realizing risk content identification on the user privacy data under the condition of protecting the user privacy, and then the risk content information or the carrier of the risk content information is managed and controlled in time.

It should be noted that, the embodiment of the risk content identification system related to the user privacy data in this specification and the embodiment of the risk content identification method related to the user privacy data in this specification are based on the same inventive concept, and therefore, for specific implementation of this embodiment, reference may be made to implementation of the risk content identification method related to the user privacy data described above, and repeated details are not described again.

Further, corresponding to the methods shown in fig. 2 to fig. 8, based on the same technical concept, one or more embodiments of the present specification further provide a device for identifying the risky content of the user privacy data, which is configured to perform the method for identifying the risky content of the user privacy data, as shown in fig. 11.

The risk content identification device for user privacy data may have a relatively large difference due to different configurations or performances, and may include one or more processors 1101 and a memory 1102, where the memory 1102 may store one or more stored applications or data. Wherein memory 1102 may be transient or persistent. The application stored in memory 1102 may include one or more modules (not shown), each of which may include a series of computer-executable instructions in a risk content identification device for user privacy data. Still further, the processor 1101 may be configured to communicate with the memory 1102 to execute a series of computer-executable instructions in the memory 1102 on a risky content identification device for user privacy data. The risky content identification apparatus of user privacy data may also include one or more power supplies 1103, one or more wired or wireless network interfaces 1104, one or more input-output interfaces 1105, one or more keyboards 1106, and the like.

In a particular embodiment, a device for risk content identification of user privacy data includes a memory, and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer-executable instructions in the device for risk content identification of user privacy data, and the one or more programs configured to be executed by one or more processors include computer-executable instructions for:

monitoring user privacy data generated by a user using a target application in a client;

performing semantic recognition on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at the client;

and if the semantic representation information comprises a first semantic feature vector used for representing suspicious risk content sentences, sending the first semantic feature vector to a server, so that the server identifies the risk content of the user privacy data based on the first semantic feature vector.

Optionally, the computer executable instructions, when executed, monitor user privacy data generated by a user using a target application in the client, comprising:

Optionally, the computer executable instructions, when executed, the target application comprises: accessing an applet or Html5 web page of a specified host application;

after monitoring the trigger operation of the user for the target application in the client, acquiring the operation page display information of the user under the target application, including:

Optionally, the computer executable instructions, when executed, the user privacy data comprises: page text information browsed by the user under the target application;

the semantic recognition is carried out on the user privacy data by utilizing a preset semantic representation model to obtain corresponding semantic representation information, and the semantic representation information comprises the following steps:

Optionally, when executed, the performing semantic recognition on each text content statement by using a preset semantic representation model to obtain a semantic feature vector and a risk weight of each text content statement includes:

Optionally, the computer executable instructions, when executed, further comprise: a second semantic feature vector for characterizing the risk-free content statement,

Optionally, when executed, the computer-executable instructions perform splitting processing on the page text information to obtain a plurality of text content sentences to be recognized, including:

Optionally, when executed, the computer-executable instructions, after performing semantic recognition on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information, further include:

Optionally, when executed, the computer-executable instruction sets a risk weight corresponding to the text content sentence with the suspicious risk word number of zero to zero; determining whether the page text information contains a suspicious text content sentence according to the risk weight of each text content sentence, including:

the sending the first semantic feature vector to the server includes:

Optionally, when executed, the computer-executable instructions send the first semantic feature vector to a server, so that the server performs risk content identification on the user privacy data based on the first semantic feature vector, including:

Optionally, when executed, the computer executable instructions upload the first semantic feature vector and identification information of the target application to a server, so that the server performs risk content identification on the suspicious risk content sentence based on the first semantic feature vector, and after performing preset processing on the target application if it is determined that the suspicious risk content sentence is a risk content sentence, the method further includes:

Optionally, the computer-executable instructions, when executed, further comprise, prior to monitoring user privacy data generated by a user using a target application in the client:

Optionally, the computer executable instructions, when executed, further comprise: BERT model.

It should be noted that, the embodiment of the risk content identification device related to the user privacy data in this specification and the embodiment of the risk content identification method related to the user privacy data in this specification are based on the same inventive concept, and therefore, for specific implementation of this embodiment, reference may be made to implementation of the risk content identification method related to the user privacy data described above, and repeated details are not described again.

Further, based on the same technical concept, corresponding to the methods shown in fig. 2 to fig. 8, one or more embodiments of the present specification further provide a storage medium for storing computer-executable instructions, where in a specific embodiment, the storage medium may be a usb disk, an optical disk, a hard disk, and the like, and the storage medium stores computer-executable instructions that, when executed by a processor, implement the following processes:

When executed by a processor, the computer-executable instructions stored in the storage medium in one or more embodiments of the present specification monitor, at a client side, user privacy data generated by a user using a target application in real time, perform semantic feature vector conversion and preliminary suspicious risk content sentence identification on the user privacy data by using a preset semantic representation model, and send the generated semantic feature vector for representing the suspicious risk content sentence to a server side, that is, the client side uploads the preliminarily screened suspicious privacy data to the server side in the form of the semantic feature vector, so that the server side performs final risk content identification on the user privacy data based on the semantic feature vector, thereby avoiding the problem of user privacy disclosure caused by directly uploading the user privacy data to the server side, and achieving the purpose of protecting user privacy, and identifying the risk content of the user privacy data, and then managing and controlling the risk content information or the carrier of the risk content information in time.

It should be noted that the embodiment of the storage medium in this specification and the embodiment of the method for identifying risk content of user privacy data in this specification are based on the same inventive concept, and therefore specific implementation of this embodiment may refer to implementation of the method for identifying risk content of user privacy data described above, and repeated parts are not described again.

The foregoing description has been directed to specific embodiments of this disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

In the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD), such as a Field Programmable Gate Array (FPGA), is an integrated circuit whose Logic functions are determined by programming the Device by a user. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Furthermore, nowadays, instead of manually making an Integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development and writing, but the original code before compiling is also written by a specific Programming Language, which is called Hardware Description Language (HDL), and HDL is not only one but many, such as abel (advanced Boolean Expression Language), ahdl (alternate Language Description Language), traffic, pl (core unified Programming Language), Cal, jhdware Description Language, langua, mylar, pams, hardlanguage (Hardware Description Language), vhlanguage, Language, HDL, software Language (Hardware Description Language), and vhjjjjjjjjjjjjjjjg Language, which are currently used in most fields. It will also be apparent to those skilled in the art that hardware circuitry that implements the logical method flows can be readily obtained by merely slightly programming the method flows into an integrated circuit using the hardware description languages described above.

The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be considered a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.

The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functionality of the various elements may be implemented in the same one or more software and/or hardware implementations of one or more of the present descriptions.

As will be appreciated by one skilled in the art, one or more embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, one or more of the present description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, one or more of the present description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied in the medium.

One or more of the present specification has been described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to one or more embodiments of the specification. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

One or more of the present specification can be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. One or more of the present specification can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The above description is merely illustrative of one or more embodiments of the present disclosure and is not intended to limit one or more embodiments of the present disclosure. Various modifications and alterations to one or more of the present descriptions will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement or the like made within the spirit and principle of one or more of the present specification should be included in the scope of one or more claims of the present specification.

Claims

1. A risk content identification method of user privacy data is applied to a client and comprises the following steps:

monitoring user privacy data generated by a user using a target application in the client;

2. The method of claim 1, wherein the monitoring user privacy data generated by a user using a target application in the client comprises:

3. The method of claim 2, wherein the target application comprises: accessing an applet or Html5 web page of a specified host application;

4. The method of claim 1, wherein the user privacy data comprises: page text information browsed by the user under the target application;

5. The method of claim 4, wherein the performing semantic recognition on each text content sentence by using a preset semantic representation model to obtain a semantic feature vector and a risk weight of each text content sentence comprises:

6. The method of claim 4, wherein the semantic representation information further comprises: a second semantic feature vector for characterizing risk-free content statements;

7. The method of claim 4, wherein the splitting the page text information to obtain a plurality of text content sentences to be recognized comprises:

8. The method of claim 4, wherein after performing semantic recognition on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information, the method further comprises:

9. The method according to claim 8, wherein if the number of the suspicious risk words is zero, the risk weight corresponding to the text content sentence is zero;

determining whether the page text information contains a suspicious text content sentence according to the risk weight of each text content sentence, including:

10. The method of claim 1, wherein the target application comprises: accessing an applet or Html5 web page of a specified host application;

the sending the first semantic feature vector to the server includes:

11. The method of claim 1, wherein the sending the first semantic feature vector to a server to enable the server to perform risk content identification on the user privacy data based on the first semantic feature vector comprises:

12. The method according to claim 11, wherein after uploading the first semantic feature vector and the identification information of the target application to a server, so that the server performs risk content identification on the suspicious risk content sentence based on the first semantic feature vector, and performing preset processing on the target application if it is determined that the suspicious risk content sentence is a risk content sentence, the method further comprises:

13. The method of any of claims 1 to 12, wherein prior to monitoring user privacy data generated by a user using a target application in the client, further comprising:

receiving file packet information of a preset semantic representation model issued by a server; the preset semantic representation model is obtained by training based on preset corpus samples by using a knowledge distillation method; the preset corpus sample comprises: normal corpus samples;

14. The method of any of claims 1 to 12, wherein the preset semantic representation model comprises: BERT model.

15. A risk content recognition device of user privacy data is arranged at a client and comprises:

the user privacy data monitoring module monitors user privacy data generated by a user by using a target application in the client;

the privacy data semantic identification module is used for carrying out semantic identification on the user privacy data by utilizing a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at the client;

and the semantic feature vector uploading module is used for sending the first semantic feature vector to a server if the semantic representation information comprises the first semantic feature vector used for representing the suspicious risk content statement, so that the server identifies the risk content of the user privacy data based on the first semantic feature vector.

16. The apparatus of claim 15, wherein the user privacy data monitoring module is to:

17. The apparatus of claim 16, wherein the target application comprises: accessing an applet or Html5 web page of a specified host application; the user privacy data monitoring module is configured to:

18. The apparatus of claim 15, wherein the user privacy data comprises: page text information browsed by the user under the target application; the private data semantic identification module:

19. The apparatus of claim 18, wherein the private data semantic identification module is to:

20. The apparatus of claim 18, wherein the semantic representation information further comprises: a second semantic feature vector for characterizing the risk-free content statement,

21. The apparatus of claim 18, wherein the private data semantic identification module is to:

22. The apparatus of claim 18, wherein the apparatus further comprises: a suspicious risk content determination module that:

23. The apparatus according to claim 22, wherein if the number of the suspected risk words is zero, the risk weight corresponding to the text content sentence is zero; the suspicious risk content determination module is configured to:

24. The apparatus of claim 15, wherein the target application comprises: accessing an applet or Html5 web page of a specified host application; the semantic feature vector uploading module is configured to:

25. The apparatus of claim 15, wherein the semantic feature vector upload module is to:

26. The apparatus of claim 25, wherein the apparatus further comprises: an application processing result prompting module that:

27. The apparatus of any one of claims 15 to 26, wherein the apparatus further comprises: a semantic representation model deployment module that:

28. The apparatus according to any one of claims 15 to 26, wherein the preset semantic representation model comprises: BERT model.

29. A system for risk content identification of user privacy data, comprising: a client and a server;

the client monitors user privacy data generated by a user by using a target application in the client; performing semantic recognition on the user privacy data by using a preset semantic representation model to obtain corresponding semantic representation information; the preset semantic representation model is a semantic representation model which is processed by a preset knowledge distillation method and then is deployed at the client; if the semantic representation information comprises a first semantic feature vector used for representing the suspicious risk content sentence, the first semantic feature vector is sent to a server;

the server receives the first semantic feature vector uploaded by the client; and performing risk content identification on the user privacy data based on the first semantic feature vector.

30. A risky content identification apparatus of user privacy data, comprising:

a processor; and

a memory arranged to store computer executable instructions that, when executed, cause the processor to:

31. A storage medium storing computer-executable instructions that, when executed by a processor, implement a method of: