CN109918939A - User query risk assessment and method for secret protection based on HMM - Google Patents
User query risk assessment and method for secret protection based on HMM Download PDFInfo
- Publication number
- CN109918939A CN109918939A CN201910072616.2A CN201910072616A CN109918939A CN 109918939 A CN109918939 A CN 109918939A CN 201910072616 A CN201910072616 A CN 201910072616A CN 109918939 A CN109918939 A CN 109918939A
- Authority
- CN
- China
- Prior art keywords
- risk
- user
- user query
- inquiry
- secret protection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention provides a kind of user query risk assessment based on Hidden Markov Model (Hidden Markov Model, HMM) and method for secret protection.It is analyzed by the feature to user query, the index that analysis is obtained establishes user query risk evaluation model as the quantizating index of HMM;Initialization model parameter;According to the time of day of visible state sequence and system, model is trained;Finally work as user query, user query Risk Calculation and risk class are assessed.The present invention evaluates user query security risk using HMM model, considers the dynamic in each stage, real time reaction risk status.User query risk is reduced using high-intensitive differential noise for high risk inquiry, inquires for low-risk and is protected using low intensive differential noise, the risk of privacy leakage when not only efficiently solving user query, but also saved secret protection cost.The model has very strong scalability simultaneously, can be applied in various online query services.
Description
Technical field
The user query risk assessment and method for secret protection that the present invention relates to a kind of based on HMM model, belong to WEB data
Inquiry, secret protection field.
Background technique
In recent years, online query service is to bring great convenience in people's information retrieval, but also bring therewith various
Privacy leakage problem.User is left a series of comprising personal information, hobby and looking into when being serviced using various online queries
The digital trace being intended to is ask, includes user's sensitive information abundant in these digital traces, once leakage can cause user
Serious harm.As attacker's (insincere service provider or third party's marketer) is inferred by analysis user data trace
The true query intention of user.Infer what user is look for, when and when user initiates inquiry operation, with
Just more related and customization induced inner content or advertisement are provided to induce user's blindly consumption either user cheating.This to use
Family is unable to control to abuse " curiosity " system of its personal information and carry out targeted advertisement and number discrimination, causes the public
Serious concern to privacy infringement.
In user query, the method for traditional secret protection is concentrated mainly on the recognizable aspect of privacy, i.e., sensitive letter
Breath is deleted, secure communication, and anonymity inquiry improves Privacy Protection when user's online query with data obfuscation.Although
It has done a few thing, but still there are problems that serious privacy leakage, such as encrypted at high cost, the problems such as flexibility is poor makes above
Method is not widely used.
Existing method for secret protection mainly has inquiry to obscure and cover the solution of inquiry.Inquiry, which is obscured, passes through life
It is sent collectively to service providing end at virtual inquiry and the true inquiry of user, to prevent to search service side to user query
It is accurate to infer.Covering inquiry is to generate to cover by using the method for potential applications index to inquire to hide the original of user and look into
It askes.But the problem of risk height is distinguished is inquired when having ignored user query in above method.In actual queries scene,
Each inquiry of user is not directed to privacy, i.e., inquiry is not necessarily all high risk every time, if using identical privacy
Guard method easily causes secret protection intensity is excessively high to lead to that user query accuracy is low and search efficiency is low.
HMM model is a kind of important probabilistic model of sequence data processing and statistical learning, has and models simple, data meter
The features such as calculation amount is small, the speed of service is fast, discrimination is high.HMM is widely used to pattern-recognition, part-of-speech tagging and information extraction side
Face, the method for combining qualitative and quantitative well have relatively accurate assessment accuracy.
Summary of the invention
The purpose of the present invention is: HMM is applied in user query risk assessment and secret protection, user is efficiently solved
The judgement of content risks height is inquired in inquiry, and high risk inquiry user is protected using privacy protection policy, prevents from using
The problem of family causes privacy of user to be revealed when inquiring.
In order to achieve the above object, the technical solution of the present invention is to provide a kind of, and the user query risk based on HMM is commented
Estimate and method for secret protection, which comprises the following steps:
Step 1, user initiate inquiry request, according to user query request included in inquiry content carry out query characteristics
Analysis obtains user query feature;
Step 2 is based on user query feature, establishes HMM model;
Step 3, initialization model parameter, according to the visible state sequence of HMM model and the time of day of system, to HMM
Model is trained;
Step 4 is carried out using the inquiry content that trained HMM model is included to the inquiry request that user initiates in real time
Risk assessment and value-at-risk calculate, and determine inquiry risk class;
Step 5, for different inquiry risk class, use different secret protection measures: when inquiry risk class for
When high risk is inquired, risk is inquired using high-intensitive difference privacy lower noise;When inquiry risk class is low-risk inquiry
When, it is realized and is protected using low intensive difference privacy noise;
Result after secret protection is sent to service provider by step 6, and service provider is according to the query demand of user
Carry out result queries;
Step 7, service provider return to the result inquired, carry out the operation of result ranking in user terminal, then
Secondary carry out secret protection.
Preferably, in step 1, when user initiates inquiry request every time, in user query, there are progressive and co-occurrence features.
Preferably, the method for HMM model is established described in step 2 the following steps are included:
Step 201, visible state when determining user query;
Step 202 establishes hidden Markov five-tuple parameter model, including state transition probability matrix, observation vector
Probability matrix, initial conditions probability distribution vector, status number and observation symbolic number.
Preferably, in step 201, all information of the visible state comprising system, and the sight under current state
It is independent for examining, and the inquiry content of user is only related with preceding state.
Preferably, in step 202, the safe condition probability distribution of each link is exactly the initial state probabilities of next link
Distribution.
Preferably, in step 4, if user query sequence is X=(x1, x2..., xn), wherein xiI-th of node is represented,
And there is a transition probability P for each node, then, on a hidden Markov model M, a search sequence X quilt
The probability of observation is the sum of the probability on all possible paths:
In formula, P (X | M) indicates the Joint Distribution of user query sequence and observed result, q1..., qnIndicate observed result
Node, QlIndicate observed result node set, P (qk-1→qk) indicate from qk-1To qkTransition probability, P (xk|qk) indicate to be used for
By xkIt queried qkProbability.
Preferably, step 5 is that different inquiry risks uses different method for secret protection, when the inquiry of user is high wind
When the inquiry of danger, the risk of user query is reduced using high-intensitive difference privacy noise.When the inquiry of user is looked into for low-risk
When inquiry, protected using low intensive difference random noise.
The present invention has the advantage that
(1) a kind of method that the present invention proposes dynamic assessment user query risk, with existing static risk appraisal procedure
Difference, the present invention dynamically assess threat caused by user query content, and inquiring risk size can be with user
The variation of query time, inquiry times etc. and it is different.
(2) present invention in secret protection, according to the difference of risk height using varying strength privacy protection policy into
Row protection, solves the height that user query risk is not differentiated between in existing method for secret protection, using identical secret protection plan
Defect slightly;
(3) compared with the existing technology, the present invention assesses user query risk using HMM model, considers each rank
The dynamic of section, obtains user query risk status in real time, which has very strong scalability, can be applied to
The various fields of online service.
Detailed description of the invention
Fig. 1 is user query risk assessment and method for secret protection flow chart based on HMM model;
Fig. 2 is user query risk assessment and method for secret protection model based on HMM model.
Specific embodiment
Present invention will be further explained below with reference to specific examples.It should be understood that these embodiments are merely to illustrate the present invention
Rather than it limits the scope of the invention.In addition, it should also be understood that, after reading the content taught by the present invention, those skilled in the art
Member can make various changes or modifications the present invention, and such equivalent forms equally fall within the application the appended claims and limited
Range.
1, user query model is summarized: as shown in Fig. 2, for a kind of user query risk assessment based on HMM model and hidden
Private guard method model, whole process include three phases, i.e. user query and risk assessment, and inquiry secret protection and inquiry are tied
Fruit returns to and resets name.(1) user query Risk Assessment Stage, user input the keyword for oneself needing to inquire by user terminal
Or content, client (secret protection part) carry out risk assessment to user query, determine wind involved in user query
Danger height;(2) the secret protection stage is inquired, is looked into for high risk inquiry using high-intensitive privacy protection policy to reduce user
Risk is ask, for low point of danger by the way of adding stochastic difference noise, the user query after secret protection are passed through
Internet is transferred in SP;(3) query result is returned and is reset the name stage, and SP carries out content according to the key word of the inquiry of user
Retrieval, query result is returned.It is analyzed to prevent insincere SP from clicking to user, in user terminal, combination user is original
Inquiry (for the true inquiry of user) newly inquires (inquiry after secret protection) result to user and carries out weight ranking process, final to use
Family gets true query result.
2, Hidden Markov Model (HMM) is summarized: HMM is grown up on the basis of Markov Chain.In HMM
The event observed is the random function of state, therefore the model is a dual random process, i.e. an observation state, one
Hidden state.In terms of HMM is widely used to pattern-recognition, part-of-speech tagging and information extraction.
3, HMM definition and analysis: to become apparent from description, HMM can be indicated in the form of 5 yuan of ancestrals, such as < N, M, π,
A, B >, wherein N is the status number in model, and state set is represented by S={ S1..., SN};M is observation symbolic number, observed result
Collection can be expressed as O={ O1..., OM, observed result indicates the result number that each state may export;π represents initial point
Cloth state;A is state transition probability matrix, i.e. A=(aij), the wherein a in AijIndicate t moment by state SiIt is transferred to state Sj
Probability;B represents output probability matrix, i.e. B={ bj(Ok), bj(Ok)=P (Ok|Sj) indicate t moment state SjOutput observation
Value OkProbability.The following is<N, M, π, A, B>formula indicate:
Wherein, π represents initial distribution state:
A is state transition probability matrix, i.e.,
B is the probability matrix for observing vector, i.e. B=(bij)N×M(1≤j≤N, 1≤k≤M)
bjkIt indicates in the case where state j, the probability that observation state k occurs, it may be assumed that
bjk=p (Vk|aj), 1≤j≤N, 1≤k≤M
In formula, VkIndicate observation state k, ajIndicate that the transition probability of j state, p () indicate the probability that certain state occurs.
4, user query risk analysis: when establishing HMM model, in conjunction with the progressive and co-occurrence query characteristics of user, from turning
It moves probability and observation probability angle is analyzed.
(1) transition probability, after referring to that user provides prior queries data sequence, using obtaining user's after some time
The conditional probability of another inquiry data, such as two connected node q1,q2Between there is a transition probability P (q1→q2).Due to
User is under continuous-query scene, and the differentiable risk of user query data depends on the inquiry data before user, if examined
Consider the past data in same subject, then the information gain of the data can be got higher.If XtFor recognizable sensitive information personal in HMM
Either sensitive theme, then XtTransition probability between node is p (Xt|Xt-1), it can be to the conversion occurred between node time
Number is weighted, i.e.,Then it is calculated according to the method for weighting transition probability hidden in user query
Private risk, i.e. α * p (Xt|Xt-1)。
(2) observation probability refers to the User behavior that some node may occur, such as user uiIt queried the probability of e by q
For P (e | q).Acquisition is analyzed and calculated to the value based on the historical query data of user, and each node, which includes one group, has sight
These observation probabilities are modeled as different user in past data (p (u by the observation for surveying probabilityi|Xt)) in find to fixed number
According to XtProbability.The data of user query specific subject are more, then higher to the deduction accuracy of user interest data, the inquiry
Risk is higher.Similarly, inquiry risk is determined by the way of weighted count, i.e.,Because user is more uniform,
It is higher to inquire privacy risk, i.e. β * p (ui|Xt)。
In user's continuous-query scene, if the current inquiry of user is XTOnly with previous inquiry XT-1It is related.As user exists
When inquiring a challenge (medical knowledge etc.), when current query result is unsatisfactory for the query demand of user, user can add
Add or delete previous inquiry content, so being to meet single order Markov property.
5, user query risk assessment:
If user uiSearch sequence is X1, X2..., Xn, then the observed result for user query sequence output is Y1,
Y2..., Yn.User u can be obtainediThe Joint Distribution of search sequence and observed result are as follows:
User u can be calculatediSearch sequence (X1→X2→…→Xn) caused by entirety privacy risk are as follows:
Wherein (HMM | ui) represent user uiAll paths privacy list of probabilities, these lists include that user's observation is general
Rate is greater than 0 node, finally, can obtain user's search sequence is X1, X2..., XnWhen inquiry risk be p (X1..., Xn|ui)。
6, user query risk class divides: we assume that there are five states altogether for user query, A1-A5, wherein A1 is indicated
Normal safe condition, A5 indicate substantial risk situation, and A2, A3, A4 indicate that danger classes is deepened step by step.If indicated with probability,
The probability for the SC risk grade that then each state indicates is as shown in table 1.
The risk class of 1 user query of table divides
Inquiry risk class division of the invention meets " GB/T 33132-2016 information security technology, Information Security Risk
The demand of processing implementation guide " and " GB/T 31722-2015 information technology, safe practice, Management of risk of information security ".
7, risk secret protection is inquired:
The inquiry risk class of user is higher, then secret protection intensity Ying Yueqiang.In inquiry secret protection, the present invention is adopted
Secret protection is realized with inquiry addition random noise of the difference secret protection technology to user.Difference privacy is resisting background attack
Etc. there is effect well.If Q is a group polling function, ε-difference privacy can be realized by addition random noise r,
I.e.Wherein r is random noise.The size of secret protection intensity is realized by r, and the r the big, represents addition
Random noise is more, then secret protection intensity is higher.When the inquiry of user is that high risk is inquired, then need to add more random
Noise needs to add few inquiry noise if the inquiry of user is low inquiry.
8, query result returns and user terminal resets name:
Query result is returned into user terminal by Internet according to the inquiry request SP of user.The knot that SP inquiry returns
Not only really inquire comprising user in fruit and inquire (noise inquiry) comprising vacation again, therefore, user terminal carries out rearrangement name to returning the result
Operation.The search result of ranking in the result and client user's configuration file that the present invention is returned by Fusion query.It is specific and
Speech, distribution correspond to the rank score for the position that candidate documents occur in the ranked list of each ranking person, and candidate
It sorts according to the overall ranking score:
Wherein Ri(d) ranking that represent data d is i, and α controls the weight of each query result.It is base in the ranking stage
Come to carry out ranking process to query result in original user query.
Resetting the name stage, the present invention except the weight for considering each page and in addition to the similarity problem of the true page,
Set frequency of occurrences threshold value beta and false access mechanism also to cover the true inquiry of user.Frequency of occurrences threshold value beta represents result page
No more than β information, this mechanism only has user to know for the true inquiry occurred in face, user when accessing final result,
Will appreciate that result returns to those in page is true result.While in order to confuse attacker (insincere service provider or third
Square marketer), present system can click at random vacation inquiry in the result of return, but this process is not need
User participates in.The final purpose for realizing the dual secret protection of user query.
The foregoing describe the information such as basic principles and main features of the invention and embodiment, but the present invention is not by upper
The limitation for stating implementation process, under the premise of not departing from spirit and range, the present invention can also have various changes and modifications.
Therefore, unless this changes and improvements are departing from the scope of the present invention, they should be counted as comprising in the present invention.
Claims (7)
1. a kind of user query risk assessment and method for secret protection based on HMM, which comprises the following steps:
Step 1, user initiate inquiry request, according to user query request included in inquiry content carry out query characteristics point
Analysis obtains user query feature;
Step 2 is based on user query feature, establishes HMM model;
Step 3, initialization model parameter, according to the visible state sequence of HMM model and the time of day of system, to HMM model
It is trained;
Step 4 carries out risk using the inquiry content that trained HMM model is included to the inquiry request that user initiates in real time
Assessment and value-at-risk calculate, and determine inquiry risk class;
Step 5, for different inquiry risk class, use different secret protection measures: when inquiry risk class for high wind
When the inquiry of danger, risk is inquired using high-intensitive difference privacy lower noise;When inquiring risk class is that low-risk is inquired, adopt
It is realized and is protected with low intensive difference privacy noise;
Result after secret protection is sent to service provider by step 6, and service provider carries out according to the query demand of user
Result queries;
Step 7, service provider return to the result inquired, user terminal carry out result ranking operation, again into
Row secret protection.
2. a kind of user query risk assessment and method for secret protection, feature based on HMM according to claim 1 exists
In in step 1, when user initiates inquiry request every time, in user query, there are progressive and co-occurrence features.
3. a kind of user query risk assessment and method for secret protection based on HMM model according to claim 1, special
Sign is, the method for HMM model is established described in step 2 the following steps are included:
Step 201, visible state when determining user query;
Step 202 establishes hidden Markov five-tuple parameter model, the probability including state transition probability matrix, observation vector
Matrix, initial conditions probability distribution vector, status number and observation symbolic number.
4. a kind of user query risk assessment and method for secret protection based on HMM model according to claim 3, special
Sign is, in step 201, the visible state includes all information of system, and the observation under current state is independent
, and the inquiry content of user is only related with preceding state.
5. a kind of user query secret protection based on HMM model according to claim 3 and assessment, which is characterized in that
In step 202, the safe condition probability distribution of each link is exactly the initial state probabilities distribution of next link.
6. a kind of user query risk assessment and method for secret protection based on HMM model according to claim 3, special
Sign is, in step 4, if user query sequence is X=(x1, x2..., xn), wherein xiRepresent i-th of node, and each section
All there is a transition probability P for point, then, on a hidden Markov model M, a search sequence X is observed general
Rate is the sum of the probability on all possible paths:
In formula, P (X | M) indicates the Joint Distribution of user query sequence and observed result, q1..., qnIndicate observed result node,
QlIndicate observed result node set, P (qk-1→qk) indicate from qk-1To qkTransition probability, P (xk|qk) indicate for passing through xk
It queried qkProbability.
7. a kind of user query risk assessment and method for secret protection based on HMM model according to claim 3, special
Sign is that step 5 is that different inquiry risks uses different method for secret protection, when the inquiry of user is high risk inquiry
When, the risk of user query is reduced using high-intensitive difference privacy noise.When the inquiry of user is that low-risk is inquired, adopt
It is protected with low intensive difference random noise.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910072616.2A CN109918939B (en) | 2019-01-25 | 2019-01-25 | HMM-based user query risk assessment and privacy protection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910072616.2A CN109918939B (en) | 2019-01-25 | 2019-01-25 | HMM-based user query risk assessment and privacy protection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109918939A true CN109918939A (en) | 2019-06-21 |
CN109918939B CN109918939B (en) | 2023-08-11 |
Family
ID=66960774
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910072616.2A Active CN109918939B (en) | 2019-01-25 | 2019-01-25 | HMM-based user query risk assessment and privacy protection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109918939B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113206839A (en) * | 2021-04-13 | 2021-08-03 | 广州锦行网络科技有限公司 | Data hiding and complementing method in data transmission |
CN116910223A (en) * | 2023-08-09 | 2023-10-20 | 北京安联通科技有限公司 | Intelligent question-answering data processing system based on pre-training model |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100318546A1 (en) * | 2009-06-16 | 2010-12-16 | Microsoft Corporation | Synopsis of a search log that respects user privacy |
US20120023586A1 (en) * | 2010-07-22 | 2012-01-26 | International Business Machines Corporation | Determining privacy risk for database queries |
CN103049643A (en) * | 2012-11-22 | 2013-04-17 | 无锡南理工科技发展有限公司 | Mobile ad hoc network security risk assessment method based on risk entropy method and markoff chain method |
CN103279499A (en) * | 2013-05-09 | 2013-09-04 | 北京信息科技大学 | User privacy protection method in personalized information retrieval |
CN103530603A (en) * | 2013-09-24 | 2014-01-22 | 杭州电子科技大学 | Video abnormality detection method based on causal loop diagram model |
US20140026221A1 (en) * | 2012-07-17 | 2014-01-23 | Oracle International Corporation | Defense against search engine tracking |
CN105608536A (en) * | 2015-12-22 | 2016-05-25 | 天津科技大学 | Food safety risk prediction method based on hidden Markov model |
CN107862219A (en) * | 2017-11-14 | 2018-03-30 | 哈尔滨工业大学深圳研究生院 | The guard method of demand privacy in a kind of social networks |
CN108197492A (en) * | 2017-12-29 | 2018-06-22 | 南京邮电大学 | A kind of data query method and system based on difference privacy budget allocation |
CN108520182A (en) * | 2018-04-09 | 2018-09-11 | 哈尔滨工业大学深圳研究生院 | A kind of demand method for secret protection based on difference privacy and correlation rule |
CN108537055A (en) * | 2018-03-06 | 2018-09-14 | 南京邮电大学 | A kind of privacy budget allocation of data query secret protection and data dissemination method and its system |
-
2019
- 2019-01-25 CN CN201910072616.2A patent/CN109918939B/en active Active
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100318546A1 (en) * | 2009-06-16 | 2010-12-16 | Microsoft Corporation | Synopsis of a search log that respects user privacy |
US20120023586A1 (en) * | 2010-07-22 | 2012-01-26 | International Business Machines Corporation | Determining privacy risk for database queries |
US20140026221A1 (en) * | 2012-07-17 | 2014-01-23 | Oracle International Corporation | Defense against search engine tracking |
CN103049643A (en) * | 2012-11-22 | 2013-04-17 | 无锡南理工科技发展有限公司 | Mobile ad hoc network security risk assessment method based on risk entropy method and markoff chain method |
CN103279499A (en) * | 2013-05-09 | 2013-09-04 | 北京信息科技大学 | User privacy protection method in personalized information retrieval |
CN103530603A (en) * | 2013-09-24 | 2014-01-22 | 杭州电子科技大学 | Video abnormality detection method based on causal loop diagram model |
CN105608536A (en) * | 2015-12-22 | 2016-05-25 | 天津科技大学 | Food safety risk prediction method based on hidden Markov model |
CN107862219A (en) * | 2017-11-14 | 2018-03-30 | 哈尔滨工业大学深圳研究生院 | The guard method of demand privacy in a kind of social networks |
CN108197492A (en) * | 2017-12-29 | 2018-06-22 | 南京邮电大学 | A kind of data query method and system based on difference privacy budget allocation |
CN108537055A (en) * | 2018-03-06 | 2018-09-14 | 南京邮电大学 | A kind of privacy budget allocation of data query secret protection and data dissemination method and its system |
CN108520182A (en) * | 2018-04-09 | 2018-09-11 | 哈尔滨工业大学深圳研究生院 | A kind of demand method for secret protection based on difference privacy and correlation rule |
Non-Patent Citations (9)
Title |
---|
SICONG ZHANG ET AL.: "Anonymizing Query Logs by Differential Privacy", 《SIGIR "16: PROCEEDINGS OF THE 39TH INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL》 * |
刘安等: "基于灰色-马尔可夫模型的粮食产量预测", 《计算机技术与发展》 * |
刘琳琳等: "大数据环境下个性化检索中数据隐私保护方法", 《信息与电脑(理论版)》 * |
吴云乘等: "一种基于时空相关性的差分隐私轨迹保护机制", 《计算机学报》 * |
康海燕等: "位置服务隐私保护", 《山东大学学报(理学版)》 * |
林青: "大数据背景下隐私保护方法研究", 《信息与电脑(理论版)》 * |
毋文敏等: "基于差分隐私的协同过滤推荐***的设计与实现", 《信息与电脑(理论版)》 * |
鄢发齐等: "基于隐马尔可夫模型的电力调度语音识别研究", 《武汉大学学报(工学版)》 * |
黄影: "隐马尔可夫模型演化下的隐组查询", 《电子科技》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113206839A (en) * | 2021-04-13 | 2021-08-03 | 广州锦行网络科技有限公司 | Data hiding and complementing method in data transmission |
CN113206839B (en) * | 2021-04-13 | 2022-05-20 | 广州锦行网络科技有限公司 | Data hiding and complementing method in data transmission |
CN116910223A (en) * | 2023-08-09 | 2023-10-20 | 北京安联通科技有限公司 | Intelligent question-answering data processing system based on pre-training model |
CN116910223B (en) * | 2023-08-09 | 2024-06-11 | 北京安联通科技有限公司 | Intelligent question-answering data processing system based on pre-training model |
Also Published As
Publication number | Publication date |
---|---|
CN109918939B (en) | 2023-08-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Cresci et al. | Fame for sale: Efficient detection of fake Twitter followers | |
Leroy et al. | Cold start link prediction | |
Li et al. | A ranking method of triangular intuitionistic fuzzy numbers and application to decision making | |
Balsa et al. | OB-PWS: Obfuscation-based private web search | |
Cheng et al. | A content-driven framework for geolocating microblog users | |
Edwards et al. | A systematic survey of online data mining technology intended for law enforcement | |
Horne et al. | Different spirals of sameness: A study of content sharing in mainstream and alternative media | |
US20130138428A1 (en) | Systems and methods for automatically detecting deception in human communications expressed in digital form | |
US20190222602A1 (en) | Determining digital vulnerability based on an online presence | |
Bhagat et al. | Applying link-based classification to label blogs | |
Schroeder et al. | Automated criminal link analysis based on domain knowledge | |
Dougnon et al. | Inferring user profiles in online social networks using a partial social graph | |
Bilenko et al. | Targeted, not tracked: Client-side solutions for privacy-friendly behavioral advertising | |
Jabeur et al. | Uprising microblogs: A Bayesian network retrieval model for tweet search | |
Cresci et al. | A Fake Follower Story: improving fake accounts detection on Twitter | |
CN109918939A (en) | User query risk assessment and method for secret protection based on HMM | |
Li et al. | Context-based people search in labeled social networks | |
Abdalla et al. | A review of using online social networks for investigative activities | |
Moghaddam et al. | AgeTrust: A new temporal trust-based collaborative filtering approach | |
Liu et al. | Detecting tag spam in social tagging systems with collaborative knowledge | |
Ghufran et al. | Toponym disambiguation in online social network profiles | |
Tsikerdekis et al. | Detecting online content deception | |
Abid et al. | Online testing of user profile resilience against inference attacks in social networks | |
Jäger et al. | Introducing the factor importance to trust of sources and certainty of data in knowledge processing systems-a new approach for incorporation and processing | |
Andoh-Baidoo | Explaining investors' reaction to internet security breach using deterrence theory |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |