CN107357516A - A kind of gesture query intention Forecasting Methodology based on hidden Markov model - Google Patents
A kind of gesture query intention Forecasting Methodology based on hidden Markov model Download PDFInfo
- Publication number
- CN107357516A CN107357516A CN201710556277.6A CN201710556277A CN107357516A CN 107357516 A CN107357516 A CN 107357516A CN 201710556277 A CN201710556277 A CN 201710556277A CN 107357516 A CN107357516 A CN 107357516A
- Authority
- CN
- China
- Prior art keywords
- gesture
- query intention
- events
- mrow
- active user
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
- G06F3/048—Interaction techniques based on graphical user interfaces [GUI]
- G06F3/0487—Interaction techniques based on graphical user interfaces [GUI] using specific features provided by the input device, e.g. functions controlled by the rotation of a mouse with dual sensing arrangements, or of the nature of the input device, e.g. tap gestures based on pressure sensed by a digitiser
- G06F3/0488—Interaction techniques based on graphical user interfaces [GUI] using specific features provided by the input device, e.g. functions controlled by the rotation of a mouse with dual sensing arrangements, or of the nature of the input device, e.g. tap gestures based on pressure sensed by a digitiser using a touch-screen or digitiser, e.g. input of commands through traced gestures
Abstract
The present invention relates to a kind of gesture query intention Forecasting Methodology based on hidden Markov model, start with from user gesture interaction, analyze the gesture classification commonly used in gesture search and its corresponding characteristic attribute, it is then based on hidden Markov characteristic structure gesture query intention model, go out the parameter of gesture query intention model using gesture interaction feature calculation on this basis, and go out optimal query intention corresponding to gesture interaction event using Viterbi theoretical prediction.The present invention meets the characteristic of hidden markov process based on gesture interaction process, and the query intention of user is captured using gesture operation feature, certain guiding is provided for user gesture interaction.Present invention can apply under the search scene of the heuristic towards gesture interaction so that user obtains the information for meeting intention during gesture interaction, improves the fluency and satisfaction of gesture interaction.
Description
Technical field
The present invention relates to a kind of gesture query intention Forecasting Methodology based on hidden Markov model, belong to man-machine interaction,
Technical field of information retrieval.
Background technology
With the fast development for touching interaction technique, according to statistics, to 2017,87% global intelligent networking device was flat
Plate computer and smart mobile phone, PC shares are then less than 13%, and this explanation multiple point touching interactive device is by as the main flow of man-machine interaction
Product, gesture interaction are increasingly becoming one of main mode of man-machine interaction.Thus user uses upward in information seeking processes
The gestures such as slip, slide downward, amplification, diminution and click carry out information searching, in order to deepen the understanding to looked into content, use
Family performs above-mentioned gesture repeatedly, continuous iteration ground trial and error, but during actual gesture interaction, most of users wish logical
Cross less gesture operation and obtain maximally effective information, if system can not accurately provide the information for meeting user view, then
User can perform gesture operation repeatedly, and this process to iterate will produce a large amount of useless gesture operations, heavy system
Inquiry burden, cause system response delay, user's search efficiency substantially reduces, thus needs to provide and draw for user gesture interaction
Lead, wherein how to be inferred to the query intention of user according to the gesture interaction process and alternative events of user is asking of paying close attention to
Topic.
At present, query intention modeling method is based primarily upon inquiry log data, solves query intention not in terms of different
The problem of specifying, two classes can be divided into:Query intention modeling method based on intensified learning and the query intention based on probability are built
Mould method.Query intention modeling method wherein based on intensified learning analyzes use using the exploratory evaluation procedure of intensified learning
The search behavior at family, user mutual intent model is established according to the interaction detail provided on visualization interface to infer the shape of exploration
State, this method compares emphasis using the online feedback of user to establish query intention model, but collects the online feedback of user
The cost of cost is higher, and in most cases user is displeased spends too much of his time participation evaluation procedure, the positive degree of participation
It is not high, thus cause the field feedback of collection less.
Related priori and search are mainly distributed according to user view in query intention modeling method based on probability
As a result the probability distribution situation of middle intention is to establish the contact between inquiry and intention, and defines inquiry meaning using this contact
Graph model predicts the query demand situation of user with this.A such as patent of invention《Query intention prediction based on artificial intelligence
Method and apparatus》(application number:201610728086.9 grant number:The A of CN 106372132), according to original retrieval sentence with looking into
The corresponding relation training intent classifier model being intended to is ask, probability is carried out to the hidden layer value by the output layer of intent classifier model
Analysis, obtain the query intention of the retrieval sentence.The log information that this Forecasting Methodology is interacted based on user with search engine
Analysis user corresponds to the click behavior of search result to original retrieval sentence, primary concern is that common click behavior, does not have
Consider the application demand of gesture interaction.And another patent of invention《Touch to search for》(application number:201380072159.8 award
Quan Hao:CN 104969164 A) in user by using gesture rather than can key in search inquiry to searching interface and select
Shown content on touch-screen, candidate search query set is selected based on the content subset of gesture data mark, and to every
Individual candidate search inquires about calculability score, and final choice goes out one or more candidate search inquiries, although the invention is abundant
Gesture data is make use of, but is not combined gesture data with user's query intention.
To sum up, most of method does not have emphasis to consider the relation between gesture operation feature and user's query intention, therefrom
Extraction meets the information of user view.
The content of the invention
It is pre- that the technical problems to be solved by the invention are to provide a kind of gesture query intention based on hidden Markov model
Survey method, the actual mechanical process of gesture interaction is considered, using gesture interaction feature and the relation of user's query intention, based on hidden
Markov property builds gesture query intention model, and goes out gesture interaction event based on Viterbi theoretical prediction on this basis
Corresponding optimal query intention.
In order to solve the above-mentioned technical problem the present invention uses following technical scheme:The present invention devises one kind and is based on hidden Ma Er
Can husband's model gesture query intention Forecasting Methodology, comprise the following steps:
Step A. initializes gesture path information aggregate Gesture and gesture interaction contextual information set Context
Sky, and enter step B;
Step B. detections obtain the gesture path information aggregate Gesture and gesture corresponding to active user's gesture to be predicted
Interaction context information aggregate Context, subsequently into step C;
Gesture path information aggregate Gestures and gesture interaction of the step C. according to corresponding to active user's gesture to be predicted
Contextual information set Context, the gesture interaction event corresponding to active user's gesture to be predicted is built, subsequently into step
D;
The gesture interaction historical data of active user in step D. extraction search engines, and combine active user's hand to be predicted
Alternative events corresponding to gesture, active user's gesture query intention model is built based on hidden Markov characteristic, subsequently into step
E;
Step E. is calculated according to the gesture interaction historical data of active user and is obtained active user's gesture query intention model
Initiation parameter, subsequently into step F;
Step F. is according to the initiation parameter of active user's gesture query intention model, and active user's gesture inquiry meaning
Graph model, it is theoretical based on Viterbi, predict the optimal inquiry meaning corresponding to alternative events corresponding to active user's gesture to be predicted
Figure, submit and perform, realize that gesture query intention is predicted.
It is as follows as a preferred technical solution of the present invention, in addition to step G, after the execution of step F, enter
Step G;
Step G. empties gesture path information aggregate Gesture and gesture interaction contextual information set Context, and returns
Return step B.
As a preferred technical solution of the present invention, gesture path information aggregate Gesture is as follows in the step A:
Gesture={ Pi,t(xi,t,yi,t,timei,t) | i ∈ [1, I], I ∈ [1,10] }, wherein, Pi,t(xi,t,yi,t,
timei,t) for the positional information of each touch point in gesture operation detection plate, i ∈ [1, I], I ∈ [1,10], I represent touch point
Quantity, i are integer, represent that in all touch points i-th of touch point in gesture operation detection plate, xi,t、yi,tRepresent respectively current
Horizontal and vertical coordinate of i-th of the touch point of moment t in gesture operation detection plate, timei,tRepresent in gesture operation detection plate
Timestamp caused by i-th of touch point in all touch points;
Gesture interaction contextual information set Context is as follows:
Context={ Contexti,t| i ∈ [1, I], I ∈ [1,10] }, wherein,
For corresponding to each touch point above and below gesture interaction
Literary information, Ki,tFor i-th of touch point of current time in gesture operation detection plate corresponding keyword,When representing current
Carve each touch points of t and correspond to keyword Ki,tThe r articles Query Result title, R is keyword Ki,tQuery Result sum, r is
Integer.
As a preferred technical solution of the present invention, the step B comprises the following steps:
Current time t is assigned to time by step B1.start, and enter step B2;
Step B2. judges to whether there is touch point in gesture operation detection plate, is then to enter step B3;Otherwise order is current each
It is sky to touch dot position information and gesture interaction contextual information, and updates gesture path information aggregate Gesture and gesture
Interaction context information aggregate Context, subsequently into step B4;
Step B3., which is recorded, respectively touches dot position information and gesture interaction contextual information in current gesture operation detection plate,
Add and update gesture path information aggregate Gesture and gesture interaction contextual information set Context, and return to step
B2;
Whether step B4. judges gesture path information aggregate Gesture and gesture interaction contextual information set Context
It is then return to step B1 for sky;Otherwise current time t is assigned to timeend, then [timestart,timeend] hand in the period
Gesture trace information set Gesture and gesture interaction contextual information set Context, as active user gesture institute to be predicted
Corresponding gesture path information aggregate Gesture and gesture interaction contextual information set Context, subsequently into step C.
As a preferred technical solution of the present invention, the step C specifically includes as follows:
Gesture path information aggregate Gesture and gesture interaction context according to corresponding to active user's gesture to be predicted
Information aggregate Context, define [timestart,timeend] gesture interaction event in the period for Events=gb, dt, v,
K, T }, the as gesture interaction event corresponding to active user's gesture to be predicted, wherein, gb is to be identified based on gesture path information
The gesture gone out;Dt be gesture interaction event Events residence time, dt=timestart-timeend;V is the speed of gesture operation
Degree, and according to gesture path information Gesture, as follows:
The speed v for calculating gesture operation is;K={ Ki,t|i∈[1,I],I∈[1,10],t∈[timestart,
timeend] it is [timestart,timeend] gesture selection in the period keyword set;For [timestart,timeend] hand in the period
The keyword query result head stack of gesture selection.
As a preferred technical solution of the present invention, the step D specifically includes as follows:
The gesture interaction historical data of active user in step D. extraction search engines, and combine active user's hand to be predicted
Alternative events corresponding to gesture are as follows based on hidden Markov characteristic structure active user's gesture query intention model:
GQIM=<Intents,Events,π,A,B>
Wherein, Intents is the hidden state set in active user's gesture query intention model, by gesture interaction history
The keyword set that user selects in data is combined into, Intents={ Kn| n ∈ [1, N] }, wherein n be query intention sequence number, Kn
For n-th of keyword of gesture selection, N is the keyword sum of gesture selection, i.e. query intention is total, 1≤n≤N, and n is
Integer;
Events is the Observable state set in active user's gesture query intention model, by gesture interaction historical data
Alternative events corresponding to middle gesture interaction event and active user's gesture to be predicted form, Events={ Eventsh∪EventsM
| h ∈ [1, M-1] }, wherein EventsMThe alternative events corresponding to active user's gesture to be predicted for being built in step E, M is works as
The sequence number of alternative events, Events corresponding to preceding user's gesture to be predictedhFor the hand in active user's gesture interaction historical data
The mutual event sets of power-relation, Eventsh={ gbh,dth,vh,Kh,Th| h ∈ [1, M-1] }, h is gesture in gesture interaction historical data
The sequence number of alternative events, 1≤h≤M-1, and h are integer;
π is the probability matrix of active user's gesture query intention model, corresponding to hidden in hidden Markov model
The matrix of probability containing state, π={ πn| n ∈ [1, N] }, wherein, πn=P (Kn) represent that query intention K occursnIt is initial general
Rate;
A is the query intention transition probability matrix of active user's gesture query intention model, corresponding to hidden Markov mould
Hidden state transition probability matrix in type, A={ ast| s, t ∈ [1, N] }, wherein ast=P (Kt,Ks) represent active user's
Gesture interaction event is KsWhen, the query intention transfer of gesture interaction is K next timetProbability, wherein for arbitrary 1≤s≤
N, meetAnd 0≤ast≤1;
Transition probability matrixs of the B between gesture query intention and gesture interaction event, corresponding to hidden Markov model
In observer state transition probability matrix, B={ bmn| m ∈ [1, M], n ∈ [1, N] }, wherein, bmn=P (Eventsm|Kn) represent
Based on query intention Kn, the gesture interaction event transfer of active user is EventmProbability, wherein for arbitrary 1≤n≤N,
MeetAnd 0≤bmn≤1。
As a preferred technical solution of the present invention, the step E specifically includes as follows:
Step E01. calculates the probability matrix π of active user's gesture query intention model, counts gesture interaction history
Each query intention K in datanThe number of appearance, it is designated as Count (Kn), then calculate query intention KnProbabilityAnd enter step E02;
Step E02. calculates the query intention transition probability matrix A of active user's gesture query intention model, and statistics is sold
User's query intention K in the mutual historical data of power-relationsAnd KtBetween the frequency changed, be designated as cnt (Ks,Kt), then calculate inquiry meaning
Transition probability between figureAnd enter step E03;
Step E03. calculates the transition probability matrix B between query intention and gesture interaction event:
Count each gesture interaction event Events in gesture interaction historical datamThe number of generation, is designated as Count
(Eventsm), and utilize formulaCalculate and gesture interaction event occurs
EventsmProbability;
Statistical query is intended to KnIn gesture interaction event EventsmThe number of middle appearance, it is designated as cnt (Kn,Eventsm), so
Formula is utilized afterwardsCalculate
Query intention KnIn gesture interaction event EventsmUnder the premise of the conditional probability that occurs, wherein utilizing formulaMeter
The weight size using the alternative events for sliding class gesture is calculated, utilizes formulaCalculate and click on class and scaling class
The weight size of the alternative events of gesture;
Query intention K is calculated using Bayesian formulanWith gesture interaction event EventsmBetween transition probabilityAnd enter step F.
As a preferred technical solution of the present invention, the step F specifically includes as follows:
Viterbi theoretical prediction based on hidden Markov model is obtained corresponding to active user's gesture to be predicted most
Excellent query intention, it is converted into known N number of query intention Intents={ Kn| n ∈ [1, N] } and M gesture interaction event Events
={ Eventsh∪EventsM| h ∈ [1, M-1] }, and model parameter λ=(π, A, the B) calculated, ask active user to be predicted
Alternative events Events corresponding to gestureMOptimal query intention, and be described as follows:
And 0≤asn≤1
And 0≤bmn≤1
Wherein, PM(Kn) represent that gesture interaction event is EventsMWhen, all query intention Intents={ Kn|n∈[1,
N] } probability that is likely to occur,For value maximum in these probability, now corresponding KnFor the optimal inquiry of prediction
It is intended to;The query intention for ensureing current gesture interaction is KsUnder conditions of, gesture interaction is looked into next time
It is K to ask intention transfernAll possible probability summation be 1;Ensure that current queries are intended to KnFeelings
Under condition, it is Events that gesture interaction event, which occurs,mAll possible probability summation be 1.
As a preferred technical solution of the present invention, in the step F, according to active user's gesture query intention model
Initiation parameter, and active user's gesture query intention model is theoretical based on Viterbi, prediction obtain active user treat it is pre-
The optimal query intention corresponding to gesture is surveyed, is specifically included as follows:
Step F01. utilizes formula P1(Kn)=πnb1n, it is Events that 1≤n≤N, which calculates gesture interaction event,1It is all can
The probability P of the query intention of energy1(Kn), and enter step F02;
Step F02. using the obtained all query intentions of step F01 probability P1(Kn), 1≤n≤N, and formulaIt is Events to calculate gesture interaction event2All possible query intention it is general
Rate P2(Kn), 1≤n≤N, subsequently into step F03;
Step F03. utilizes step F02 result, and according to formula
Gesture interaction event Events is calculated successively3,Events4,...,EventsMAll possible query intention probability P3
(Kn)、P4(Kn)、…、PM(Kn), subsequently into step F04;
The P that step F04. selecting steps F03 is obtainedM(Kn) in the optimal objective value of maximum probable value as the problem, this
When query intention K corresponding to the valuenFor the optimal query intention of prediction, the query intention is submitted and performed.
A kind of gesture query intention Forecasting Methodology based on hidden Markov model of the present invention uses above technical side
Case compared with prior art, has following technique effect:
(1) a kind of gesture query intention Forecasting Methodology based on hidden Markov model that the present invention designs, gesture is considered
Interactive actual search process, during hidden markov process is applied into gesture interaction, analyze query intention between, gesture
Transfer relationship between alternative events and query intention, the gesture query intention mould based on hidden Markov characteristic is built with this
Type, preferably express the incidence relation between query intention and gesture interaction event;
(2) present invention designs a kind of gesture query intention Forecasting Methodology based on hidden Markov model, is used according to current
The gesture interaction historical data at family calculates the initiation parameter of gesture query intention model, has fully demonstrated the personalization of user
Operating habit, and gesture operation feature and user's query intention are closely linked, query intention and hand have been used well
Transfer relationship between gesture alternative events so that the gesture query intention model based on hidden Markov characteristic preferably applies to
In gesture interaction environment;
(3) present invention designs a kind of gesture query intention Forecasting Methodology based on hidden Markov model, based on Viterbi
Optimal query intention corresponding to theoretical prediction gesture interaction event, will be complex the problem of, are converted into simple Dynamic Programming meter
Calculation problem, it is theoretical most possibly to produce the query intention of gesture interaction event according to searching with Viterbi, can at utmost it protect
Demonstrate,prove the accuracy of prediction so that the gesture query intention modeling method based on hidden Markov characteristic have it is strong it is theoretical according to
According to.
Brief description of the drawings
Fig. 1 is a kind of signal of the gesture query intention Forecasting Methodology based on hidden Markov model designed by the present invention
Figure;
Fig. 2 is the gesture interaction Operation interface diagram in the embodiment of the present invention;
Fig. 3 is the Operation interface diagram of user gesture search key Machine Learning in the embodiment of the present invention;
Fig. 4 is the Operation interface diagram that user slides Machine Learning search result lists in the embodiment of the present invention;
Fig. 5 is that user amplifies a certain result title in Machine Learning search result lists in the embodiment of the present invention
Operation interface diagram;
Fig. 6 is the gesture query intention model example figure based on hidden Markov characteristic in the embodiment of the present invention.
Embodiment
The embodiment of the present invention is described in further detail with reference to Figure of description.
As shown in figure 1, the present invention devises a kind of gesture query intention Forecasting Methodology based on hidden Markov model, it is real
In the application of border, specifically comprise the following steps:
Step A. initializes gesture path information aggregate Gesture and gesture interaction contextual information set Context
Sky, and enter step B.
Wherein, gesture path information aggregate Gesture is as follows:
Gesture={ Pi,t(xi,t,yi,t,timei,t) | i ∈ [1, I], I ∈ [1,10] }, wherein, Pi,t(xi,t,yi,t,
timei,t) for the positional information of each touch point in gesture operation detection plate, i ∈ [1, I], I ∈ [1,10], I represent touch point
Quantity, i are integer, represent that in all touch points i-th of touch point in gesture operation detection plate, xi,t、yi,tRepresent respectively current
Horizontal and vertical coordinate of i-th of the touch point of moment t in gesture operation detection plate, timei,tRepresent in gesture operation detection plate
Timestamp caused by i-th of touch point in all touch points.
Gesture interaction contextual information set Context is as follows:
Context={ Contexti,t| i ∈ [1, I], I ∈ [1,10] }, wherein,
For corresponding to each touch point on gesture interaction
Context information, Ki,tFor i-th of touch point of current time in gesture operation detection plate corresponding keyword,Represent current
Each touch points of moment t correspond to keyword Ki,tThe r articles Query Result title, R is keyword Ki,tQuery Result sum, r
For integer.
Step B. detections obtain the gesture path information aggregate Gesture and gesture corresponding to active user's gesture to be predicted
Interaction context information aggregate Context, subsequently into step C.
Above-mentioned steps B comprises the following steps:
Current time t is assigned to time by step B1.start, and enter step B2.
Step B2. judges to whether there is touch point in gesture operation detection plate, is then to enter step B3;Otherwise order is current each
It is sky to touch dot position information and gesture interaction contextual information, and updates gesture path information aggregate Gesture and gesture
Interaction context information aggregate Context, subsequently into step B4.
Step B3., which is recorded, respectively touches dot position information and gesture interaction contextual information in current gesture operation detection plate,
Add and update gesture path information aggregate Gesture and gesture interaction contextual information set Context, and return to step
B2。
Whether step B4. judges gesture path information aggregate Gesture and gesture interaction contextual information set Context
It is then return to step B1 for sky;Otherwise current time t is assigned to timeend, then [timestart,timeend] hand in the period
Gesture trace information set Gesture and gesture interaction contextual information set Context, as active user gesture institute to be predicted
Corresponding gesture path information aggregate Gesture and gesture interaction contextual information set Context, subsequently into step C.
Gesture path information aggregate Gestures and gesture interaction of the step C. according to corresponding to active user's gesture to be predicted
Contextual information set Context, the gesture interaction event corresponding to active user's gesture to be predicted is built, subsequently into step
D.Specifically include as follows:
Gesture path information aggregate Gesture and gesture interaction context according to corresponding to active user's gesture to be predicted
Information aggregate Context, define [timestart,timeend] gesture interaction event in the period for Events=gb, dt, v,
K, T }, the as gesture interaction event corresponding to active user's gesture to be predicted, wherein, gb is to be identified based on gesture path information
The gesture gone out, the click class click commonly used in the main search including gesture, slide class swipeup/down and scaling class
zoomin/out;Dt is gesture interaction event Events residence time;Dt is gesture interaction event Events residence time,
Dt=timestart-timeend;V is the speed of gesture operation, and according to gesture path information Gesture, as follows:
The speed v for calculating gesture operation is;K={ Ki,t|i∈[1,I],I∈[1,10],t∈[timestart,
timeend] it is [timestart,timeend] gesture selection in the period keyword set;For [timestart,timeend] hand in the period
The keyword query result head stack of gesture selection.
The gesture interaction historical data of active user in step D. extraction search engines, and combine active user's hand to be predicted
Alternative events corresponding to gesture, it is as follows based on hidden Markov characteristic structure active user's gesture query intention model, subsequently into
Step E.
GQIM=<Intents,Events,π,A,B>
Wherein, Intents is the hidden state set in active user's gesture query intention model, by gesture interaction history
The keyword set that user selects in data is combined into, Intents={ Kn| n ∈ [1, N] }, wherein n be query intention sequence number, Kn
For n-th of keyword of gesture selection, N is the keyword sum of gesture selection, i.e. query intention is total, 1≤n≤N, and n is
Integer.
Events is the Observable state set in active user's gesture query intention model, by gesture interaction historical data
Alternative events corresponding to middle gesture interaction event and active user's gesture to be predicted form, Events={ Eventsh∪EventsM
| h ∈ [1, M-1] }, wherein EventsMThe alternative events corresponding to active user's gesture to be predicted for being built in step E, M is works as
The sequence number of alternative events, Events corresponding to preceding user's gesture to be predictedhFor the hand in active user's gesture interaction historical data
The mutual event sets of power-relation, Eventsh={ gbh,dth,vh,Kh,Th| h ∈ [1, M-1] }, h is gesture in gesture interaction historical data
The sequence number of alternative events, 1≤h≤M-1, and h are integer.
π is the probability matrix of active user's gesture query intention model, corresponding to hidden in hidden Markov model
The matrix of probability containing state, π={ πn| n ∈ [1, N] }, wherein, πn=P (Kn) represent that query intention K occursnIt is initial general
Rate.
A is the query intention transition probability matrix of active user's gesture query intention model, corresponding to hidden Markov mould
Hidden state transition probability matrix in type, A={ ast| s, t ∈ [1, N] }, wherein ast=P (Kt,Ks) represent active user's
Gesture interaction event is KsWhen, the query intention transfer of gesture interaction is K next timetProbability, wherein for arbitrary 1≤s≤
N, meetAnd 0≤ast≤1。
Transition probability matrixs of the B between gesture query intention and gesture interaction event, corresponding to hidden Markov model
In observer state transition probability matrix, B={ bmn| m ∈ [1, M], n ∈ [1, N] }, wherein, bmn=P (Eventsm|Kn) represent
Based on query intention Kn, the gesture interaction event transfer of active user is EventmProbability, wherein for arbitrary 1≤n≤N,
MeetAnd 0≤bmn≤1。
Step E. is calculated according to the gesture interaction historical data of active user and is obtained active user's gesture query intention model
Initiation parameter, subsequently into step F.
Above-mentioned steps E specifically includes as follows:
Step E01. calculates the probability matrix π of active user's gesture query intention model, counts gesture interaction history
Each query intention K in datanThe number of appearance, it is designated as Count (Kn), then calculate query intention KnProbabilityAnd enter step E02.
Step E02. calculates the query intention transition probability matrix A of active user's gesture query intention model, and statistics is sold
User's query intention K in the mutual historical data of power-relationsAnd KtBetween the frequency changed, be designated as cnt (Ks,Kt), then calculate inquiry meaning
Transition probability between figureAnd enter step E03.
Step E03. calculates the transition probability matrix B between query intention and gesture interaction event:
Count each gesture interaction event Events in gesture interaction historical datamThe number of generation, is designated as Count
(Eventsm), gesture interaction event EventsmThe number of generation is more, illustrates gesture interaction event Events occursmCan
Energy property is bigger, therefore utilizes formulaCalculate and gesture interaction thing occurs
Part EventsmProbability.
Statistical query is intended to KnIn gesture interaction event EventsmThe number of middle appearance, it is designated as cnt (Kn,Eventsm), go out
Existing number is more, illustrates query intention KnIn gesture interaction event EventsmUnder the premise of the possibility that occurs it is bigger;Gesture is handed over
During mutually, when user is using class gesture is slided, sliding speed shows user to when the front slide page is interested more slowly,
Query intention KnIn gesture interaction event EventsmUnder the premise of the possibility that occurs it is larger, otherwise it is smaller;When user uses click class
When gesture and scaling class gesture, gesture residence time is longer to show that user is interested in current operation interface, query intention Kn
In gesture interaction event EventsmUnder the premise of the possibility that occurs it is larger, otherwise it is smaller, therefore utilize formulaCalculate query intention Kn
In gesture interaction event EventsmUnder the premise of the conditional probability that occurs, wherein utilizing formulaCalculate using cunning
The weight size of the alternative events of dynamic class gesture, utilizes formulaCalculate the interaction clicked on class and scale class gesture
The weight size of event.
Query intention K is calculated using Bayesian formulanWith gesture interaction event EventsmBetween transition probabilityAnd enter step F.
Step F. is according to the initiation parameter of active user's gesture query intention model, and active user's gesture inquiry meaning
Graph model, it is theoretical based on Viterbi, predict the optimal inquiry meaning corresponding to alternative events corresponding to active user's gesture to be predicted
Figure, submit and perform, realize that gesture query intention is predicted, subsequently into step G.Step F will specifically be based on hidden Markov model
Viterbi theoretical prediction obtain optimal query intention corresponding to active user's gesture to be predicted, be converted into known N number of inquiry
It is intended to Intents={ Kn| n ∈ [1, N] } and M gesture interaction event Events={ Eventsh∪EventsM|h∈[1,M-
1] }, and calculate model parameter λ=(π, A, B), seek alternative events Events corresponding to active user's gesture to be predictedM's
Optimal query intention, and be described as follows:
And 0≤asn≤1
And 0≤bmn≤1
Wherein, PM(Kn) represent that gesture interaction event is EventsMWhen, all query intention Intents={ Kn|n∈[1,
N] } probability that is likely to occur,For value maximum in these probability, now corresponding KnFor the optimal inquiry of prediction
It is intended to;The query intention for ensureing current gesture interaction is KsUnder conditions of, gesture interaction is looked into next time
It is K to ask intention transfernAll possible probability summation be 1;Ensure that current queries are intended to KnFeelings
Under condition, it is Events that gesture interaction event, which occurs,mAll possible probability summation be 1.
In practical application, according to the initiation parameter of active user's gesture query intention model in above-mentioned steps F, and work as
Preceding user gesture query intention model, theoretical based on Viterbi, prediction obtains optimal corresponding to active user's gesture to be predicted
Query intention, specifically include as follows:
Step F01. utilizes formula P1(Kn)=πnb1n, it is Events that 1≤n≤N, which calculates gesture interaction event,1It is all can
The probability P of the query intention of energy1(Kn), and enter step F02.
Step F02. using the obtained all query intentions of step F01 probability P1(Kn), 1≤n≤N, and formulaIt is Events to calculate gesture interaction event2All possible query intention it is general
Rate P2(Kn), 1≤n≤N, subsequently into step F03.
Step F03. utilizes step F02 result, and according to formula
Gesture interaction event Events is calculated successively3,Events4,...,EventsMAll possible query intention probability P3
(Kn)、P4(Kn)、…、PM(Kn), subsequently into step F04.
The P that step F04. selecting steps F03 is obtainedM(Kn) in the optimal objective value of maximum probable value as the problem, this
When query intention K corresponding to the valuenFor the optimal query intention of prediction, the query intention is submitted and performed.
Step G. empties gesture path information aggregate Gesture and gesture interaction contextual information set Context, and returns
Return step B.
The above-mentioned designed gesture query intention Forecasting Methodology based on hidden Markov model is applied among reality,
The detailed process of the gesture query intention modeling method based on hidden Markov characteristic is provided such as Fig. 1.Assuming that user uses Fig. 2 institutes
The gesture operation interface shown carries out the knowledge in terms of gesture search key Machine Learning, starts user using click
Gesture inputs Machine Learning keywords in " input keyword search " frame, gesture operation as shown in Figure 3 occurs
Surface chart, now user, which can use, slides class gesture slip scan the results list and further obtain information needed, such as Fig. 4 occurs
Shown gesture operation surface chart, user, which can also use, scales the heading message that certain search result is amplified/reduced to class gesture,
The details of a certain search result are gone through, gesture operation surface chart as shown in Figure 5 occur.Actual mould in the present embodiment
Intend the gesture interaction process of user, extract the gesture interaction historical data of active user, historical operation process is:User is defeated first
Enter Machine Learning keywords to scan for, part gesture path information is Gesture1={ (" x ":0.1,"y":
0.7125,"t":0),("x":0.1,"y":0.7125,"t":223) }, gesture interaction contextual information is Context1=
{Machine Learning};
Then quick sliding search result list 3 times, wherein a certain second part gesture path information is Gesture2=("
x":0.1,"y":0.6737,"t":0),("x":0.1737,"y":0.6526,"t":52),("x":0.4053,"y":
0.5263,"t":100),("x":0.6053,"y":0.4526,"t":151),("x":0.7842,"y":0.3684,"t":
209),("x":0.9,"y":0.3263,"t":268) }, gesture interaction contextual information is Context2={ Machine
Learning,Online machine learning};
Amplify Online machine learning search result titles, part gesture path information is Gesture5=
{("x1":0.1,"y1":0.7024,"t1":0),("x1":0.1096,"y1":0.7024,"t1":211),("x1":0.9,"
y1":0.2976,"t1":408),("x2":0.9,"y":0.2756,"t2":0),("x2":0.5488,"y2":0.4902,"
t":200),("x2":0.1,"y2":0.7244,"t2":464) }, gesture interaction contextual information is Context5=
{Machine Learning,Online machine learning}。
Thus according to gesture interaction historical data, build the gesture based on hidden Markov characteristic using step D and inquire about meaning
Graph model GQIM=<Intents,Events,π,A,B>, as shown in fig. 6, wherein:
Intents is the hidden state set in gesture query intention model, in the present embodiment, Intents={ Kn|n
=1,2 }, i.e. Intents={ Machine Learning, Online machine learning };
Events is the Observable state set in gesture query intention model, in the present embodiment, Events=
{Eventsm| m ∈ [1,5] }, 4 gesture interaction historical datas and 1 current gesture interaction event are included in the present embodiment, point
It Wei not click on Machine Learning keywords, quick sliding Machine Learning search result lists 3 times and put
Big Online machine learning search result titles, with quick sliding Machine Learning search result lists
Exemplified by, Events now2=Machine Learning, Online machine learning, swipe up, 268,
0.00325};
π is the probability matrix of gesture query intention model, general corresponding to the hidden state in hidden Markov model
Rate matrix, π={ πn| n=1,2 }, wherein πn=P (Kn) represent that query intention K occursnProbability;
A is the query intention transition probability matrix of gesture query intention model, corresponding to hidden in hidden Markov model
Containing state transition probability matrix, A={ ast| s, t=1,2 }, wherein ast=P (Kt,Ks) represent user in current gesture interaction
Query intention is KsWhen, the query intention transfer of gesture interaction is K next timetProbability, it is full wherein for arbitrary 1≤s≤2
FootAnd 0≤ast≤1;
Transition probability matrixs of the B between gesture query intention and gesture interaction event, corresponding to hidden Markov model
In observer state transition probability matrix, B={ bmn| m ∈ [1,5], n=1,2 }, wherein bmn=P (Eventsm|Kn) represent base
In query intention Kn, the transfer of gesture interaction event is EventmProbability, wherein for arbitrary 1≤n≤2, meet
And 0≤bmn≤1。
Initiation parameter λ=(π, A, the B) for obtaining active user's gesture query intention model is calculated according to step E, specifically
Including as follows:
Step E01. calculates the probability matrix π of active user's gesture query intention model, counts gesture interaction history
Each query intention K in datanThe number of appearance, it is designated as Count (K1)=4, Count (K2)=1, then calculates query intention
KnProbabilityProbability matrix π in practical application=[0.8,
0.2], and step E02 is entered.
Step E02. calculates the query intention transition probability matrix A of active user's gesture query intention model, counts gesture
The query intention of the current gesture interaction of user is K in interactive history datasUnder conditions of, the query intention of gesture interaction is next time
KtNumber, be designated as cnt (K1,K1)=3, cnt (K1,K2)=1, utilize
Calculate P (K1,K1)=0.75, P (K1,K2)=0.25, the then transition probability matrix between query intention
And enter step E03;
Step E03. calculates the transition probability matrix B between query intention and gesture interaction event, and statistics gesture interaction is gone through
Gesture interaction event Events in historymThe number of generation, Count (Events are designated as respectively1)=1, Count (Events2)=1,
Count(Events3)=1, Count (Events4)=1, Count (Events5)=1, utilizes formulaCalculate gesture interaction event EventsmThe probability that may occur, point
Wei not P (Events1)=0.2, P (Events2)=0.2, P (Events3)=0.2, P (Events4)=0.2, P (Events5)
=0.2;
Utilize formulaWithThe weight of 5 gesture interaction events is calculated respectively, is remembered respectively
For w1=0.138, α2=0.302, α3=0.305, α4=0.214, w5=0.287;
Statistical query is intended to KnIn gesture interaction event EventsmThe number of middle appearance, cnt (K are designated as respectively1,
Events1)=1, cnt (K2,Events1)=0, cnt (K1,Events2)=2, cnt (K2,Events2)=1, cnt (K1,
Events3)=2, cnt (K2,Events3)=1, cnt (K1,Events4)=2, cnt (K2,Events4)=1, cnt (K1,
Events5)=1, cnt (K2,Events5)=1, utilizes formula
Calculate inquiry
It is intended to KnIn gesture interaction event EventsmUnder the premise of the conditional probability that occurs, be designated as P (K respectively1|Events1)=0.138, P
(K1|Events2)=0.201, P (K1|Events3)=0.203, P (K1|Events4)=0.143, P (K1|Events5)=
0.1435, P (K2|Events1)=0, P (K2|Events2)=0.101, P (K2|Events3)=0.102, P (K2|Events4)
=0.071, P (K2|Events5)=0.1435;
Utilize formula
Calculate query intention KnWith gesture interaction event EventsmBetween transition probability, then query intention and gesture interaction event it
Between transition probability matrixAnd enter step F.
Initiation parameter according to step F according to active user's gesture query intention model, and active user's gesture are looked into
Intent model is ask, it is theoretical based on Viterbi, predict alternative events Events corresponding to active user's gesture to be predicted5Corresponding
Optimal query intention, submit and perform, realize that gesture query intention is predicted, specifically include as follows:
Step F01. utilizes formula P1(Kn)=πnb1n, it is Events that 1≤n≤2 calculate gesture interaction event respectively1Institute
The probability P of possible query intention1(Kn), it is designated as P1(K1)=0.1336, P1(K2)=0, and enter step F02;
The P that step F02. is obtained using step F011(K1) and P1(K2) and formula
It is Events to calculate gesture interaction event respectively2All possible query intention probability P2(Kn), it is designated as:
P2(K1)=max [P1(K1)a11,P1(K2)a21]b21=0.024,
P2(K2)=max [P1(K1)a12,P1(K2)a22]b22=0.008,
It is Events that step F03., which continues to calculate gesture interaction event,3All possible query intention probability P3(Kn),
It is designated as respectively:
P3(K1)=max [P2(K1)a11,P2(K2)a21]b31=0.00441,
P3(K2)=max [P2(K1)a12,P2(K2)a22]b32=0.001464,
It is Events that step F04., which continues to calculate gesture interaction event,4All possible query intention probability P4(Kn),
It is designated as respectively:
P4(K1)=max [P3(K1)a11,P3(K2)a21]b41=0.00057,
P4(K2)=max [P3(K1)a12,P3(K2)a22]b42=0.000187,
It is Events that step F05., which continues to calculate gesture interaction event,5All possible query intention probability P5(Kn),
It is designated as respectively:
P5(K1)=max [P4(K1)a11,P4(K2)a21]b51=0.000074,
P5(K2)=max [P4(K1)a12,P4(K2)a22]b52=0.000049,
Step F06. chooses P according to step F05 result of calculation5(Kn) in the optimal objective of maximum value as the problem
Value, i.e. P5(K1), the optimal query intention now predicted is K1, by query intention K1Submit and perform, then empty current gesture rail
Mark information aggregate Gesture and gesture interaction contextual information set Context, return to step B.
Gesture query intention Forecasting Methodology based on hidden Markov model designed by above-mentioned technical proposal, (1) consider hand
The mutual actual search process of power-relation, during hidden markov process is applied into gesture interaction, analyze query intention between, hand
Transfer relationship between gesture alternative events and query intention, the gesture query intention mould based on hidden Markov characteristic is built with this
Type, preferably express the incidence relation between query intention and gesture interaction event;(2) gone through according to the gesture interaction of active user
History data calculate the initiation parameter of gesture query intention model, have fully demonstrated the individual operation custom of user, and will
Gesture operation feature is closely linked with user's query intention, has used well between query intention and gesture interaction event
Transfer relationship so that the gesture query intention model based on hidden Markov characteristic preferably applies to gesture interaction environment
In;(3) optimal query intention corresponding to Viterbi theoretical prediction gesture interaction event is based on, will be complex the problem of is converted into
Simple Dynamic Programming computational problem, it is theoretical for according to the most possible inquiry meaning for producing gesture interaction event of searching with Viterbi
Figure, it can at utmost ensure the accuracy of prediction so that the gesture query intention modeling method tool based on hidden Markov characteristic
There is strong theoretical foundation.
Embodiments of the present invention are explained in detail above in conjunction with accompanying drawing, but the present invention is not limited to above-mentioned implementation
Mode, can also be on the premise of present inventive concept not be departed from those of ordinary skill in the art's possessed knowledge
Make a variety of changes.
Claims (9)
1. a kind of gesture query intention Forecasting Methodology based on hidden Markov model, it is characterised in that comprise the following steps:
Step A. initialization gesture path information aggregate Gesture and gesture interaction contextual information set Context are sky, and
Into step B;
Step B. detections obtain the gesture path information aggregate Gesture and gesture interaction corresponding to active user's gesture to be predicted
Contextual information set Context, subsequently into step C;
Above and below gesture path information aggregate Gesture and gesture interaction of the step C. according to corresponding to active user's gesture to be predicted
Literary information aggregate Context, the gesture interaction event corresponding to active user's gesture to be predicted is built, subsequently into step D;
The gesture interaction historical data of active user in step D. extraction search engines, and combine active user's gesture pair to be predicted
The alternative events answered, active user's gesture query intention model is built based on hidden Markov characteristic, subsequently into step E;
Step E. is calculated according to the gesture interaction historical data of active user and is obtained the first of active user's gesture query intention model
Beginningization parameter, subsequently into step F;
Step F. is according to the initiation parameter of active user's gesture query intention model, and active user's gesture query intention mould
Type, it is theoretical based on Viterbi, the optimal query intention corresponding to alternative events corresponding to active user's gesture to be predicted is predicted, is carried
Hand over and perform, realize that gesture query intention is predicted.
2. a kind of gesture query intention Forecasting Methodology based on hidden Markov model, its feature exist according to claim 1
In, in addition to step G is as follows, after the execution of step F, into step G;
Step G. empties gesture path information aggregate Gesture and gesture interaction contextual information set Context, and returns to step
Rapid B.
3. a kind of gesture query intention Forecasting Methodology based on hidden Markov model according to claim 1 or claim 2, its feature
It is, gesture path information aggregate Gesture is as follows in the step A:
Gesture={ Pi,t(xi,t,yi,t,timei,t) | i ∈ [1, I], I ∈ [1,10] }, wherein, Pi,t(xi,t,yi,t,
timei,t) for the positional information of each touch point in gesture operation detection plate, i ∈ [1, I], I ∈ [1,10], I represent touch point
Quantity, i are integer, represent that in all touch points i-th of touch point in gesture operation detection plate, xi,t、yi,tRepresent respectively current
Horizontal and vertical coordinate of i-th of the touch point of moment t in gesture operation detection plate, timei,tRepresent in gesture operation detection plate
Timestamp caused by i-th of touch point in all touch points;
Gesture interaction contextual information set Context is as follows:
Context={ Contexti,t| i ∈ [1, I], I ∈ [1,10] }, wherein,
Believe for gesture interaction context corresponding to each touch point
Breath, Ki,tFor i-th of touch point of current time in gesture operation detection plate corresponding keyword,Represent that current time t is each
Touch point corresponds to keyword Ki,tThe r articles Query Result title, R is keyword Ki,tQuery Result sum, r is integer.
4. a kind of gesture query intention Forecasting Methodology based on hidden Markov model, its feature exist according to claim 3
In the step B comprises the following steps:
Current time t is assigned to time by step B1.start, and enter step B2;
Step B2. judges to whether there is touch point in gesture operation detection plate, is then to enter step B3;Otherwise the current each touch of order
Dot position information and gesture interaction contextual information are sky, and update gesture path information aggregate Gesture and gesture interaction
Contextual information set Context, subsequently into step B4;
Step B3., which is recorded, respectively touches dot position information and gesture interaction contextual information in current gesture operation detection plate, add
And gesture path information aggregate Gesture and gesture interaction contextual information set Context are updated, and return to step B2;
Step B4. judge gesture path information aggregate Gesture and gesture interaction contextual information set Context whether be
Sky, it is then return to step B1;Otherwise current time t is assigned to timeend, then [timestart,timeend] gesture in the period
Trace information set Gesture and gesture interaction contextual information set Context, as active user gesture institute to be predicted are right
The gesture path information aggregate Gesture and gesture interaction contextual information set Context answered, subsequently into step C.
5. a kind of gesture query intention Forecasting Methodology based on hidden Markov model, its feature exist according to claim 4
In the step C specifically includes as follows:
Gesture path information aggregate Gesture and gesture interaction contextual information according to corresponding to active user's gesture to be predicted
Set Context, define [timestart,timeend] gesture interaction event in the period for Events=gb, dt, v, K,
T }, the as gesture interaction event corresponding to active user's gesture to be predicted, wherein, gb is identifies based on gesture path information
Gesture;Dt be gesture interaction event Events residence time, dt=timestart-timeend;V is the speed of gesture operation,
And according to gesture path information Gesture, as follows:
<mrow>
<mi>v</mi>
<mo>=</mo>
<mfrac>
<msqrt>
<mrow>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>t</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>t</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>y</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
<mrow>
<mo>|</mo>
<mrow>
<msub>
<mi>time</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>t</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>time</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
</mrow>
<mo>|</mo>
</mrow>
</mfrac>
</mrow>
The speed v for calculating gesture operation is;K={ Ki,t|i∈[1,I],I∈[1,10],t∈[timestart,timeend] be
[timestart,timeend] gesture selection in the period keyword set;
For [timestart,timeend] gesture selection in the period keyword query result head stack.
6. a kind of gesture query intention Forecasting Methodology based on hidden Markov model, its feature exist according to claim 5
In the step D specifically includes as follows:
The gesture interaction historical data of active user in step D. extraction search engines, and combine active user's gesture pair to be predicted
The alternative events answered are as follows based on hidden Markov characteristic structure active user's gesture query intention model:
GQIM=<Intents,Events,π,A,B>
Wherein, Intents is the hidden state set in active user's gesture query intention model, by gesture interaction historical data
The keyword set of middle user's selection is combined into, Intents={ Kn| n ∈ [1, N] }, wherein n be query intention sequence number, KnFor hand
N-th of keyword of gesture selection, N are the keyword sum of gesture selection, i.e. query intention is total, 1≤n≤N, and n is integer;
Events is the Observable state set in active user's gesture query intention model, by hand in gesture interaction historical data
Alternative events corresponding to gesture alternative events and active user's gesture to be predicted form, Events={ Eventsh∪EventsM|h∈
[1, M-1] }, wherein EventsMThe alternative events corresponding to active user's gesture to be predicted for being built in step E, M use to be current
The sequence number of alternative events, Events corresponding to the gesture to be predicted of familyhHanded over for the gesture in active user's gesture interaction historical data
Mutual event sets, Eventsh={ gbh,dth,vh,Kh,Th| h ∈ [1, M-1] }, h is gesture interaction in gesture interaction historical data
The sequence number of event, 1≤h≤M-1, and h are integer;
π is the probability matrix of active user's gesture query intention model, corresponding to the implicit shape in hidden Markov model
State probability matrix, π={ πn| n ∈ [1, N] }, wherein, πn=P (Kn) represent that query intention K occursnProbability;
A is the query intention transition probability matrix of active user's gesture query intention model, corresponding in hidden Markov model
Hidden state transition probability matrix, A={ ast| s, t ∈ [1, N] }, wherein ast=P (Kt,Ks) represent active user gesture
Alternative events are KsWhen, the query intention transfer of gesture interaction is K next timetProbability, it is full wherein for arbitrary 1≤s≤N
FootAnd 0≤ast≤1;
Transition probability matrixs of the B between gesture query intention and gesture interaction event, corresponding in hidden Markov model
Observer state transition probability matrix, B={ bmn| m ∈ [1, M], n ∈ [1, N] }, wherein, bmn=P (Eventsm|Kn) represent to be based on
Query intention Kn, the gesture interaction event transfer of active user is EventmProbability, wherein for arbitrary 1≤n≤N, meetAnd 0≤bmn≤1。
7. a kind of gesture query intention Forecasting Methodology based on hidden Markov model, its feature exist according to claim 6
In the step E specifically includes as follows:
Step E01. calculates the probability matrix π of active user's gesture query intention model, counts gesture interaction historical data
In each query intention KnThe number of appearance, it is designated as Count (Kn), then calculate query intention KnProbabilityAnd enter step E02;
Step E02. calculates the query intention transition probability matrix A of active user's gesture query intention model, counts gesture friendship
User's query intention K in mutual historical datasAnd KtBetween the frequency changed, be designated as cnt (Ks,Kt), then calculate query intention it
Between transition probabilityAnd enter step E03;
Step E03. calculates the transition probability matrix B between query intention and gesture interaction event:
Count each gesture interaction event Events in gesture interaction historical datamThe number of generation, is designated as Count
(Eventsm), and utilize formulaCalculate and gesture interaction event occurs
EventsmProbability;
Statistical query is intended to KnIn gesture interaction event EventsmThe number of middle appearance, it is designated as cnt (Kn,Eventsm), Ran Houli
Use formulaCalculate inquiry
It is intended to KnIn gesture interaction event EventsmUnder the premise of the conditional probability that occurs, wherein utilizing formulaCalculate
Using the weight size for the alternative events for sliding class gesture, formula is utilizedCalculate and click on class and scaling class gesture
Alternative events weight size;
Query intention K is calculated using Bayesian formulanWith gesture interaction event EventsmBetween transition probabilityAnd enter step F.
8. a kind of gesture query intention Forecasting Methodology based on hidden Markov model, its feature exist according to claim 7
In the step F specifically includes as follows:
Viterbi theoretical prediction based on hidden Markov model is obtained into optimal looking into corresponding to active user's gesture to be predicted
Ask and be intended to, be converted into known N number of query intention Intents={ Kn| n ∈ [1, N] } and M gesture interaction event Events=
{Eventsh∪EventsM| h ∈ [1, M-1] }, and model parameter λ=(π, A, the B) calculated, seek active user's hand to be predicted
Alternative events Events corresponding to gestureMOptimal query intention, and be described as follows:
<mrow>
<munder>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
<mrow>
<mn>1</mn>
<mo>&le;</mo>
<mi>n</mi>
<mo>&le;</mo>
<mi>N</mi>
</mrow>
</munder>
<msub>
<mi>P</mi>
<mi>M</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>K</mi>
<mi>n</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
<mrow>
<mn>1</mn>
<mo>&le;</mo>
<mi>s</mi>
<mo>&le;</mo>
<mi>N</mi>
</mrow>
</munder>
<mo>&lsqb;</mo>
<msub>
<mi>P</mi>
<mrow>
<mi>M</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>K</mi>
<mi>s</mi>
</msub>
<mo>)</mo>
</mrow>
<msub>
<mi>a</mi>
<mrow>
<mi>s</mi>
<mi>n</mi>
</mrow>
</msub>
<mo>&rsqb;</mo>
<msub>
<mi>b</mi>
<mrow>
<mi>M</mi>
<mi>n</mi>
</mrow>
</msub>
<mo>,</mo>
<msub>
<mi>P</mi>
<mn>1</mn>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>K</mi>
<mi>n</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>&pi;</mi>
<mi>n</mi>
</msub>
<msub>
<mi>b</mi>
<mrow>
<mn>1</mn>
<mi>n</mi>
</mrow>
</msub>
<mo>,</mo>
<mi>n</mi>
<mo>&Element;</mo>
<mo>&lsqb;</mo>
<mn>1</mn>
<mo>,</mo>
<mi>N</mi>
<mo>&rsqb;</mo>
<mo>,</mo>
<mi>s</mi>
<mo>&Element;</mo>
<mo>&lsqb;</mo>
<mn>1</mn>
<mo>,</mo>
<mi>N</mi>
<mo>&rsqb;</mo>
</mrow>
And 0≤asn≤1
1≤n≤N and 0≤bmn≤1
Wherein, PM(Kn) represent that gesture interaction event is EventsMWhen, all query intention Intents={ Kn| n ∈ [1, N] } can
The probability that can occur,For value maximum in these probability, now corresponding KnFor the optimal query intention of prediction;1≤s≤N ensures that the query intention of current gesture interaction is KsUnder conditions of, the query intention of gesture interaction next time
Shift as KnAll possible probability summation be 1;1≤n≤N ensures that current queries are intended to KnIn the case of,
Generation gesture interaction event is EventsmAll possible probability summation be 1.
9. a kind of gesture query intention Forecasting Methodology based on hidden Markov model, its feature exist according to claim 8
In in the step F, according to the initiation parameter of active user's gesture query intention model, and the inquiry of active user's gesture
Intent model, theoretical based on Viterbi, prediction obtains the optimal query intention corresponding to active user's gesture to be predicted, specific bag
Include as follows:
Step F01. utilizes formula P1(Kn)=πnb1n, it is Events that 1≤n≤N, which calculates gesture interaction event,1It is all possible
The probability P of query intention1(Kn), and enter step F02;
Step F02. using the obtained all query intentions of step F01 probability P1(Kn), 1≤n≤N, and formulaIt is Events to calculate gesture interaction event2All possible query intention it is general
Rate P2(Kn), 1≤n≤N, subsequently into step F03;
Step F03. utilizes step F02 result, and according to formulaCount successively
Calculate gesture interaction event Events3,Events4,...,EventsMAll possible query intention probability P3(Kn)、P4
(Kn)、…、PM(Kn), subsequently into step F04;
The P that step F04. selecting steps F03 is obtainedM(Kn) in the optimal objective value of maximum probable value as the problem, now should
Query intention K corresponding to valuenFor the optimal query intention of prediction, the query intention is submitted and performed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710556277.6A CN107357516B (en) | 2017-07-10 | 2017-07-10 | A kind of gesture query intention prediction technique based on hidden Markov model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710556277.6A CN107357516B (en) | 2017-07-10 | 2017-07-10 | A kind of gesture query intention prediction technique based on hidden Markov model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107357516A true CN107357516A (en) | 2017-11-17 |
CN107357516B CN107357516B (en) | 2019-10-01 |
Family
ID=60292937
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710556277.6A Active CN107357516B (en) | 2017-07-10 | 2017-07-10 | A kind of gesture query intention prediction technique based on hidden Markov model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107357516B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108197191A (en) * | 2017-12-27 | 2018-06-22 | 神思电子技术股份有限公司 | A kind of scene of more wheel dialogues is intended to interrupt method |
CN108284444A (en) * | 2018-01-25 | 2018-07-17 | 南京工业大学 | Multi-mode human action prediction technique based on Tc-ProMps algorithms under man-machine collaboration |
CN108563321A (en) * | 2018-01-02 | 2018-09-21 | 联想(北京)有限公司 | Information processing method and electronic equipment |
CN110032948A (en) * | 2019-03-22 | 2019-07-19 | 中国科学院软件研究所 | A kind of sketch gesture identification method based on interaction timing information |
CN114578959A (en) * | 2021-12-30 | 2022-06-03 | 惠州华阳通用智慧车载系统开发有限公司 | Gesture recognition method and system based on touch pad |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090051648A1 (en) * | 2007-08-20 | 2009-02-26 | Gesturetek, Inc. | Gesture-based mobile interaction |
CN102467905A (en) * | 2010-10-28 | 2012-05-23 | 鸿富锦精密工业(深圳)有限公司 | Gesture recognition appparatus and method |
CN103235812A (en) * | 2013-04-24 | 2013-08-07 | 中国科学院计算技术研究所 | Method and system for identifying multiple query intents |
CN104463250A (en) * | 2014-12-12 | 2015-03-25 | 广东工业大学 | Sign language recognition translation method based on Davinci technology |
CN105930785A (en) * | 2016-04-15 | 2016-09-07 | 丁盛 | Intelligent concealed-type interaction system |
-
2017
- 2017-07-10 CN CN201710556277.6A patent/CN107357516B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090051648A1 (en) * | 2007-08-20 | 2009-02-26 | Gesturetek, Inc. | Gesture-based mobile interaction |
CN102467905A (en) * | 2010-10-28 | 2012-05-23 | 鸿富锦精密工业(深圳)有限公司 | Gesture recognition appparatus and method |
CN103235812A (en) * | 2013-04-24 | 2013-08-07 | 中国科学院计算技术研究所 | Method and system for identifying multiple query intents |
CN104463250A (en) * | 2014-12-12 | 2015-03-25 | 广东工业大学 | Sign language recognition translation method based on Davinci technology |
CN105930785A (en) * | 2016-04-15 | 2016-09-07 | 丁盛 | Intelligent concealed-type interaction system |
Non-Patent Citations (1)
Title |
---|
程春玲 等: "支持手势交互的数据库技术研究挑战和进展", 《南京邮电大学学报(自然科学版)》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108197191A (en) * | 2017-12-27 | 2018-06-22 | 神思电子技术股份有限公司 | A kind of scene of more wheel dialogues is intended to interrupt method |
CN108197191B (en) * | 2017-12-27 | 2018-11-23 | 神思电子技术股份有限公司 | A kind of scene intention interrupt method of more wheel dialogues |
CN108563321A (en) * | 2018-01-02 | 2018-09-21 | 联想(北京)有限公司 | Information processing method and electronic equipment |
CN108284444A (en) * | 2018-01-25 | 2018-07-17 | 南京工业大学 | Multi-mode human action prediction technique based on Tc-ProMps algorithms under man-machine collaboration |
CN108284444B (en) * | 2018-01-25 | 2021-05-11 | 南京工业大学 | Multi-mode human body action prediction method based on Tc-ProMps algorithm under man-machine cooperation |
CN110032948A (en) * | 2019-03-22 | 2019-07-19 | 中国科学院软件研究所 | A kind of sketch gesture identification method based on interaction timing information |
CN110032948B (en) * | 2019-03-22 | 2021-01-12 | 中国科学院软件研究所 | Sketch gesture recognition method based on interaction time sequence information |
CN114578959A (en) * | 2021-12-30 | 2022-06-03 | 惠州华阳通用智慧车载系统开发有限公司 | Gesture recognition method and system based on touch pad |
CN114578959B (en) * | 2021-12-30 | 2024-03-29 | 惠州华阳通用智慧车载系统开发有限公司 | Gesture recognition method and system based on touch pad |
Also Published As
Publication number | Publication date |
---|---|
CN107357516B (en) | 2019-10-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107357516B (en) | A kind of gesture query intention prediction technique based on hidden Markov model | |
WO2020211566A1 (en) | Method and device for making recommendation to user, computing apparatus, and storage medium | |
CN103294815B (en) | Based on key class and there are a search engine device and method of various presentation modes | |
US20220284327A1 (en) | Resource pushing method and apparatus, device, and storage medium | |
CN109165301A (en) | Video cover selection method, device and computer readable storage medium | |
CN105378764A (en) | Interactive concept editing in computer-human interactive learning | |
WO2021135562A1 (en) | Feature validity evaluation method and apparatus, and electronic device and storage medium | |
CN103559252A (en) | Method for recommending scenery spots probably browsed by tourists | |
CN102207960B (en) | Search engine for touch equipment and method | |
CN105320719A (en) | Crowdfunding website project recommendation method based on project tag and graphical relationship | |
CN105549885A (en) | Method and device for recognizing user emotion during screen sliding operation | |
CN106294661A (en) | A kind of extended search method and device | |
Li et al. | PURS: Personalized unexpected recommender system for improving user satisfaction | |
CN105915987A (en) | Implicit interaction method facing smart television set | |
CN111523940B (en) | Deep reinforcement learning-based recommendation method and system with negative feedback | |
CN108595882B (en) | Yinyi planning pipeline BIM design pipeline cost generation system | |
Xu et al. | Quantifying predictability of sequential recommendation via logical constraints | |
CN113110905B (en) | Control method, device, equipment and computer readable storage medium | |
CN111506760B (en) | Depth integration measurement image retrieval method based on difficult perception | |
Li et al. | Modeling touch-based menu selection performance of blind users via reinforcement learning | |
Wang et al. | Intention-aware user modeling for personalized news recommendation | |
CN112948707A (en) | Collaborative filtering recommendation algorithm for optimizing LFM (Linear frequency modulation) through reinforcement learning | |
JPH07129593A (en) | Text selecting device | |
CN107168633A (en) | A kind of gesture interaction query event construction method based on data dependence | |
CN114348019B (en) | Vehicle track prediction method, device, computer equipment and storage medium |
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 | ||
EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20171117 Assignee: NUPT INSTITUTE OF BIG DATA RESEARCH AT YANCHENG Assignor: NANJING University OF POSTS AND TELECOMMUNICATIONS Contract record no.: X2020980007071 Denomination of invention: A gesture query intention prediction method based on Hidden Markov model Granted publication date: 20191001 License type: Common License Record date: 20201026 |
|
EE01 | Entry into force of recordation of patent licensing contract |