CN101473647A - Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising - Google Patents
Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising Download PDFInfo
- Publication number
- CN101473647A CN101473647A CNA200780023082XA CN200780023082A CN101473647A CN 101473647 A CN101473647 A CN 101473647A CN A200780023082X A CNA200780023082X A CN A200780023082XA CN 200780023082 A CN200780023082 A CN 200780023082A CN 101473647 A CN101473647 A CN 101473647A
- Authority
- CN
- China
- Prior art keywords
- user
- signal
- filter
- assets
- information
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims description 81
- 239000002245 particle Substances 0.000 claims abstract description 64
- 238000005259 measurement Methods 0.000 claims abstract description 13
- 239000000203 mixture Substances 0.000 claims description 32
- 238000001914 filtration Methods 0.000 claims description 30
- 230000014509 gene expression Effects 0.000 claims description 10
- 238000004891 communication Methods 0.000 claims description 5
- 230000008676 import Effects 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 230000010365 information processing Effects 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 230000005540 biological transmission Effects 0.000 claims description 2
- 239000000470 constituent Substances 0.000 claims 2
- 230000007704 transition Effects 0.000 abstract description 5
- 230000006870 function Effects 0.000 description 27
- 230000008859 change Effects 0.000 description 24
- 230000008569 process Effects 0.000 description 18
- 230000015572 biosynthetic process Effects 0.000 description 11
- 238000012952 Resampling Methods 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 5
- 238000006243 chemical reaction Methods 0.000 description 5
- 230000004069 differentiation Effects 0.000 description 5
- 230000006399 behavior Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000009795 derivation Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000003542 behavioural effect Effects 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- GOLXNESZZPUPJE-UHFFFAOYSA-N spiromesifen Chemical compound CC1=CC(C)=CC(C)=C1C(C(O1)=O)=C(OC(=O)CC(C)(C)C)C11CCCC1 GOLXNESZZPUPJE-UHFFFAOYSA-N 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000036962 time dependent Effects 0.000 description 2
- 244000287680 Garcinia dulcis Species 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000008602 contraction Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004836 empirical method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000009206 extralife Substances 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000032696 parturition Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000007670 refining Methods 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N7/00—Television systems
- H04N7/16—Analogue secrecy systems; Analogue subscription systems
- H04N7/173—Analogue secrecy systems; Analogue subscription systems with two-way working, e.g. subscriber sending a programme selection signal
- H04N7/17309—Transmission or handling of upstream communications
- H04N7/17318—Direct or substantially direct transmission and handling of requests
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
- H04N21/442—Monitoring of processes or resources, e.g. detecting the failure of a recording device, monitoring the downstream bandwidth, the number of times a movie has been viewed, the storage space available from the internal hard disk
- H04N21/44213—Monitoring of end-user related data
- H04N21/44222—Analytics of user selections, e.g. selection of programs or purchase activity
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
- H04N21/23—Processing of content or additional data; Elementary server operations; Server middleware
- H04N21/234—Processing of video elementary streams, e.g. splicing of video streams or manipulating encoded video stream scene graphs
- H04N21/23424—Processing of video elementary streams, e.g. splicing of video streams or manipulating encoded video stream scene graphs involving splicing one content stream with another content stream, e.g. for inserting or substituting an advertisement
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
- H04N21/23—Processing of content or additional data; Elementary server operations; Server middleware
- H04N21/24—Monitoring of processes or resources, e.g. monitoring of server load, available bandwidth, upstream requests
- H04N21/2407—Monitoring of transmitted content, e.g. distribution time, number of downloads
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
- H04N21/442—Monitoring of processes or resources, e.g. detecting the failure of a recording device, monitoring the downstream bandwidth, the number of times a movie has been viewed, the storage space available from the internal hard disk
- H04N21/44213—Monitoring of end-user related data
- H04N21/44218—Detecting physical presence or behaviour of the user, e.g. using sensors to detect if the user is leaving the room or changes his face expression during a TV program
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/60—Network structure or processes for video distribution between server and client or between remote clients; Control signalling between clients, server and network components; Transmission of management data between server and client, e.g. sending from server to client commands for recording incoming content stream; Communication details between server and client
- H04N21/65—Transmission of management data between client and server
- H04N21/658—Transmission by the client directed to the server
- H04N21/6582—Data stored in the client, e.g. viewing habits, hardware capabilities, credit card number
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/80—Generation or processing of content or additional data by content creator independently of the distribution process; Content per se
- H04N21/81—Monomedia components thereof
- H04N21/812—Monomedia components thereof involving advertisement data
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04H—BROADCAST COMMUNICATION
- H04H60/00—Arrangements for broadcast applications with a direct linking to broadcast information or broadcast space-time; Broadcast-related systems
- H04H60/61—Arrangements for services using the result of monitoring, identification or recognition covered by groups H04H60/29-H04H60/54
- H04H60/63—Arrangements for services using the result of monitoring, identification or recognition covered by groups H04H60/29-H04H60/54 for services of sales
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04H—BROADCAST COMMUNICATION
- H04H60/00—Arrangements for broadcast applications with a direct linking to broadcast information or broadcast space-time; Broadcast-related systems
- H04H60/61—Arrangements for services using the result of monitoring, identification or recognition covered by groups H04H60/29-H04H60/54
- H04H60/64—Arrangements for services using the result of monitoring, identification or recognition covered by groups H04H60/29-H04H60/54 for providing detail information
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Social Psychology (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- Marketing (AREA)
- Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
- Testing, Inspecting, Measuring Of Stereoscopic Televisions And Televisions (AREA)
- Complex Calculations (AREA)
Abstract
Input measurements from a measurement device are processed as a Markov chain whose transitions depend upon the signal. The desired information related to the device can then be obtained by estimating the state of the signal at a time of interest. A nonlinear filter system can be used to provide an estimate of the signal based on the observation model. The nonlinear filter system may involve a nonlinear filter model and an approximation filter for approximating an optimal nonlinear filter solution. The approximation filter may be a particle filter or a discrete state filter for enabling substantially real-time estimates of the signal based on the observation model. In one application, a click stream (208) entered with respect to a digital set top box (200) of a cable television network is analyzed to determine information regarding users (205) of the digital set top box (206) so that ads (204) can be targeted to the users (205).
Description
The cross reference of related application
The application require to the sequence number that on May 2nd, 2006 proposed be 60/746,244, the priority of U.S. Provisional Application under U.S.C.119 35 chapters of " METHODAND APPARATUS TO PERFORM REAL-TIME ESTIMATION ANDCOMMERCIAL SELECTION SUITABLE FOR TARGRETEDADVERTIDSING " by name.The application also require to the sequence number that proposed on January 12nd, 2006 be 11/331,853, the priority of the U.S. Patent application of " TARGETEDIMPRESSION MODEL FOR BROADCAST NETWORK ASSET DELIVERY " by name.The spy incorporates the content of these two applications herein into, to carry out complete elaboration.
Technical field
The present invention relates to the innovation of non-linear filtration, wherein observation process is modeled as Markov chain (Markov chain), and utilize one embodiment of the present of invention to estimate that user in the communication network is equipped with user's composition of equipment, for example, TV viewer's number and demographics in top box of digital machine (DSTB) environment.In addition, the present invention also provides the method for determining which assets collection (for example, commercial advertisement) is inserted the available network bandwidth based on the sampling that the optimum condition of current network operating position (for example, group of viewers) is estimated, best.
Background technology
Generally, in the past 50 in the period of, submit to the mode of commercial advertisement (commercial) not have too big variation to the televiewer.Distribution company and advertiser attempt using historical Nielson
TMViewership information judges what their target audience watching.These data provide the estimation to the number of the family of the specific episodes of watching TV programme at special time, and demographics detailed catalogue (dividing based on age, sex, income and race usually).In case the user shows that they are watching TV, then use ' personnel's measuring appliance ' (people meter) data to collect such data (another kind of audience rating data) widely, personnel's measuring appliance data monitor automatically which program is just viewed.These samples are quite little-and current, only use about 8000 families to estimate all-american whole group of viewers.Along with increasing of available television channels number, and group of viewers from broadcasting to the transfer of cable TV, the increase of the number of television set in the single in addition family more and more is difficult to accurately estimate according to little sample like this actual spectators of TV programme.Therefore, less shared cable channel can not correctly be estimated their group of viewers, thereby advertiser can not correctly catch favourable target demographic statistics situation.
Because the growing demand to the digital cable TV supply causes constantly popularizing of DSTB, can obtain the more accurate information of each family in theory.That is, set-top box can obtain about watching which channel, watching this channel how long to wait information.So a large amount of information if correctly handled, can provide understanding in depth family's behavior.Yet none can provide directly that advertiser's desired information type-in particular moment, what kind of person is watched these information.Advertiser wishes with the highest accuracy their advertisement to be shown to their target audience, to reduce the marketing cost and to improve its effect.And they wish to avoid playing the passiveness propaganda cost that commercial advertisement brings to unsuitable spectators.Provide the key of ability of its investment of maximization to be to advertiser, change the mode that group of viewers is counted, this " [has changed] all types of comparison values and whole demographic field " potentially (Gertner, J; Our Rating, Ourselves; The New York Times; On April 10th, 2005).
The various systems that are used to identify current beholder or their demographics have been proposed or have realized.Some such system has formed interference, and they require the user to import identity or demographic information clearly.The behavior overview of basis from the information extraction beholder of each provenance attempted by other system.Yet these systems all have following one or more shortcoming usually: 1) they attentiveness be placed on current who in the family rather than who watch; 2) they only can provide the coarse information of family-related subclass; 3) they require the user to participate in, and this is that the certain user is undesirable, and may cause mistake; 4) they can not be provided for judging the framework that when has a plurality of beholders or be used for explication demographics under a plurality of beholder's situations; 5) their prerequisite is, they are static fully, and can not correctly dispose the family's composition (composition) and the demographics of variation; With and/or 6) they have used time good technology, require comprehensively training, require too much resource, thereby limited actual application.
Summary of the invention
The present invention relates to analyzing, to obtain the information of relevant institute attention signal from the observed result that measuring equipment obtained.In an application, (for example the present invention relates to analysis user at the user of the outfit equipment of communication network input, the user who imports at the top box of digital machine (DSTB) of cable TV network imports click steam), to determine that relevant user is equipped with the user's of equipment information (for example, spectators' sorting parameter of one or more user).Some aspect of the present invention relates to handling from ruined, distortion and/or the incomplete data observation result that measuring equipment received, inferring the information of relevant signal, and be provided for obtaining the basic filtration system of estimating in real time to the state of pay close attention to signal constantly.Particularly, such filtration system can provide approximate to the reality of Optimal Nonlinear filtering scheme according to some restriction to the admissible state of inferring according to observing environment or its combination.
According to an aspect of the present invention, provide the method and apparatus (" system ") that is used at produce observation model from the data of being analyzed that equipment obtained (being measurement result).Particularly, this system is modeled as Markov chain to the measurement result of being imported, and its conversion depends on signal.Observation model can be considered exogenous information, or the information of the outside of the measurement result of being imported (although needn't be independent of).In an implementation, the measurement result of being imported has reflected the click steam of DSTB.Click steam can reflect and control relevant channel selection incident and/or other input such as volume.In the case, observation model also can relate to the programme information relevant with selected channel (for example, from download such as the network platform of head end).So, click steam information is handled as Markov chain.
Then, can pay close attention to the state of the signal at place constantly, obtain device-dependent expectation information by estimating.In the example of the click steam of analyzing DSTB, signal can representative of consumer composition (demographics that relates to one or more user and/or be associated) and influence such as channel change the extra factor of the click steam of form, below will be discussed this in more detail.In case signal is estimated, then can determine in the past, now or following constantly the state of signal, for example, in order to the user who uses together with the assets orientation system to be provided composition information.
According to a further aspect in the invention, a kind of system generates the real-time substantially estimation of signal condition according to observation model.Therefore, can use the non-linear filtration system so that the estimation of signal to be provided based on observation model.The non-linear filtration system can comprise non-linear filtration model and the approximate filter that is used near optimal non-linear filtration scheme.For example, approximate filter can comprise basic particle filter or the discrete state filter of estimating in real time that is used for realizing according to observation model signal.In the DSTB example, the non-linear filtration system allows the user's composition that comprises more than one beholder is identified, and is adapted to potential spectators and changes, for example, for potential spectators, previous unknown personnel's increase or leaving away of user before.
According to a further aspect in the invention, a kind of system uses the information by filter application is obtained in observation model, to obtain the institute's concern information at this observation model.Particularly, can obtain in the past according to the estimated state of paying close attention to signal constantly, now or following constantly information.Under the situation of the use of analyzing DSTB, can determine concrete identity and/or the demographics of locating one or more user of DSTB constantly according to signal condition.For example, can use this information " ballot ", the suitable assets of i.e. predetermined commercial advertisement of sign or program location, with from being used for submitting to the assets option of (delivery) to select assets at DSTB, and/or at one or more user's judgement that receives the assets of submitting to or report the grade of fit (goodness of fit) of these assets.
The various aspects of the invention described above can be provided by any suitable combination.And, also can realize above mentioned any or all aspects together with directed assets submission system.
In one embodiment of the invention, provide a kind of system that is used for the user that assets are oriented to communication network (as wired TV network) is equipped with the user of equipment.This system relates to: at the input that the user is equipped with equipment, produce (develop) observation model according to one or more user; Incorporate described observation model into reflect the signal of at least one user's composition that described user is equipped with one or more user of equipment at the time; The pay close attention to state that user's composition constantly is defined as signal; And when the equipment that is equipped with at the user carries out orientation to assets, use determined user's composition.In this way, can be equipped with input (as click steam) the application filtration theory of equipment at the user, to obtain indicating the signal of user's composition.
Can be modeled as Markov chain to input.And the model permission of signal is shown user's component list and comprises two or more users.Therefore, can identify a plurality of users' situation, to be used for directed assets and/or to estimate audience scale better and composition (for example, to improve appraisal and the charge that assets are submitted to).In addition, preferably, signal model allows the change of expression user composition, for example removes personnel to spectators' user interpolation or from spectators user.
Can define the non-linear filtration device, with according to the observation model picked up signal.Thus, signal can be represented the user's composition at the family of time, and can be defined as the function of the state of signal constantly of paying close attention to spectators' sorting parameter (for example, one or more active user's demographics).For the estimation of Optimal Nonlinear being filtered the practicality of separating is provided, can be provided for the approximate filter of the operation of approximate non-linear filtration device.For example, approximate filter can comprise particle filter or discrete space filter.And approximate filter can be realized at least one constraint at one or more signal component.Thus, constraint can be used for the one-component of signal is used as the invariant of the time period that changes for the permission second component.In addition, constraint also can be used at least one state of first component is used as irrational, perhaps is used as a certain combination of the state of unlike signal component irrational.For example, under the situation of the click steam of DSTB, at least one individual's existence is represented in the appearance of click event.Therefore, when click event, only can permit user's composition corresponding at least one individual's existence.Other allowance or nonlicet combination can be associated income with the place.Can realize constraint explicitly with the approximate filter of the confined space.For example, can reorientate the value of following on the unreasonable unit, for example, it be shifted to adjacent reasonable unit in proportion.In this way, approximate filter can converge on rationally soon and separate, and need not unsuitable processing resource.Be used at least one potential state that is calculated is defined as under irrational situation in constraint, approximate filter can redistribute one or more counting associated therewith.
In addition, approximate filter also can be operated and be used to forbid converging on unreasonable state.So, approximate filter be designed to avoid converging on infeasible in logic or can not (click event when no user exists) or rule think user's composition of the DSTB of irrational (for the nonlicet income range in given place).In an implementation, add seed counting (seed count) by reasonable unit, to forbid realizing this point at the convergence of unreasonable unit to the discrete space filter.
Preferably, determine user's composition information at the top box of digital machine place.That is, user profile is calculated on top box of digital machine, and be used to choose, portfolio selection and/or report.Replacedly, can as head end, wherein can determine user's composition information to the clickstream data independently platform that leads, for example, wherein message to send bandwidth be sufficient and DSTB to handle resource be limited.As another replacement, can be transferred to head end or other platform content to user's composition information (for example, opposite) with the assets vote information to be used to select insert.
Determined user's composition information can be used by the assets orientation system.For example, can offer the network platform to information, as be used for assets are inserted the head end of the content stream of network.Thus, this platform can be used to the input from a plurality of DSTB, selects to be inserted into the assets in the available network bandwidth.Thus, can utilize extraneous information, the information that the every user who submits to as the reflection assets is worth.This platform can be an observation model with the information processing that is equipped with equipment from a plurality of users, and uses the filter that suitably disposes at this observation model to estimate that the network spectators' constantly that pay close attention to integral body constitutes.
According to a further aspect in the invention, stochastic control theory is applied to the problem that directed assets are submitted to, selects as the person's of dynamically watching classification and/or television advertising.Traditionally, stochastic control theory is applied to directly to calculate and only can be according to the situation that may come signal calculated or function for noise or incomplete observed result.In the present circumstance, can handle measurement result, to estimate to can be used to determine the signal of state information according to stochastic control theory from measuring equipment.For example, measurement result, for example from the click steam of remote controller, as the noise observed result from input equipment, and use STOCHASTIC CONTROL to handle, with the signal of the behavioral formation estimating represent family or the spectators that watch and be associated with the clauses and subclauses of importing.STOCHASTIC CONTROL allows this signal is followed the trail of, and makes the state of concrete signal constantly reflect that the beholder of this moment constitutes and form.For example, by the spectators' that watch the sorting parameter and the orientation parameter of advertisement available are mated the targeted ads that can use this Information Selection to be used to submit to.
Description of drawings
In order more fully to understand the present invention and more advantage thereof, incite somebody to action now in conjunction with the accompanying drawings with reference to following detailed description, in the accompanying drawing:
Fig. 1 is the schematic diagram according to orientation of the present invention (targeted) advertisement distributing system;
Fig. 2 has illustrated according to REST structure of the present invention;
Fig. 3 has illustrated the cellular construction according to the unit of filter of the present invention;
Fig. 4 is the flow chart that has illustrated according to filter evolution process of the present invention; And
Fig. 5 has illustrated the block diagram that is used for the process of modeling event according to of the present invention.
Embodiment
In the following description, just submit the situation of (for example targeted ads issue) system to, set forth the present invention at the directed assets of cable TV network.Yet, should be appreciated that various aspects of the present invention are not limited to this situation, and scope of the present invention is limited by following given claim.
Various targeted ads delivery systems have been proposed or have realized at cable TV network.Therefore these systems, can be mated commercial advertisement and spectators usually based on the understanding to current spectators' composition, with the value of maximization commercial advertisement.To recognize that various such systems all can have benefited from being used to identify the structure of the present invention and the function of current beholder's sorting parameter (for example, demographics).Thereby, although following only illustrative introduced concrete directed assets submission system, yet should be appreciated that the present invention can be used more widely.
The sequence number that proposed on January 12nd, 1 is to have described a kind of directed assets submission system that can use together with the present invention in 11/331,853 the U.S. Patent application.For the purpose of brief, will not repeat the full details of this system herein.Generally speaking, in this system,, a plurality of assets options are provided at the some preset time on the given program channel.Although thus can directed various types of assets, set forth in describing as described, yet targeted ads issue (for example, the orientation of commercial advertisement) only be illustrative application herein, and used as simply reference easily.So, can provide a plurality of assets (for example, advertisement) channel of advertisement options to support the given program channel by one or more location advertising to commercial insert.DSTB is used for that not visible ground (from beholder's angle) is converted into the correspondent advertisement channel during commercial insert, so that the targeted ads issue to be provided to one or more current beholder.
Beholder's marking structure of the present invention and function can be used for described directed assets submission system by variety of way.In described system, before commercial insert, send the advertising listing that comprises orientation parameter to DSTB.DSTB determines sorting parameter at one or more current beholders, the target component of each advertisement in these sorting parameters and the tabulation mated, and to " ballot " of head end (Head End) transmission at one or more advertisement.Head end compiles the ballot from a plurality of DSTB, and the optimal advertising group is come together in available bandwidth (it can comprise program channel and a plurality of advertisement channel).When commercial insert, DSTB selects to run through " path " of this group, to submit correspondent advertisement to.Then, DSTB can report and submit which advertisement to, and actual spectators and the target component of indication has the grade of fit information of how to mate.
Can in described directed assets submission system, directly realize the present invention.That is, use technology described herein, can determine spectators' sorting parameter of one or more current beholder at the DSTB place.As described in described in the pending application, this information can be used for ballot, advertisement selection and/or grade of fit and determines.Replacedly, below described a kind of filtration theory based on head end advertisement selection system, it substitutes as a kind of of described voting process.Substitute as another, can offer head end or another network platform to click steam information, wherein can calculate spectators' sorting parameter.So, spectators' sorting parameter, advertisement selection and other functional can change, and can be distributed between DSTB, head end or other platform by variety of way.
Following chapters and sections are divided into several sections.In first, some background discussion are provided to relevant non-linear filtration theory.In second portion, architecture and model class have been discussed.
1.1 non-linear filtration
In order suitably to solve targeted ads group of viewers (potential with current) problem, can be conceived to filtration field (field of filtering) best on the mathematics.
1.1.1 traditional non-linear filtration general introduction
Non-linear filtration is according to ruined, distortion or the incomplete data observation result of signal, handles past of a certain non-linear stochastic dynamic process (being commonly referred to ' signal '), now and/or the best estimate of to-be in real time.Generally speaking, X
tBe considered as being defined in a certain probability space (Ω,
, the P) Markov process on, it is separated for a certain Martigale problem.Observation is usually at discrete moment t
mCarry out, and depend on the use sensor function
The signal of certain random fashion.In fact, traditional theory and method are set up around such observation, wherein, measurement result is the distortion of signal (because nonlinear function h), ruined (because noise V), incomplete (because h is only to possible dependence of the part of the state of signal) sample.Given until available observed result of current time, optimum filter provides the condition of the state of signal to distribute:
This filter not only can provide best estimate to the current state of signal, can also provide best estimate to the entire path of previous and to-be and signal:
Under some linear case, can obtain effective best recurrence formula.Suppose that signal follows the Ito random difference equation
, wherein A is linear, B is a constant.And observation function is
Form, wherein
It is independent Gaussian random variable.This formula is called as Kalman (Kalman) filter.Although kalman filter is very effective in the process of its estimation, because the strictness of the handled signal of observation is described, there is intrinsic restriction in its use in application.In the dynamic characteristic of signal is non-linear or observed result has under the situation of non-interpolation and/or relevant noise, and kalman filter provides time good estimation.Therefore, in order under more general situation, to provide best estimate, need seek other method.
Although there has been many decades in the equation that Optimal Nonlinear is estimated, just found the purposes that they are few as of late.Can not realize best equation on a computer, it requires to use unlimited memory and computational resource.Yet, in the past ten years,, created some approximate expression of these optimum filtration equations for overcoming this problem.These approximate expressions are generally asymptotic the best, this means owing to used the increasing resource of quantity in their computational process, so they converge on optimum solution.Two kinds of most popular such methods are particle (particle) method and discrete space method.
1.1.2 particle filter
The particulate filter method relates to establishment and is represented as
The separate copies (being called ' particle ') of signal, N wherein
tThe number of the particle that uses for moment t place.According to the rule at random of signal, these particles develop in time.Give weighted value to each particle then
Thereby, form observation sequence (Y with pooling information effectively
1..., Y
m).This can be done in such a manner: the weight after m the observation depends on observation Y the m time for the weight after m-1 the observation multiply by
mThe factor.Yet it is extremely inhomogeneous that these weights always become, and this means that many particles (those have the particle of low relatively weight) become inessential, and except the cycle of consumption calculations machine, do not do.More suitably be, by removing these particles and calculating being reduced to the particle that number constantly reduces, particle is resampled, this means particle position and weight are all adjusted, to guarantee that all particles all contribute to the condition distribution by meaningful ways and calculate, guarantee simultaneously statistic bias not to be introduced this adjustment.Early stage particle method tends to carry out too much resampling, thereby the too much noise of resampling is introduced in the system of particle, deterioration estimation.Suppose after resampling, the weight table of the particle after m the observation to be shown
Then particle filter is to being approximately that the condition of optimum filter distributes:
Work as N
tDuring → ∞, particulate filter is estimated to obtain Optimal Nonlinear and is filtered estimation.
U.S. Patent application 2002/0198681 at " Flexible Efficient Branching Particle Tracking Algorithms " by name (is incorporated it herein into, with for referencial use) in, in each time step, duplicate, destroy to probabilistic particle or make particle remain unchanged.Based on the weight (W that is only calculated at the current time step-length
m(ξ
j)), revise particle according to following routine:
1. if
Then create particle ξ
j Individual copy, and establishment has probability
Additional copies.
3. carry out the zero deflection control algolithm, the number of particle is back to the existing amount before of resampling.
The scheme of this ' careful ' is a kind of improvement to the previously known particle filter, this application also comprises the system of described algorithm of effective realization and historical variable, wherein, people wish to estimate the path of the X before time t, and are not only its current state.By name " Selectively ResamplingParticle Filter ", sequence number is 7,058,550 United States Patent (USP) has been released still less performance degradation of a kind of introducing in (it being incorporated into herein, with for referencial use), and has improved the further improvement of computational efficiency.This method is carried out in couples as follows and is resampled:
1. for the highest weighting particle j and lowest weighted particle i, when
The time, then:
2. probability of use
The state of particle i to j is set, and probability of use
The state of particle j to i is set.
3. the weight of particle i and j is rearranged into
In this method, introduced Control Parameter p, with the performed amount of resampling of suitable adjusting.Described in application 2005/0049830, the dynamic change in time of this value is with the current state that adapts to filter and concrete application.This application also comprises the effective system of storing on computers and calculating the desired value of this algorithm.
1.1.3 discrete space filter
When the state space of signal is on the limited dimensional space of a certain bounded, then can use discrete space and amplitude approximate.Sequence number is 7,188,048, be entitled as in the United States Patent (USP) (it being incorporated into herein, with for referencial use) of " Refining Stochastic Grid Filter " and describe a kind of discrete space filter (REST filter) in detail.Under this form, state space D is divided into discrete unit η
CFor example, this space can be d dimension Euler space or certain Counting Measures space.The amplitude that each unit is called the discretization of ' particle counting ' (is expressed as
), its condition that is used to form the discrete space filter distributes:
According to the operator and the handled observation data of signal, change the particle counting of each state cell.When the number of unit became infinity, then the estimation of REST filter converged on optimum filter.For clear, this application is directly considered the equation of filtration of discretization, rather than to signal discreteization and design attainable equation of filtration at the discretization signal.
In application 2005/0071123, this invention is according to the real-time processing of observation, and the y-bend index tree that utilizes dynamic interleaving has with tissue so that the unit of the data structure estimated of the condition of recursive calculation filter effectively.Although this structure can be complied with some and use, yet under the less situation of state space dimension complexity, the expense of this data structure has reduced the practicality of this method.
1.2 STOCHASTIC CONTROL
Select problem in order to solve directed commercial advertisement up hill and dale, should be conceived to mathematical best of STOCHASTIC CONTROL.
In concept, can invent the approximate on computers particle method that solves the direct discretization method of STOCHASTIC CONTROL problem.Yet these work are not finished as yet, are perhaps generally admitted as yet at least.And various implementation methods are found the solution the discretization problem then usually whole problem discretization.
2.1 targeted ads delivery system architecture
Fig. 1 has described whole targeted ads delivery system.This system is made of head end 100, head end 100 one or more top box of digital machines 200 of control.DSTB 200 attempts to use DSTB filter 202 to estimate to comprise the conditional probability of the state of the potential beholder in one or more the current kinsfolk's who watches TV the family 205.DSTB filter 202 uses a pair of model 201 of describing signal (family) and observed result (clickstream data 206).Carry out initialization by the 302 pairs of DSTB filters of downloading from head end 100 202 that are provided with.In order to estimate the state of family, DSTB filter 202 also can use programme information 207 (its can for current or recently in the past or following), programme information 207 can obtain from the memory 208 of programme information.
Condition is distributed DSTB filter 202 or the estimation of therefrom derivation is sent to commercial advertisement selection algorithm 203, commercial advertisement selection algorithm 203 judges according to the output of filter, the commercial advertisement 301 of being downloaded and any regular 302 (its which type of commercial advertisement of control under the situation that given beholder estimates is permitted) which commercial advertisement 204 is shown to current beholder then.Record and storage are shown to beholder's commercial advertisement.
Can estimate and commercial advertisement is submitted to statistics and out of Memory to carry out grab sample 303 and compiled 304 DSTB filter 202, to provide information to head end 100.Head end filter 102 uses these information, and head end filter 102 is gathered at relative DSTB and calculated the condition distribution that (obeying its available resources) compiles potential and actual group of viewers.Head end filter 101 uses and compiles family and DSTB feedback model 101, and its estimation is provided.Head end commercial advertisement selective system 103 uses these to estimate, which commercial advertisement judgement should send to DSTB set that head end 100 is controlled.Commercial advertisement selective system 103 has also been considered the relevant any obtainable market information 105 of economic conditions with current commercial advertisement contract and those contracts.Subsequently the selected commercial advertisement 301 that obtains is downloaded to DSTB100.The commercial advertisement that selection is used to download affects rank and is provided with 104, and rank is provided with 104 constraints that the personal broadcasting particular commercial of subtend particular type is provided.
Below two chapters and sections some details of this system has been described.
2.2 family's signal and observation model are described
In these chapters and sections, provided the example of the possible embodiment of the description of resultant signal and observation model and this model.
2.2.1 signal model is described
Generally speaking, be the signal modeling of family one group of individual and one family state.In a preferred embodiment, this family's representative may be watched the people of the concrete TV that uses DSTB potentially.Each individual at given time t place (is expressed as X
i) have state from state space s ∈ S, wherein, the characteristic set determined in the family everyone is wished in S representative.For example, in one embodiment, may wish each individual age, sex, the situation of taking in and watch are classified.Can be thought of as real number value to age and income, perhaps be thought of as discrete range.In this example, will be defined as state space:
S={0-12,12-18,18-24,24-38,38+} * { man, woman } * { 0-$50,000 , $50,000+} * { be, not }
So kinsfolk's state space is
Wherein, k represents individual's number, S
0Expression nobody's independent state.The kinsfolk
Member with time dependent random number, wherein n
tNumber of members during for moment t.Because the inferior ordered pair problem of the member in this group is unimportant, so we use member's experience to estimate
Represent family.
Family's form (regime) has been described current the watching ' tendency ' of the family of generation that may the appreciable impact clickstream data.The current form r of family
tIt is value from state space R.In one embodiment of the invention, form can be by forming such as the value of ' normally ', ' channel upset ', ' situation inspection ' and ' preference surfing '.
So complete signal is by kinsfolk and morphosis:
X
t=(X
t,R
t)。
The state of signal develops in time by the audience ratings function lambda, the variation of audience ratings function lambda probabilistic ground control signal state.So at a time the probability of the state variation after the t, from state i to j is:
Existence is at the independently audience ratings function of each individual's differentiation, kinsfolk itself and family's form.In one embodiment of the invention, the experience that can only depend on given individual, signal at the audience ratings function of individual i estimate, current time and some external environment condition variable
The number n of individual in the family
tChange in time according to giving birth to dead speed.Living and dead speed not only represents new survivor's birth and existing person's death-they can represent the incident that causes one or more people to enter and withdraw from family.Current state according to a guy of institute in the family is calculated these speed.For example, in one embodiment of the invention, can calculate and describe the rate function that the bachelor has the room-mate that enters family or spouse's possibility.
In one embodiment of the invention, can be formulated as math equation to these rate functions with parameter, wherein said parameter is by estimated probability with because of available demographics, macroeconomy and watch the desired value of the state variation that behavioral data produces to be complementary, and is determined to experience.In another embodiment, the age can develop on certainty ground in continuous state space [0,120].
2.2.2 observation model is described
Generally speaking, observation model is made of the mutual click steam information that generates of one or more people and DSTB.In a preferred embodiment of the invention, in observation model, only represented the channel change information current and past.Given whole M channel is at moment t
kHas in the past the Y that watches in B the discrete time step-length
k=(y
k..., y
K-B+1) channel of individual channel changes formation.In a preferred embodiment of the invention, only write down channel and change the moment of taking place and that channel that changes to it, to reduce expense.
In a more general case, watch formation to comprise this current and past channel and such as contents such as volume (volume) history.In these cases, watch formation to degenerate to channel and change formation.
Then, according to the state of signal and some downloadable content D
t(be expressed as P
I → j(D
t, X
t)), determine when moment t, to change to the probability of watching formation of state j from state i.In a preferred embodiment, except that other content, this downloadable content also comprises the programme information of the property class description of the current available program of some refinement, for example, for each program, described programme information is " action movie " or " serial " for this program, and the time started of the duration of this program, this program, plays the channel of this program etc.
Lacking under the situation of specific form, creating the empirical method of calculating the Markov chain transition probability.These probability depend on all members' of family current state and available programs.Use observes watches the behavior and the Waradarajian law of large numbers, verifies this method.Suppose that P is discrete probability measure, and give Ω={ ω probability
1..., ω
k, then we have N separate copies of the test of selecting element.Then, the law of large numbers provides
Wherein, ω
iBe i the output at random that draws element according to Ω.
In one embodiment of the invention, this method is 1 channel formation (that is Y, at size
k=y
k) calculating probability.Can be by determining other probability of transformation classes of program, find out then to be converted into the probability of concrete channel in this classification, at first calculating observation probability, i.e. two probability of watching conversion between the formation on next discrete steps.First step-length is calculated the relative scale that changes the classification variation that causes because of program on channel change and/or the same channel by offline mode usually.In order to carry out this calculating, all possible member condition X
tMapping be mapped to discrete state space ∏, make for certain π
t∈ ∏ is for all possible X
t, f (X
t)=π
tSuppose to exist the classification C ∈ { C of fixed number
1, C
2..., C
k.In addition, suppose to exist N
vIndividual beholder's record, each beholder writes down and represents constant time period Δ t, and wherein, each tlv triple is watched record V (k)=(π, C
1, C
2), k=1,2 ... N
vComprise the discretization state of relevant family (π) and relevant time period to beginning (C
1) and finish (C
2) time the information of classification.So, at each π ∈ ∏ and
C
J∈ C, we calculate:
When best estimate system during just at real time execution, given current available programs by calculating the probability that classification changes, at first calculates now on step-length preset time from C
ITo C
JThe probability of classification conversion:
Wherein, A exchanges through all possible effective classification according to available current program.Then, by following formula this probability is converted to required channel switch probability:
Wherein, n
t(J) for have the number of the channel of all programs that fall into classification J in the end of current time step-length.
Employed replacement probability measure is, calculates ' pouplarity ' of channel on each discrete time step-length, rather than the conversion between the channel.Said method can be used for providing this form by at given classification transition probability being sued for peace simply:
Equally, by using multiplication example then, this probability is converted to required channel switch probability:
Wherein, n
t(J) for have the number of the channel of all programs that fall into classification J in the end of current time step-length.
In one embodiment of the invention, under the situation of given fine granularity (level of granularity), some or all classification will be program itself.In other example, preferably, have wide classification, to reduce the number of the probability that need be stored.
2.3 use the best estimate of Markov chain observation
In above traditional filtration theory of summing up, according to following formula, resulting distortion, ruined, the incomplete measurement that is observed signal:
Wherein, t
kBe the observation time of the k time observation,
Be a certain driving noise process, perhaps a certain continuous time variable.Yet for the DSTB model described in the former chapters and sections that are right after, Y depends on the discrete time Markov chain of signal for its transition probability.In the case, new state Y
kMay depend on original state, cause above-mentioned standard theory invalid.In these chapters and sections,, new analog theory and system have been proposed in order to solve the problem that wherein is observed Markov chain.The noticeable general character of this system is, can only allow Markov chain observation to be transformed into the subclass of all states, and wherein this subclass depends on its current state of living in.In targeted ads issue was used, this was useful characteristic, because after observation and inserting some new data, watched most of past data of formation to remain on and watched in the formation.For easy to understand, will use at the targeted ads issue and describe this point, although it obviously is applicable to ordinary circumstance.
Suppose that we have markov signal X
t, it has generator
With initial distribution v.For the purpose of accurately, be this signal definition to satisfy
Unique D of-halter strap (Martingale) random process problem
E[0, ∞) process:
P(X
0,ε,·)=v(·),
And
Wish according to depending on X
tAnd a certain external source information D
t1,2 ..., X is estimated in the discrete time Markov chain observation of M} value
tCondition distribute.For the purpose of clear and definite, suppose
For being independent of the independent random variable series of signal and observation, make for i=1,2 ..., M and k ∈ Z,
Observation y
kOccur in t constantly
k, have the incident that can get finite state space 1 ..., M} and Y
k=(y
k, y
K-1... y
K-B+1), wherein
When having at moment t from the even transition probability p of state i steering state j
I → j(D
t, X
t) 1 ..., M}
BIn value between change.Herein, D
tAnd X
tCurrent state and possibility state change signal condition constantly for relevant outer source information.
For the ease of expression, we define
And establish
V
k=υ
k, υ
K-1..., υ
K-B+1)
T, for k=1,2 ...
Wherein,
In order to represent conveniently definition Z
0=1.So some mathematical computations shows:
For t
j≤ T, wherein f:E → R and
P(A)=E[1
AZ(T)]
Order
Respectively for g=1 and g=f, wherein,
For the enough abundant class of function f: E → R, for
We only need an equation.
Wherein,
And n
s=max{k:t
k≤ s}.
2.4 filter approximate
In order in real-time computer system, to use above-mentioned derivation, must be similar to, make and can on Computer Architecture, realize resulting equation.In order to use particle filter or discrete space filter, must carry out different being similar to.Following chapters and sections have been described these emphatically and have been similar to.
2.4.1 particle filter is approximate
By equation (6), we only need to be similar to
Wherein,
Be weighting function.Now, suppose that we have introduced the independent signal particle
Each independent signal particle has the rule identical with historical signal, and the definition weight
So it follows the deFinnetti theorem and the law of large numbers
2.4.2 discrete space is approximate
If X
iState space be made as E, then, work as N for each N ∈ N
During ∞, make L
NAnd M
NSatisfy l
N ∞ and M
N ∞.For
Suppose
Be the part of E, thereby
And suppose that all discrete states are in the different unit, that is, it only according to age expansion and contraction, causes making α=α
N→ 0.Then, we get
And definition
And carry out following proof directly perceived:
Get η (C
N)=j means for all i ∈ D
NWith
And test function
Substitution (10), we obtain
Wherein,
So, have
2.5 use the meticulous random grid filter of Discrete Finite state space
At United States Patent (USP) 7,188, in 048, describe the general type of REST filter in detail.Proved that this method and system can be used in the multiple application, particularly can be used for Euler's spatial pursuit problem and discrete counts problem of measure.Yet, had been found that some improvement of relevant this method, to one embodiment of the present of invention, they have obviously reduced storage and calculation requirement., described the new method and system of REST filter herein, wherein can use discrete and finite state space carries out modeling to signal.For the sake of clarity, provided the example that uses targeted ads issue model, yet it is the problem of feature with following environment that this method also can be used for any.
2.5.1 environment description
In some problem, signal is by zero or more a plurality of targets
And zero or more a plurality of forms
Constitute.For example, in the targeted ads issue, an embodiment of signal model is X
t=(X
t, R
t) form, wherein, X
tFor the experience of target (perhaps, more particularly, being the kinsfolk) is estimated, and only there is a form.In addition, each target and form only have discrete and a limited number of state, and have limited number target and form (therefore, having the possible combination of limited number target and form).A limited number of combination needn't be all possible combination-only require a limited number of rationally (legitimate) combination.For example, can derive family's (meaning the family that wherein has the particular demographic situation) of limited possible type from depending on geographical demographic census information by the level of relative particulate.Replacement has the individual and (counts n until a certain maximum kinsfolk
MAX) all potential combinations, only need those combinations that may find in given geographic area are considered as reasonably, and they be included in the state space.
In these restricted problems, some in the state of one or more target and/or one or more form can be constant in the short-term that carries out best estimate.In these cases, such state information remains unchanged, and the other parts of state information keep changing.In an embodiment of family's signal model, can be considered as each individual age, sex, income and education degree in the family constantly, because these values just can change after the long time period, and DSTB estimates only to carry out several weeks.Yet the current situation of watching of family's shape information changed in the quite short time period, the result, and these states still change in estimation problem.We will be expressed as the constant part of signal
And the changing unit of signal is expressed as
Exist the possible invariant state of N (by
Represent i such state) and at the M of i invariant state
iIndividual possible variable condition (by
Represent j state).
2.5.2 REST finite state space system survey
Fig. 2 has described a preferred embodiment of REST filter in the finite state space environment.REST is made of one group of invariant state unit, a possible set and their the invariant state characteristic of the target of each invariant state unit representation signal and form.Each constant unit comprises one group of variable condition unit, and the possible time dependent state of given constant unit is represented in each variable condition unit.Impliedly, change unit comprises the invariant state information of their the constant unit of father, this means the concrete sneak condition of each change unit representation signal.For the purpose of, constant unit itself is only represented and is compiled container object, and purpose for convenience and using.Can the group that become unit and constant unit be stored on the computer media with the form of array, vector, tabulation or formation.Can from such container, remove the unit that does not have particle counting at given time t, to reduce space and calculation requirement, although need to insert again the mechanism of such unit in the future.
As shown in Figure 3, each variable condition unit comprises particle counting
This particle counting has been represented the amplitude of the discretization of this unit.As mentioned previously, this amplitude is used to calculate the conditional probability of given state.Each variable condition unit also comprises one group of imagination clock
These imaginary clock representatives are from the possible state variation of given state unit.For each variable condition unit, there is Q
I, jIndividual possible state exchange.In this environment, all effective statuses are handled and are occurred in the same invariant state unit.Change when distributing,, use to be called the particle counting increment in case finished processing in succession for the condition of considering the REST filter
Interim counter, storage will add to the number of given variable condition unit or the particle removed from given variable condition unit.Having from having state
Those unit of effective status conversion of variable condition unit be called the neighbours of this unit.
As previously discussed, the invariant state unit is the container that is used to simplify information processing.The particle counting of each invariant state unit
Be compiling to its sub-variable condition unit particle counting.Similarly, the imaginary clock of invariant state unit is compiling from all clocks of change unit.This compiles the differentiation that helps filter, because can skip the invariant state that those do not have current particle counting in each stage of handling.
2.5.3 the REST filter develops
Fig. 4 has described the typical differentiation of REST filter.This differentiation method is by using imaginary clock value transfer particles between adjacent cells, and the condition of upgrading filter on section dt sometime distributes.Particle mobile between adjacent cells called incident (we replace moving of particle with extra life with dead usually, occur to allow more audience ratings to offset).Integrally simulate such incident, with the computing cost that reduces to develop.The number of the incident of simulating is based on the total imaginary clock and the λ of all unit
tFig. 5 has described the method how many particles are shifted to each neighbour that defines.When finishing the simulation of incident, can upgrade particle counting, and readjustment (seal back) imaginary clock, with the variation of state of expression filter.
In U.S. Patent No. 7,188, the method described in 048 was compared with before, had added additional step, to improve the efficient of filter.Particularly, now under push away observation procedure before, carry out the adjustment of unit particle counting, and before particle control, added and time drawn routine.In some problem, according to observation information, some state may not become the possibility of current demand signal state.For example, if write down the channel change, then family must have at least one member to watch.In these cases, must redistribute effective status to the particle in all disarmed states pro rata.So, to redistribute if exist
Individual particle, then all effective variable condition unit will receive
Individual particle, and reception had probability
Extra particle.When use this type based on the adjustment of observation the time, the speed suitably change probably of the differentiation of restriction signal is with consistent with the use of observation data under this mode.
In order to strengthen the robustness of REST filter, added reflow method.This method is used a certain function
With initial distribution v according to signal,
Individual particle adds the variable condition unit to.The number that adds the particle of each unit to depends on the integrality of time, given unit and filter.The method has guaranteed that filter does not converge on one or more invariant state under situation about can't recover from incorrect localization.
2.6 head end is estimated
In order to maximize the profit ability of Multi-Service Operator advertisement promulgating service, the decision which commercial advertisement is distributed in one group of DSTB is vital.Since can obtain relevant asymptotic Optimal Nonlinear filter based on DSTB, based on the distribute more information of actual group of viewers of commercial advertisement of (perhaps estimating) of condition from the condition of its derivation, the price of concrete commercial advertisement period can more dynamically change, thereby has improved gross profit.
In order to utilize this potentiality, at the head end place, the grab sample of estimating according to condition DSTB is estimated comprising the family that compiles such as the information of the number in each demographics.Comprise preferred embodiment with drag.
2.6.1 head end signal model
The head end signal model is made up of related characteristics (trait) information of the potential and current TV viewer with the DSTB box that is connected in concrete head end.The definition status space S, it represents such one group of characteristic at single individual.In one embodiment of the invention, this space can and watch history to constitute by individual's the range of age, sex recently.In order to keep tracking, order to the individual
Be unmanned H/S, C
nFor having the set of n individual H/S
C
m={ ((s
1, n
1) ..., (s
r, n
r)): s
i∈ S, obviously n
1+ n
2+ ...+n
r=n}.
Like this, the set of family will be the union of the family that wherein has n people
In fact, will exist us can the treatable maximum N of family, we be set to by family's state space
Wherein N is a certain big number.
In order to handle the estimation that oppositely transmits from DSTB by grab sample mechanism, we also wish to follow the trail of current channel at each DSTB.This means, obtain comprising potential family group of viewers, watch each DSTB state of situation and current channel from following formula
Wherein, exist DSTB may turn to its M possible channel.
We do not worry single DSTB, even do not worry which DSTB is in particular state yet, and worry that how many DSTB are in state d ∈ D.Therefore, the signal X that our order will be followed the tracks of is limited Counting Measures, in each classification d ∈ D the number of DSTB is counted.
In one embodiment of the invention, can follow the trail of the compiling of possible number of DSTB in each classification, with the minimization calculation requirement.Under these circumstances, use the atom (atom) of size, make summation still reach the maximum number of DSTB as o.For example, suppose to exist 100 ten thousand DSTB.Then we will have 100,000 atoms (each atom is made up of a=10 DSTB) that are distributed on the D.Suppose the Counting Measures on M (D) the expression D, M (D) expression has the subclass of the M (D) of 100,000 atoms just.Signal will develop according to martingale problem mathematics ground
Wherein, t → M
t(f) be halter strap at each the continuous bounded function f on the M (D),
For substantially supposing determined certain operator naturally according to what DSTB audience ratings and family independently took action.
Do not provide the family of its demographics to be thought of as the part of signal by exposed mode any.
2.6.2 head end observation model
Herein, we have described two observation models: a grab sample that is used for DSTB, one is used for submitting to statistics.
For the grab sample observation model, we, consider channel and group of viewers, and make V as in the last chapters and sections by making the signal of X for us
kRepresent in the sampling process at moment t
kSelection at random.For the purpose of accurately, suppose to have M DSTB at specific head end, and supposition (recall) once more: believe that the current DSTB that is watching of at least one individual will provide the sample with 5% fixation probability for one.So, V
kTo be to have the matrix of line number at random, each row be formed by M with proper what a nonzero term, and nonzero term is corresponding to the index (index) of the concrete DSTB that sample is provided.Line number should be corresponding to the number of the DSTB that sample is provided.On each row, the position of nonzero term is different naturally, and will be selected equably in possible arrangement, to reflect the actual sampling of being carried out.
Now, order
Be the corresponding channel of (row) vector sum of the condition distribution group of viewers of M DSTB, described both all are in t constantly
kSo this observation process will be
Herein, V
kTo select at random, h will provide selected function with feedack.
Submit statistical model to for compiling advertisement, we have function H
K, jThe time index sequence, this function provides before at moment t
k-t
jThe counting of the various advertisements of submitting to.Because some DSTB may not return any information because of temporary derangement (that is, ' observation of losing '), and, will there be a spot of noise W owing to do not guarantee to be used for determining that the estimated group of viewers of successfully submitting to is correct
K, j
From compile submit statistics to second observation information will for
Herein, inverse change on the position section of the scope of j in the report period, t
kBe report time period.
2.6.3 head end filter
Become probability distribution at the signal of head end from DSTB.
2.7 the head end commercial advertisement is selected
In certain embodiments of the present invention, can also be can be used for compiling the out of Memory of group of viewers.For example, compiling (and may postpone) advertisement submits to statistics that deduction to the estimated group of viewers of DSTB also can be provided, and any ' exposed mode ' information, the exchange of their state information (demographics, consumer mentality etc.) as some compensation selected whether to provide by family thus.
In this is provided with, with regard to the details of the treaty, available resources and following signal condition, the commercial advertisement contract is modeled as the figure of the profit that increases progressively.We call contract figure to these figure, and it arrives with the audience ratings that depends on the details of the treaty, signal condition and economic environment.Some details of the treaty comprise:
With the number of times (can comprise minimum and max-thresholds) of the commercial advertisement of playing, it may be thousands of times.
Time range with the day/time-of-week of the commercial advertisement of playing;
One or more target demographic statistics situation of commercial advertisement;
The concrete channel or the program of commercial advertisement will be play;
Write the client of contract,
Wherein some are optional.
The arrival at random of contract figure is called contract figure process.And, if under the situation of given state (now with following) and environment, the resource of being distributed is no more than available resources, be the available commercial advertisement position on each classification, then resource (needing not be maximum assignable for any contract) called feasible selection to the distribution of contract figure process.At this moment because these Limited resources become exhausted when accepting contract, by utility function to current and following potential profit modeling.This utility function is gathered the stream (current and following arrival at random) of available contract figure, and returns the number of indication profit with dollar or some other satiable form.Because the future behaviour at random of contract figure, utility function can not provide maximum profit simply under the situation of not considering the expectation profit about departing from, and can not bring significant low profit risk to guarantee maximization.
In order to carry out the best business advertisement selection, need the following model of definition: head end signal model, head end observation model, contract generation model and practicality (profit) model.
2.7.1 contract model
The commercial advertisement contract that produces is modeled as gauge point process on the contract figure.The rate dependent that contract arrives is in previous performed contract and the external factor such as economic condition.
Suppose that l represents that Lesbegue measures.Then we make C represent to have on it the space of possible contract figure of a certain topology, { η
t, t 〉=0} represents at the Counting Measures random process until the arrival of the contract figure of moment t, ξ represent to have a certain C that on average estimates v * l * l * [0, ∞) * [0, ∞) Poisson (Poisson) on is estimated.And [0, t) 0 until moment t, but when not comprising t constantly during the arrival of contract figure, we make λ (c, η to record from constantly as η
[0, t), t) for the speed that will arrive with contract figure c ∈ C at the new contract of moment t (with respect to v).Then, we are according to following random difference equation modeling:
To all A ∈ B (C).
When accepting contract, can change the above mentioned details of the treaty.As a result, the details of the treaty are modeled as depend on the external environment condition that may develop in time.
2.7.2 utility function is described
For the ease of expression, but the program downloading information D during based on moment s
s, we make R (D
s) be the present and following available resources.
We can not accept whole contracts of being produced, and we must determine can't see under following situation, are to accept or contract of refusal.We are shown feasible selection permitting option table, make each resource allocation decision not use following contract or following observation information.According to the mark of last chapters and sections, we suppose n
tRepresentative is up to (and comprising) t constantly, the number of the various types of contracts that arrived, and get
Wherein, Q represents all potential customers' set, { l
s, s 〉=0} is a selection course, that is, and and to each contract c Resources allocation.So, if for each s 〉=0 l
s≤ R (D
s) and l
sDo not use following contract not use observation information, then { l yet
s, s 〉=0} can permit selecting, promptly for each s 〉=0, relatively
Can survey.Now, r
t(l) representative is until moment t, by the profit that can permit selecting l to obtain.For the ease of expression, we make, and Λ is all such set of permitting selecting.
Current profit of utility function J balance and following profit, and be equilibrated on the concrete contract chance that obtains high profit and do not have or the risk of low profit.In order to ensure reasonably beginning, we will remove the weight of following profit by exponential manner.And for excessively not radical, we will comprise class variable (variance-like) condition.For little constant λ, a〉0, an embodiment of the utility function of gained is:
J(X,l)=∫
[0,∞)e
-λt[r
t(l)-a(r
t(l))
2]dt,
Like this, the commercial advertisement selection course be on l ∈ Λ, maximize E[J (X, l)].Can use one or more asymptotic optimum filter to realize this goal.
Illustrative and descriptive provided above description the of the present invention.And this description is not intended to the present invention is limited to form disclosed herein.Therefore, the present invention is carried out knowing variation and the modification suitable with knowledge with the technical ability of correlation technique with above-mentioned religion, all be in the scope of the present invention.Embodiment described above also is intended to explain to realize known best mode of the present invention, and makes other those of skill in the art in this technology utilize the present invention and of the present invention one or more specifically to use or use desired various modification by such or other embodiment.Also be intended to make claims to comprise the alternative embodiment of prior art institute allowed band.
Claims (32)
1. method that is used for the user that assets are oriented to communication network is equipped with the user of equipment may further comprise the steps:
At the input that the user is equipped with equipment, produce observation model according to one or more user;
Incorporate described observation model into reflect the signal of at least one user's composition that described user is equipped with one or more user of equipment at the time;
Approximate condition by the described signal under the situation of given measurement data distributes, and estimates at pay close attention to described user's composition at place constantly; And
When the equipment that is equipped with at described user carries out orientation to assets, use described estimated user's composition.
2. method according to claim 1, wherein, the described click steam that is input as along with user's input of time, and described observation model is modeled as Markov chain to described click steam.
3. method according to claim 2, wherein, described observation model consideration is used for the program-related information by at least some indicated Web contents of described input.
4. method according to claim 3, also comprise step: use Mathematical Modeling to handle described Markov chain, the observation of wherein said Markov chain only can be transformed into state complete or collected works' subclass, and wherein said subclass depends on the current state of described Markov chain.
5. method according to claim 1, wherein, described modeling procedure comprises: described observation model is modeled as Markov chain on k step Markov chain.
6. method according to claim 5, wherein, the transfer function of observation Markov chain depends on the position of wanting estimated signals.
7. method according to claim 1, wherein, the independent factor that described signal is set up as the described user's composition of representative and influences described user's input.
8. method according to claim 1, wherein, feasible described user's component list is shown of the model of described signal comprises two or more users.
9. method according to claim 1, wherein, the variation of the described user's composition of the feasible expression of the model of described signal.
10. method according to claim 9, wherein, described variation is the variation of the number of users that is associated with described user's outfit equipment.
11. method according to claim 1, wherein, described modeling procedure comprises the definition filter, to obtain the probability Estimation of described signal according to described observation model and measurement data.
12. method according to claim 11, wherein, described modeling procedure comprises definition non-linear filtration device, to obtain the probability Estimation of described signal according to described observation model.
13. method according to claim 12, wherein, described modeling procedure also comprises sets up approximate filter, is used for the operation of approximate described non-linear filtration device.
14. method according to claim 13, wherein, described approximate filter is a particle filter.
15. method according to claim 13, wherein, described approximate filter is the discrete space filter.
16. method according to claim 1, wherein, described use step comprises: provide information based on described user's composition the network platform that assets are inserted the content stream of described network to operating.
17. method according to claim 16, wherein, the described user of described message identification is equipped with one or more user's of equipment demographics.
18. method according to claim 17, wherein, described platform can be operated and compile the user's composition information that is associated with a plurality of user's outfit equipment, and one or more assets that are used to insert according to the described Information Selection of compiling.
19. method according to claim 16, wherein, it is observation model that described platform can be operated the information processing that is equipped with equipment from a plurality of users, and uses filter to estimate the described constituent of being paid close attention to network spectators constantly at described observation model.
20. method according to claim 17, wherein, described platform can operate the extraneous information according to the submission value of described constituent and the concrete assets of influence, selects the assets that are used to insert.
21. method according to claim 16, wherein, described information is according to described user's composition, and sign is used to submit to one or more suitable assets that described user is equipped with equipment.
22. method according to claim 1, wherein, the step of described use comprises: be equipped with the assets that described one or more user selects to be used to submit in the equipment place described user.
23. method according to claim 1, wherein, the step of described use comprises: report is equipped with the appropriate degree of the assets of submitting at the equipment place described user at described one or more user.
24. one kind is used for assets are oriented to the device that the communication network user is equipped with the user of equipment, comprises:
Port is used to receive the input information at the input of user's outfit equipment about one or more user; And
Processor, be used for providing observation model according to described input, observation model is modeled as to depend at the time reflects that described user is equipped with one or more user's of equipment the signal of at least one user's composition, pay close attention to constantly that user's composition at place is defined as the state of signal, and when the equipment that is equipped with at the user carries out orientation to assets, use determined user's composition.
25. device according to claim 24, wherein, described processor is used to define the non-linear filtration device, to obtain the estimation of described signal according to described observation model and measurement data.
26. device according to claim 25, wherein, described processor is used to set up the approximate filter of the operation that is used for approximate described non-linear filtration device.
27. device according to claim 26, wherein, described non-linear filtration device is a kind of in particle filter and the discrete space filter.
28. device according to claim 24 comprises that also transmission is used for assets are oriented to the independently port of the information of the network platform, wherein, described information is based on described determined user's composition.
29. a method that is used in the directed assets of radio network may further comprise the steps:
Group analysis is corresponding to the data flow of a series of user's inputs; And
Be applied to the characteristic that is associated with user's spectators classification being used to mate this logic that flows described pattern.
30. method according to claim 29, wherein, the step of described group analysis comprises: set up observation model, wherein said a series of user's inputs are modeled as Markov chain.
31. method according to claim 29, wherein, the step of described applied logic comprises: use non-linear filtration device model to import from described a series of users and extract the signal estimation and distribute, and use described signal to obtain described characteristic.
32. method according to claim 29, wherein, the step of described applied logic comprises: the approximate filter of operation, and with the operation of approximate described non-linear filtration device.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US74624406P | 2006-05-02 | 2006-05-02 | |
US60/746,244 | 2006-05-02 |
Publications (1)
Publication Number | Publication Date |
---|---|
CN101473647A true CN101473647A (en) | 2009-07-01 |
Family
ID=38668530
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CNA200780023082XA Pending CN101473647A (en) | 2006-05-02 | 2007-05-02 | Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising |
Country Status (6)
Country | Link |
---|---|
EP (1) | EP2016766A4 (en) |
JP (1) | JP5172826B2 (en) |
CN (1) | CN101473647A (en) |
AU (1) | AU2007247995B2 (en) |
CA (1) | CA2651165C (en) |
WO (1) | WO2007131068A2 (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102541877A (en) * | 2010-12-07 | 2012-07-04 | 腾讯科技(北京)有限公司 | Image-text information issuing method and system |
CN103329559A (en) * | 2011-04-06 | 2013-09-25 | 兰屈克有限公司 | Method and system for detecting non-powered video playback devices |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8761658B2 (en) | 2011-01-31 | 2014-06-24 | FastTrack Technologies Inc. | System and method for a computerized learning system |
CN104584001B (en) * | 2012-08-22 | 2017-10-27 | 兰屈克有限公司 | System and method for predicting viewing-data |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1223757B1 (en) * | 2001-01-09 | 2006-03-22 | Metabyte Networks, Inc. | System, method, and software application for targeted advertising via behavioral model clustering, and preference programming based on behavioral model clusters |
US7424409B2 (en) * | 2001-02-20 | 2008-09-09 | Context-Based 4 Casting (C-B4) Ltd. | Stochastic modeling of time distributed sequences |
US7818206B2 (en) | 2001-10-29 | 2010-10-19 | Teradata Us, Inc. | System and method for profiling different users having a common computer identifier |
US20030126601A1 (en) * | 2001-12-31 | 2003-07-03 | Koninklijke Philips Electronics N.V. | Visualization of entertainment content |
US8245251B2 (en) * | 2002-12-06 | 2012-08-14 | General Instrument Corporation | Method and apparatus for predictive tuning in digital content receivers |
US20040172650A1 (en) * | 2003-02-28 | 2004-09-02 | Hawkins William J. | Targeted content delivery system in an interactive television network |
US7188048B2 (en) * | 2003-06-25 | 2007-03-06 | Lockheed Martin Corporation | Refining stochastic grid filter |
-
2007
- 2007-05-02 JP JP2009510062A patent/JP5172826B2/en active Active
- 2007-05-02 EP EP07797322A patent/EP2016766A4/en not_active Withdrawn
- 2007-05-02 AU AU2007247995A patent/AU2007247995B2/en active Active
- 2007-05-02 WO PCT/US2007/068075 patent/WO2007131068A2/en active Application Filing
- 2007-05-02 CA CA2651165A patent/CA2651165C/en active Active
- 2007-05-02 CN CNA200780023082XA patent/CN101473647A/en active Pending
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102541877A (en) * | 2010-12-07 | 2012-07-04 | 腾讯科技(北京)有限公司 | Image-text information issuing method and system |
CN102541877B (en) * | 2010-12-07 | 2014-10-22 | 腾讯科技(北京)有限公司 | Image-text information issuing method and system |
CN103329559A (en) * | 2011-04-06 | 2013-09-25 | 兰屈克有限公司 | Method and system for detecting non-powered video playback devices |
CN104581209A (en) * | 2011-04-06 | 2015-04-29 | 兰屈克有限公司 | Method and system for detecting non-powered video playback devices |
CN103329559B (en) * | 2011-04-06 | 2016-11-16 | 兰屈克有限公司 | The method and system of playback apparatus is shot with video-corder for detecting non-powered |
CN104581209B (en) * | 2011-04-06 | 2017-07-04 | 兰屈克有限公司 | For detecting that non-powered shoots with video-corder the method and system of playback apparatus |
Also Published As
Publication number | Publication date |
---|---|
EP2016766A4 (en) | 2012-03-28 |
JP2009536412A (en) | 2009-10-08 |
WO2007131068A3 (en) | 2008-09-25 |
JP5172826B2 (en) | 2013-03-27 |
CA2651165A1 (en) | 2007-11-15 |
AU2007247995B2 (en) | 2012-03-15 |
WO2007131068A9 (en) | 2008-02-28 |
WO2007131068A2 (en) | 2007-11-15 |
AU2007247995A1 (en) | 2007-11-15 |
CA2651165C (en) | 2014-02-04 |
EP2016766A2 (en) | 2009-01-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113473187B (en) | Cross-screen optimization of advertisement delivery | |
US20160142754A1 (en) | Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising | |
CN104035926B (en) | A kind of dispensing of internet information and system | |
CN108352025A (en) | Television advertising period positioning based on the online behavior of consumer | |
US20160086208A1 (en) | Systems, methods, and articles of manufacture to measure online audiences | |
CN105897616B (en) | Resource allocation method and server | |
CN106776928B (en) | Position recommendation method based on memory computing framework and fusing social contact and space-time data | |
CN102314488A (en) | Distribute to specific demography and utilization rate and to obtain the method and apparatus of anonymous audience measurement data from webserver data | |
CN104063801A (en) | Mobile advertisement recommendation method based on cluster | |
WO2012024316A2 (en) | Unified data management platform | |
CN104737152A (en) | A system and method for transferring information from one data set to another | |
US10911821B2 (en) | Utilizing combined outputs of a plurality of recurrent neural networks to generate media consumption predictions | |
US20200074498A1 (en) | Systems and methods for improving social media advertising efficiency | |
CN109936860B (en) | Network information flow distribution method, flow distribution model training method and device | |
CN103649981A (en) | Method and apparatus for delivering targeted content | |
CN109417644A (en) | The income optimization launched across screen advertisement | |
CN107657487A (en) | A kind of advertisement placement method and device based on verification | |
CN103069794A (en) | Improved network data transmission system and method | |
CN108965938A (en) | Potential paying customer prediction technique and system in smart television | |
WO2020250723A1 (en) | Information processing method, information processing device, and program | |
CN101473647A (en) | Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising | |
CN104871547A (en) | Method and system for efficiently compiling media content items for a media-on-demand platform | |
CN102754117A (en) | Method, system and controller for providing goods and /or services to consumers | |
CN115330428A (en) | Data calibration method and device, computer equipment and readable storage medium | |
CN111861542A (en) | Method, device, terminal and non-transitory computer readable storage medium for delivering advertisement |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C12 | Rejection of a patent application after its publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20090701 |