CN109710778B - Multimedia information processing method, device and storage medium - Google Patents
Multimedia information processing method, device and storage medium Download PDFInfo
- Publication number
- CN109710778B CN109710778B CN201811627152.9A CN201811627152A CN109710778B CN 109710778 B CN109710778 B CN 109710778B CN 201811627152 A CN201811627152 A CN 201811627152A CN 109710778 B CN109710778 B CN 109710778B
- Authority
- CN
- China
- Prior art keywords
- data
- multimedia
- data set
- user
- machine learning
- 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.)
- Active
Links
- 230000010365 information processing Effects 0.000 title claims abstract description 30
- 238000003672 processing method Methods 0.000 title claims abstract description 17
- 238000010801 machine learning Methods 0.000 claims abstract description 29
- 238000007781 pre-processing Methods 0.000 claims abstract description 14
- 238000012545 processing Methods 0.000 claims description 41
- 238000000034 method Methods 0.000 claims description 26
- 230000006399 behavior Effects 0.000 claims description 21
- 238000007906 compression Methods 0.000 claims description 11
- 230000006835 compression Effects 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 5
- 230000002441 reversible effect Effects 0.000 claims description 3
- 238000013144 data compression Methods 0.000 claims 4
- 230000000875 corresponding effect Effects 0.000 description 15
- 239000011159 matrix material Substances 0.000 description 12
- 230000008569 process Effects 0.000 description 12
- 230000007704 transition Effects 0.000 description 12
- 238000004422 calculation algorithm Methods 0.000 description 10
- 238000005516 engineering process Methods 0.000 description 9
- 238000004364 calculation method Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 238000009826 distribution Methods 0.000 description 5
- 238000004891 communication Methods 0.000 description 4
- 230000008451 emotion Effects 0.000 description 4
- 230000009471 action Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 2
- 238000005266 casting Methods 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 230000036961 partial effect Effects 0.000 description 2
- 230000010267 cellular communication Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a multimedia information processing method, which comprises the following steps: acquiring data of a first type of multimedia file; preprocessing the data for adapting a machine learning model; and in the machine learning model, determining the interest degree of the user in the first type of multimedia files according to the preprocessed data. The invention also discloses a multimedia information processing device and a storage medium.
Description
Technical Field
The present invention relates to the field of communications, and in particular, to a method, an apparatus, and a storage medium for processing multimedia information.
Background
In the related art, based on modeling methods such as a decision tree model, an optimal combination model, a neural network model and the like, a social network link prediction technology, a user attribute prediction technology in a social network, an emergency hot event technology in the social network, a social network popularity prediction technology, a commodity-user click rate prediction technology, a commodity-purchase amount prediction technology and other data-based user behavior prediction technologies are adopted to predict the interest degree of a user in the content of a multimedia file.
However, when predicting the interest degree of the user in the multimedia file content in the related technology, the influence between the micro individuals and the influence of the macro trend cannot be considered; in addition, the weight of the data adopted when predicting the interest degree of the user in the multimedia file content and the interest degree of the future user in the multimedia file content are not considered. Based on the above factors, the related art cannot comprehensively predict the user's interest level in the content of the multimedia file.
Disclosure of Invention
In view of the above, embodiments of the present invention are directed to a multimedia information processing method, apparatus and storage medium, capable of processing multimedia information
In order to achieve the above purpose, the technical solution of the embodiment of the present invention is realized as follows:
in a first aspect, an embodiment of the present invention provides a multimedia information processing method, where the method includes:
acquiring data of a first type of multimedia file;
preprocessing the data for adapting a machine learning model;
and in the machine learning model, determining the interest degree of the user in the first type of multimedia files according to the preprocessed data.
In the foregoing scheme, the acquiring data of the first type of multimedia file includes:
and acquiring data of the first-class multimedia files from at least two users in a first time interval.
In the above scheme, the data includes at least one of:
usage data of the first type of multimedia file;
attribute data of at least two multimedia files of the first class of multimedia files.
In the foregoing solution, the preprocessing the data for adapting to the machine learning model includes:
dividing the use data in the first time interval into at least four first data sets, wherein the time interval between every two first data sets is a fixed value;
performing first merging processing on the first data set according to a first compression ratio to obtain at least two second data sets;
and carrying out second combination processing on the second data set according to a second compression ratio to obtain at least one third data set.
In the foregoing solution, the preprocessing the data for adapting to the machine learning model includes:
dividing the third data set into at least two fourth data sets based on a multimedia file Identification (ID);
dividing the fourth data set into fourth data subsets with corresponding dimensions based on operations of users in different dimensions of the first multimedia files;
and processing the fourth data subset to acquire the attribute corresponding to the fourth data subset.
In the foregoing solution, the preprocessing the data for adapting to the machine learning model includes:
and carrying out reverse combination processing on the fourth data subset to obtain a fifth data set representing the behavior attribute of the first multimedia file at a first time point of the user and a sixth data set representing the behavior attribute of the first multimedia file at a second time period of the user.
In the foregoing solution, the determining the interest level of the user in the first type of multimedia file according to the preprocessed data includes:
determining the machine learning model based on the fifth dataset;
and determining the interest degree of the user in the first type of multimedia files based on the sixth data set and the machine learning model.
In a second aspect, an embodiment of the present invention provides a multimedia information processing apparatus, where the apparatus includes:
the acquiring unit is used for acquiring data of the first type of multimedia files;
a processing unit for preprocessing the data for adapting a machine learning model;
and the determining unit is used for determining the interest degree of the user in the first type of multimedia files according to the preprocessed data in the machine learning model.
In a third aspect, an embodiment of the present invention provides a multimedia information processing apparatus, including a processor and a memory for storing a computer program capable of running on the processor, wherein,
the processor is used for executing the steps of the multimedia information processing method when the computer program is run.
In a fourth aspect, an embodiment of the present invention provides a storage medium, which stores an executable program, and when the executable program is executed by a processor, the storage medium implements the multimedia information processing method described above.
The multimedia information processing method, the device and the storage medium provided by the embodiment of the invention are used for acquiring the data of a first-class multimedia file and preprocessing the data for adapting to a machine learning model; and in the machine learning model, determining the interest degree of the user in the first type of multimedia files according to the preprocessed data. In this way, the first-class multimedia file data is obtained by taking the weight of the interest degree of the future user for the content of the multimedia file as a reference factor; the interest degree of the group users in the multimedia file is abstracted to a random process of a set of individual user behaviors on a discrete time set, so that the interest degree of the users in the content of the multimedia file can be comprehensively and accurately predicted.
Drawings
FIG. 1 is a diagram illustrating an alternative hardware configuration of a multimedia information processing apparatus according to an embodiment of the present invention;
FIG. 2 is a diagram illustrating a die configuration according to the related art;
FIG. 3 is a diagram illustrating implicit state relationship transition in the related art;
FIG. 4 is a diagram illustrating a hidden Markov model in the related art;
FIG. 5 is a schematic view of an alternative flow chart of a multimedia information processing method according to an embodiment of the present invention;
fig. 6 is a schematic flow chart of processing data according to an embodiment of the present invention;
FIG. 7 is a block diagram of a multimedia information processing apparatus according to an embodiment of the present invention.
Detailed Description
So that the manner in which the features and aspects of the embodiments of the present invention can be understood in detail, a more particular description of the embodiments of the invention, briefly summarized above, may be had by reference to the embodiments, some of which are illustrated in the appended drawings.
In practical applications of the method and apparatus for processing multimedia information provided in the embodiments of the present invention, each functional module in the apparatus for processing multimedia information may be cooperatively implemented by hardware resources of a device (such as a terminal device, a server, or a server cluster), such as computing resources of a processor and communication resources (such as being used for supporting communications in various manners, such as optical cables and cellular communications). Fig. 1 schematically shows an alternative hardware configuration of a multimedia information processing apparatus 10, which includes a processor 11, an input/output interface 13 (e.g., one or more of a display, a keyboard, a touch screen, a speaker microphone), a memory 14, and a network interface 12, and which may be communicatively coupled via a system bus 15.
Of course, the embodiments of the present invention are not limited to being provided as methods and hardware, and may be provided as a storage medium (storing instructions for executing the multimedia information processing method provided by the embodiments of the present invention), and different implementations are illustrated below.
Mobile terminal application program and module
The embodiment of the invention can provide a software module designed by using programming languages such as C/C + +, Java and the like, and the software module is embedded into various mobile terminal Apps based on systems such as Android or iOS and the like (stored in a storage medium of the mobile terminal in the form of executable instructions and executed by a processor of the mobile terminal), so that the relevant data mining tasks are completed by directly using the computing resources of the mobile terminal, and data, intermediate results or final results are transmitted to a remote server in various network communication modes periodically or aperiodically or are stored locally at the mobile terminal.
Second, server application program and platform
The embodiment of the invention can provide application software designed by using programming languages such as C/C + +, Java and the like or a special software module in a large-scale software system, operate in a server end (stored in a storage medium of the server end in an executable instruction mode and operated by a processor of the server end), calculate at least one of various received original data, intermediate data of each level and final results from other equipment together with some data or results existing on the server to obtain an updated result, then output the updated result to other application programs or modules in real time or non-real time for use, and also write the updated result into a database or file at the server end for storage.
The embodiment of the invention can also provide a data mining platform, a credit evaluation platform (for evaluating the credit of users) and the like used by individuals, groups or enterprises by carrying a customized and easily interactive network (Web) Interface or other User Interfaces (UI) on a distributed and parallel computing platform formed by a plurality of servers. The user can upload the existing data packets to the platform in batch to obtain various calculation results, and can also transmit the real-time data stream to the platform to calculate and refresh each stage of results in real time.
Third, server side Application Program Interface (API) and plug-in
The embodiment of the invention can be provided for realizing the API, the Software Development Kit (SDK) or the plug-in of the server side, is called by other server side application program developers and is embedded into various application programs.
Fourth, mobile device client API and plug-in
The embodiment of the invention can also provide an API, an SDK or a plug-in for the mobile equipment end, is called by other mobile end application program developers, and is embedded into various application programs.
Before describing the embodiments of the present invention in detail, a hidden markov model will be briefly described.
Hidden Markov Models (HMM) are statistical models used to describe a Markov process with Hidden unknown parameters. The processing procedure is to determine the implicit parameters of the process from the observable parameters. These parameters are then used for further analysis, such as pattern recognition.
The hidden markov model is exemplified below.
As shown in fig. 2, assume that there are three different dice, the first of which is a hexahedron (called this die D6) with 6 faces, each face (1, 2, 3, 4, 5, 6) having a probability of occurrence of 1/6. The second die is a tetrahedron (this die is called D4) and the probability of occurrence of each face (1, 2, 3, 4) is 1/4. The third die is an octahedron, (this die is called D8), and the probability of occurrence of each face (1, 2, 3, 4, 5, 6, 7, 8) is 1/8.
The dice are then rolled, one of the three dice is picked, and the probability of picking each dice is 1/3. We then roll the dice to get one of the numbers 1, 2, 3, 4, 5, 6, 7, 8. Repeating the above process without stop will result in a series of numbers, each number being one of 1, 2, 3, 4, 5, 6, 7, 8. To get the following string of numbers (throw dice 10 times): 1635273524. this string of numbers is called the visible state chain. However, in hidden markov models, there is not only such a chain of visible state but also a chain of hidden state. This chain of hidden states is the sequence of dice used. For example, the implicit state chain might be: D6D 8D 8D 6D 4D 8D 6D 6D 4D 8
In general, a markov chain spoken in an HMM is actually referred to as a chain of hidden states because there is a transition probability (transition probability) between hidden states (dice). An implicit state relationship transition diagram is shown in fig. 3, and in the above example, the probabilities that the next state of D6 is D4, D6 and D8 are 1/3. The next states of D4 and D8 are D4, D6 and D8, and the same is true for the transition probabilities 1/3. Of course, the transition probability can be flexibly set according to needs. For example, it is defined that D4 cannot be connected after D6, the probability of D6 after D6 is 0.9, and the probability of D8 is 0.1. This is a new HMM.
Similarly, although there is no transition probability between visible states, there is a probability between an implicit state and a visible state called an output probability (output probability). Continuing with the above example, the output probability of six sided dice (D6) producing a 1 is 1/6. The probabilities of generating 2, 3, 4, 5, 6 are also 1/6. We can also make other definitions of the output probabilities. For example, a six sided die that is moved by the casino to the hands and feet has a greater probability of casting a 1, 1/2, and a probability of casting a 2, 3, 4, 5, 6, 1/10.
HMM model schematic, as shown in fig. 4, algorithms related to HMM models are mainly classified into three categories, which respectively solve three problems:
1) several dice are known (number of implicit states), what each die is (transition probability), and depending on the outcome of the throw of the dice (visible state chain), it is desirable to determine which die is thrown each time (implicit state chain).
2) Several dice are known (implicit state number), what each die is (transition probability), and from the outcome of the throw of a die (visible state chain), it is desirable to determine the probability of the throw of this outcome.
3) Several dice are known (implicit number of states), it is not known what each die is (transition probability), the results of a number of dice rolls are observed (visible state chain), and it is desirable to determine what each die is (transition probability).
The following describes the multimedia information processing method provided by the present invention in detail.
As shown in fig. 5, an optional flow diagram of the multimedia information processing method provided in the embodiment of the present invention includes the following steps:
step S201, acquiring data of a first type of multimedia file.
In the embodiment of the invention, the multimedia information processing device acquires the data of the first-class multimedia files from at least two users in a first time interval.
The first time interval can be flexibly set, and can be in a week unit, a day unit or a month unit; taking the surrounding units as an example, if the first time interval is three weeks, the multimedia information processing apparatus acquires the multimedia file data within three weeks.
The first type of multimedia file may be obtained by dividing the type of the multimedia file according to an existing division standard or a newly defined division standard. Such as art multimedia files, art-integrated multimedia files, youth multimedia files, love multimedia files, war multimedia files, science and fiction multimedia files, and the like. The multimedia file may be a variety of videos, micro-movies, advertisements, etc.
Optionally, the data of the first type of multimedia file comprises at least one of:
usage data of the first type of multimedia files, and attribute data of at least two of the first type of multimedia files. The usage data may be a user ID, a video ID, an occurrence time of a behavior of a user viewing a multimedia file, a data type (e.g., a click, or a score, or a comment for one video), and a data content (e.g., a viewing duration, a score value, or a comment content). The attribute data may be a video ID, a video duration, etc.
In the embodiment of the invention, the user data is correlated through the video IP to form a behavior data set P of multi-dimensional group users for the multimedia files of specific types. Therefore, the behavior data set P is a two-dimensional matrix formed by a plurality of one-dimensional tuples; optionally, the one-dimensional tuple is in the form of [ unique code, user ID, video ID, behavior occurrence time of the multimedia file, data type, data content, video duration ], wherein the unique code represents one click or one score or one comment.
In the related art, the historical data adopted when determining the interestingness of the user in the multimedia file is generally acquired at a fixed time interval, and the influence of long-term data far away from the current time point, medium-term data moderate to the current time point and short-term data close to the current time point on the prediction result is not considered. In the embodiment of the invention, future data are predicted by using current behavior data such as clicking, grading and commenting of the first-class multimedia files by the user, so that the interest degree of the user in the first-class multimedia files is comprehensively predicted.
Step S202, preprocessing is carried out on the data for adapting to a machine learning model.
In the embodiment of the invention, the multimedia information processing device divides the use data in the first time interval into at least four first data sets, and the time interval between every two first data sets is a fixed value; performing first merging processing on the first data set according to a first compression ratio to obtain at least two second data sets; and carrying out second combination processing on the second data set according to a second compression ratio to obtain at least one third data set.
For example, the usage data in the first time interval is a data set P consisting of a first data set Pi、Pi-1、Pi-2、…Pi-m-2 nForming; the time interval between every two first data sets is a fixed value, i.e. PiAnd Pi-1Time interval between, Pi-1And Pi-2The time intervals between them are equal. Based on the fact that the influence trend of historical data on a future time period in the generalized relativity theory is a descending curve with almost exponential power, the embodiment of the invention decomposes integers in all data into a form of sum of powers of 2, namely 2 is used as a base when data is processed. Due to the native computing mechanism of binary computer storage and computing processing, the embodiment of the invention can greatly improve the overall computing efficiency by adopting 2 as a base when processing data.
In particular implementation, as shown in FIG. 6, for data set P, the first two data sets P are selected firstiAnd Pi-1Performing a first merging process by a dot product algorithm of the matrix to obtain a second data set Mi1. Due to this, the first data set PiAnd Pi-1All are matrixes of 1 row and N columns; thus, the second data set Mi1Is a matrix of 2 rows and N columns, and a second data set Mi1First behavior P in (1)iElement of (1), second data set Mi1Second behavior P in (1)i-1The elements of (1); or a second data set Mi1Is a matrix of 2 rows and N columns, and a second data set Mi1First behavior P in (1)i-1Element of (1), second data set Mi1Second behavior P in (1)iOf (1). Next, for the thirdFrom one dataset to a sixth dataset Pi-2、Pi-3、Pi-4、Pi-5Carrying out first merging processing through a dot product algorithm of the matrix to obtain a second data set Mi2. Due to the first data set Pi-2、Pi-3、Pi-4And Pi-5Second data set M of matrix with 1 row and N columnsi2Is a matrix of 4 rows and N columns, so that the second data set is a matrix of 4 rows and N columns, and Pi-2、Pi-3、Pi-4And Pi-5Respectively a row of elements in the second data set. Then, the seventh to fourteenth data sets Pi-6To Pi-13Carrying out first merging processing through a dot product algorithm of the matrix to obtain a second data set Mi3. By analogy, performing first combination processing on the fifteenth data set to the thirty data set in the first data set through a dot product algorithm of the matrix to obtain a second data set Mi4. It is understood that the first compression ratio is 2nThat is, the data in the first data set is subjected to n times of first combination processing, and the data processed each time is 2nA plurality of; the first time 2 data sets of the first data set are subjected to a first merging process, and the second time 2 data sets of the first data set are subjected to a second merging process2The first merging process is performed on the data sets, and the third time is to perform the first merging process on 2 data sets in the first data set3The first merging process is performed on each data set, and so on.
In particular implementation, as shown in FIG. 6, for a data set Mi1、Mi2、Mi3…M(i-n)2 nFirst, the first two data sets M are selectedi1And Mi2Performing a second combination process by a cross multiplication algorithm of the matrix to obtain a third data set Di. Next, M is selected againi3、Mi4、Mi5And Mi6Performing a second combination process by a cross multiplication algorithm of the matrix to obtain a third data set Di-1. By analogy, the second data set is subjected to the second merging processing for the Nth time, and the number of the third data sets corresponding to the second merging processing for the Nth time is 2NAnd (4) respectively. The third data set is a data set having a two-dimensional row-column structure, wherein each column of dataAn attribute characterizing the behavior of the user, and the number of rows in each of the two third data sets is different.
In the embodiment of the invention, the mode of sampling and compressing data by inverse power fingers is introduced when the first merging processing and the second merging processing are carried out on the data, the problem of low prediction capability caused by excessive unknown variables in pure time series prediction is solved, and the calculation difficulty is reduced.
For the third data set, the embodiment of the present invention divides the third data set into at least two fourth data sets based on the multimedia file ID, and deletes the column element corresponding to the multimedia file ID in the fourth data sets. For example, since the fourth data set is obtained by dividing based on the multimedia file IDs, the multimedia file IDs corresponding to each row of elements in the fourth data set are the same; if the third data set comprises 6 columns of elements, wherein 1 column of elements is used for characterizing the multimedia file ID; the fourth data set comprises 5 column elements and the fourth data set lacks the column elements corresponding to the multimedia file ID compared to the third data set. Fourth data set with SijAnd (4) showing.
Next, the fourth data set is divided into fourth data subsets with corresponding dimensions based on the operations of the user on the first multimedia files with different dimensions. For example, the operations of the user on the first type of multimedia file in different dimensions include: when clicking, commenting and grading the multimedia file, the fourth data set is divided into Sij1、Sij2And Sij3Three fourth subsets of data; each data subset is divided according to one-dimensional operation; if the user clicks the first multimedia file, S is obtained from the fourth data setij1(ii) a Obtaining S from the fourth data set according to the comment operation of the user on the first-class multimedia fileij2(ii) a Obtaining S from the fourth data set according to the grading operation of the user on the first type of multimedia filesij3. For Sij1、Sij2And Sij3The three fourth data subsets, the user ID and the occurrence time of the user's action of viewing the multimedia file, have no reference value to the overall user action in the macroscopic time sense, and thereforeFourth subset of data Sij1、Sij2And Sij3And deleting the elements corresponding to the user ID and the occurrence time of the behavior of the user for watching the multimedia file.
And combining the fourth data subsets based on the user ID to obtain the attribute corresponding to each fourth data subset. For example, for the fourth subset of data Sij1Merging based on the user ID to obtain Sij1The corresponding attribute is average duration; for a fourth subset of data Sij2Merging based on the user ID to obtain Sij2The corresponding attribute is a weighted average of the scores; for a fourth subset of data Sij3Merging based on the user ID to obtain Sij3The corresponding attribute is a comment value, and the comment value is equal to the product of the word frequency of the positive words of the comment and the positive emotion judgment coefficient and then divided by the total number of the positive and negative word frequencies; or the comment value is equal to the product of the word frequency of the negative words of the comment and the negative emotion judgment coefficient, and then the product is divided by the total number of the positive and negative word frequencies. Wherein, the value of the positive emotion judging coefficient is 1, and the value of the negative emotion judging coefficient is-1.
Aiming at the fourth data subsets, obtaining a fifth data set M for representing the behavior attribute of the first-class multimedia file at the first time point by carrying out weighted average on elements representing the same attribute in a plurality of fourth data subsets01i、M02iAnd M03iAnd a sixth data set characterizing behavioral attributes of a user for the first type of multimedia file over a second time period. Optionally, the behavior attribute of the first type of multimedia file includes attention, interest, and propagation, and the first time point is a current i time point. The collection of the fifth data sets from the i-n time point to the i time point constitutes a sixth data set M characterizing the user's behavioral attributes for said first class of multimedia files over a second time period01、M02And M03。
In some embodiments, the machine learning model is a hidden markov model.
Step S203, in the machine learning model, determining the interest degree of the user in the first type of multimedia files according to the preprocessed data.
In some embodiments, the interest level of the user in a particular multimedia file varies with time, satisfying the following relationship: the state at any time t is only related to the previous state and is not related to the state at any other time:
P(it|it-1,Ot-1，...,i1,O1)＝P(it|it-1)；
wherein T is 1, 2, …, T.
In other embodiments, the interest level of the user in a particular multimedia file varies with time, and satisfies the following relationship: the state at any time t is only related to the markov chain state at that time, and not to the states at any other time:
P(Oi|iT,OT，iT-1,OT-1iT-1,...,it+1,Ot+1,it,it-1,Ot-1,...，it,Ot)＝P(Ot|it)
based on the relationship satisfied by the interest degree of the user in the specific multimedia file along with the change of time, the embodiment of the invention introduces the hidden Markov model to carry out relevant prediction on the hidden sequence corresponding to the input data sequence. The hidden markov model is determined by the following three objects: 1) an initial probability distribution pi; 2) a state transition probability distribution A; 3) a probability distribution B is observed. Thus, a hidden markov model λ can be represented in ternary notation as: λ ═ a, B, pi.
According to the current fifth data set M01i、M02iAnd M03iInitial probability distributions π 1, π 2 and π 3 can be determined. According to a sixth data set M01、M02And M03Observed probability distributions B1, B2, and B3 may be determined.
Based on the above description, the embodiment of the invention converts the interest prediction problem of the user on the first type of multimedia files into the probability estimation problem in the hidden markov model. That is, given a model λ ═ (a, B, pi) and an observed sequence, the probability P (0| λ) of the occurrence of observed sequence 0 under the model λ is calculated. By substituting the observation sequence 0 into the sixth data sets M01, M02 and M03, the total attention, total interest and total propagation of the user to the first type of multimedia files in a certain time period i + M can be obtained, and the user's interest in the first type of multimedia files can be further determined.
For different application scenarios, for example, determining the interest level of the user in music content, determining the interest level of the user in game content, determining the interest level of the user in video content, and the like, the following different calculation methods may be adopted:
1) direct calculation method
In the direct calculation method, the calculation is directly performed according to a probability formula, the joint probability P (O, I | lambda) of each state sequence I and the observation sequence O is obtained by enumerating all possible state sequences I with the length of T, and then the sum of all the possible state sequences is obtained to obtain P (O | lambda), namely:
P(O|λ)＝∑IP(O|I,λ)P(I|λ)＝∑Iπi1bi1(o1)ai1i2bi2(o2)...aiT－1iTbiT(oT)。
2) forward method of calculation
Forward probability: given a model λ, the partial observation sequences are defined as O1, O2,., Ot by time t, and the probability of state qi is a forward probability, which is shown as the following equation:
αt(i)＝P(O1，O2，...，Ot，it＝qi|λ)
in the forward algorithm for the probability of an observed sequence,
the input parameters are hidden horse model lambda and observation sequence O; the output parameter is sequencing column probability P (O | lambda);
the initial value t is 1; α 1(i) ═ P (O1, i1 ═ q1| λ) ═ pi ibi (O1); wherein i is 1, 2, …, N.
For t ═ 1, 2, …, N
αt+1(i)＝[∑j＝1Nαt(j)aji]bi(ot+1)。
Obtaining: p (O | λ) ═ Σ i ═ 1N α t (i).
3) Backward algorithm
Backward probability: given λ, defining that under the condition that the state at the time T is qi, the partial observation sequence from T +1 to T is set as Ot +1, Ot +2,.. the probability of Ot is a backward probability, which is expressed by the following formula:
Βt(i)＝P(ot+1,ot+2,...,oT|it＝qi,λ)
in a backward algorithm of observation sequence probability, input parameters are a Markov model lambda and an observation sequence O; the output parameter is the observation sequence probability P (O | λ).
An initial value β t (i) ═ 1, i ═ 1, 2, ·, N;
for T ═ T-1, T-2, …, 1
β t (i) ═ Σ j ═ 1Naijbj (ot +1) β t +1 (j); wherein, i is 1, 2.
Obtaining: p (O | λ) ═ Σ i ═ 1N pi ibi (O1) β 1 (i).
An embodiment of the present invention further provides a multimedia information processing apparatus, and a composition structure of the multimedia information processing apparatus, as shown in fig. 7, includes:
an obtaining unit 301, configured to obtain data of a first type of multimedia file;
a processing unit 302 for preprocessing the data for adapting a machine learning model;
a determining unit 303, configured to determine, in the machine learning model, a user's interest level in the first type of multimedia file according to the preprocessed data.
In this embodiment of the present invention, the obtaining unit 301 is configured to obtain data of the first-class multimedia file from at least two users in a first time interval.
In an embodiment of the present invention, the data includes at least one of:
usage data of the first type of multimedia file;
attribute data of at least two multimedia files of the first class of multimedia files.
In this embodiment of the present invention, the processing unit 302 is configured to divide the usage data in the first time interval into at least four first data sets, and a time interval between every two first data sets is a fixed value;
performing first merging processing on the first data set according to a first compression ratio to obtain at least two second data sets;
and carrying out second combination processing on the second data set according to a second compression ratio to obtain at least one third data set.
In this embodiment of the present invention, the processing unit 302 is configured to divide the third data set into at least two fourth data sets based on a multimedia file ID;
dividing the fourth data set into fourth data subsets with corresponding dimensions based on user operations on the first multimedia files with different dimensions;
and processing the fourth data subset to acquire the attribute corresponding to the fourth data subset.
In this embodiment of the present invention, the processing unit 302 is configured to perform reverse merge processing on the fourth data subset, to obtain a fifth data set representing a behavior attribute of the user to the first multimedia file at a first time point, and a sixth data set representing a behavior attribute of the user to the first multimedia file at a second time point.
In an embodiment of the present invention, the determining unit 303 is configured to determine the machine learning model based on the fifth data set;
and determining the interest degree of the user in the first type of multimedia files based on the sixth data set and the machine learning model.
An embodiment of the present invention further provides a multimedia information processing apparatus, which includes a processor and a memory for storing a computer program capable of running on the processor, wherein,
the processor is used for executing the steps of the multimedia information processing method when the computer program is run.
The embodiment of the invention also provides a storage medium, which stores an executable program, and the executable program is executed by a processor to realize the multimedia information processing method.
In the embodiment of the invention, the preposed parameters (preset parameters) related in the hidden Markov model can be dynamically optimized according to the deviation of the predicted result and the actual result, so that the accuracy of determining the interest degree of the user in the first type of video files is dynamically improved.
In addition, the embodiment of the invention abstractly quantizes the interest degree of the user to the multimedia file so as to perform probabilistic calculation processing on the data, and determines the interest degree of the user to the multimedia file by combining the hidden Markov model, thereby improving the accuracy of the determined interest degree.
It should be noted that, in the embodiment of the present invention, the multimedia file may be various types of files such as advertisements, music, games, and animations. The multimedia file processing method provided by the embodiment of the invention can also be applied to state prediction or analysis in the fields of Internet, industry, servers, information technology and the like.
The above description is only exemplary of the present invention and should not be taken as limiting the scope of the present invention, and any modifications, equivalents, improvements, etc. that are within the spirit and principle of the present invention should be included in the present invention.
Claims (9)
1. A method for processing multimedia information, the method comprising:
acquiring data of a first type of multimedia file;
preprocessing the data for adapting a machine learning model;
in the machine learning model, determining the interest degree of the user in the first type of multimedia files according to the preprocessed data;
the pre-processing the data for adapting a machine learning model comprises:
dividing the use data in a first time interval in the data into at least four first data sets, wherein the time interval between every two first data sets is a fixed value;
performing first merging processing on the first data set by utilizing an inverse power finger sampling data compression mode according to a first compression ratio to obtain at least two second data sets;
and performing second merging processing on the second data set by utilizing an inverse power finger sampling data compression mode according to a second compression ratio to obtain at least one third data set.
2. The method of claim 1, wherein the obtaining data for the first type of multimedia file comprises:
and acquiring data of the first-class multimedia files from at least two users in a first time interval.
3. The method according to claim 1 or 2, wherein the data comprises at least one of:
usage data of the first type of multimedia file;
attribute data of at least two multimedia files of the first class of multimedia files.
4. The method of claim 1, wherein the pre-processing the data to adapt a machine learning model comprises:
dividing the third data set into at least two fourth data sets based on a multimedia file Identification (ID);
dividing the fourth data set into fourth data subsets with corresponding dimensions based on user operations on the first multimedia files with different dimensions;
and processing the fourth data subset to acquire the attribute corresponding to the fourth data subset.
5. The method of claim 4, wherein the pre-processing the data to adapt a machine learning model comprises:
and carrying out reverse combination processing on the fourth data subset to obtain a fifth data set representing the behavior attribute of the first-class multimedia file at a first time point of the user and a sixth data set representing the behavior attribute of the first-class multimedia file at a second time point of the user.
6. The method of claim 5, wherein determining the interest level of the user in the first type of multimedia file according to the preprocessed data comprises:
determining the machine learning model based on the fifth dataset;
and determining the interest degree of the user in the first type of multimedia files based on the sixth data set and the machine learning model.
7. A multimedia information processing apparatus, characterized in that the apparatus comprises:
the acquiring unit is used for acquiring data of the first type of multimedia files;
a processing unit for preprocessing the data for adapting a machine learning model;
the determining unit is used for determining the interest degree of the user in the first type of multimedia files according to the preprocessed data in the machine learning model;
the processing unit is configured to:
dividing the use data in a first time interval in the data into at least four first data sets, wherein the time interval between every two first data sets is a fixed value;
performing first merging processing on the first data set by utilizing an inverse power finger sampling data compression mode according to a first compression ratio to obtain at least two second data sets;
and performing second merging processing on the second data set by utilizing an inverse power finger sampling data compression mode according to a second compression ratio to obtain at least one third data set.
8. A multimedia information processing apparatus comprising a processor and a memory for storing a computer program capable of running on the processor, wherein,
the processor is configured to execute the steps of the multimedia information processing method according to any one of claims 1 to 6 when the computer program is executed.
9. A storage medium storing an executable program which, when executed by a processor, implements the multimedia information processing method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811627152.9A CN109710778B (en) | 2018-12-28 | 2018-12-28 | Multimedia information processing method, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811627152.9A CN109710778B (en) | 2018-12-28 | 2018-12-28 | Multimedia information processing method, device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109710778A CN109710778A (en) | 2019-05-03 |
| CN109710778B true CN109710778B (en) | 2021-06-04 |
Family
ID=66257960
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811627152.9A Active CN109710778B (en) | 2018-12-28 | 2018-12-28 | Multimedia information processing method, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109710778B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111832036A (en) * | 2019-07-16 | 2020-10-27 | 北京骑胜科技有限公司 | Data processing method, data processing device, storage medium and electronic equipment |
| CN112148716B (en) * | 2020-10-26 | 2024-09-10 | 腾讯科技（深圳）有限公司 | Data processing method and device for multimedia file, storage medium and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1761997A (en) * | 2003-03-19 | 2006-04-19 | 英特尔公司 | A coupled hidden markov model (CHMM) for continuous audiovisual speech recognition |
| CN105868318A (en) * | 2016-03-25 | 2016-08-17 | 海信集团有限公司 | Multimedia data type prediction method and device |
| WO2018053438A1 (en) * | 2016-09-18 | 2018-03-22 | Newvoicemedia Us Inc. | System and method for optimizing communications using reinforcement learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9843837B2 (en) * | 2015-08-03 | 2017-12-12 | At&T Intellectual Property I, L.P. | Cross-platform analysis |
-
2018
- 2018-12-28 CN CN201811627152.9A patent/CN109710778B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1761997A (en) * | 2003-03-19 | 2006-04-19 | 英特尔公司 | A coupled hidden markov model (CHMM) for continuous audiovisual speech recognition |
| CN105868318A (en) * | 2016-03-25 | 2016-08-17 | 海信集团有限公司 | Multimedia data type prediction method and device |
| WO2018053438A1 (en) * | 2016-09-18 | 2018-03-22 | Newvoicemedia Us Inc. | System and method for optimizing communications using reinforcement learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109710778A (en) | 2019-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109460514B (en) | Method and device for pushing information | |
| US20190364123A1 (en) | Resource push method and apparatus | |
| JP7316453B2 (en) | Object recommendation method and device, computer equipment and medium | |
| US11501161B2 (en) | Method to explain factors influencing AI predictions with deep neural networks | |
| CN110266745B (en) | Information flow recommendation method, device, equipment and storage medium based on deep network | |
| US20240127058A1 (en) | Training neural networks using priority queues | |
| CN110119477B (en) | Information pushing method, device and storage medium | |
| JP2021118000A (en) | Item recommendation method, system, program, electronic equipment and storage medium | |
| CN111340221B (en) | Neural network structure sampling method and device | |
| CN110390408A (en) | Trading object prediction technique and device | |
| US12063186B2 (en) | Systems and methods for a two-tier machine learning model for generating conversational responses | |
| JP2024503774A (en) | Fusion parameter identification method and device, information recommendation method and device, parameter measurement model training method and device, electronic device, storage medium, and computer program | |
| US11811708B2 (en) | Systems and methods for generating dynamic conversational responses using cluster-level collaborative filtering matrices | |
| CN108182633B (en) | Loan data processing method, loan data processing device, loan data processing program, and computer device and storage medium | |
| CN111460384A (en) | Policy evaluation method, device and equipment | |
| US20220414343A1 (en) | Systems and methods for generating dynamic conversational responses based on historical and dynamically updated information | |
| CN113301442A (en) | Method, apparatus, medium, and program product for determining live broadcast resource | |
| CN109710778B (en) | Multimedia information processing method, device and storage medium | |
| JP2022545335A (en) | New word classification technology | |
| CN113015010A (en) | Push parameter determination method, device, equipment and computer readable storage medium | |
| WO2020211616A1 (en) | Method and device for processing user interaction information | |
| WO2023142408A1 (en) | Data processing method and method for training prediction model | |
| CN112541705B (en) | Method, device, equipment and storage medium for generating user behavior evaluation model | |
| CN113742593A (en) | Method and device for pushing information | |
| CN111949860A (en) | Method and apparatus for generating a relevance determination model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |