CN111400548A - Deep learning and Markov chain-based recommendation method and device - Google Patents

Deep learning and Markov chain-based recommendation method and device Download PDF

Info

Publication number
CN111400548A
CN111400548A CN201910000979.5A CN201910000979A CN111400548A CN 111400548 A CN111400548 A CN 111400548A CN 201910000979 A CN201910000979 A CN 201910000979A CN 111400548 A CN111400548 A CN 111400548A
Authority
CN
China
Prior art keywords
video
target
transition probability
target user
embedded vector
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910000979.5A
Other languages
Chinese (zh)
Other versions
CN111400548B (en
Inventor
张大朋
戈扬
段福高
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Mobile Communications Group Co Ltd
China Mobile Communications Ltd Research Institute
Original Assignee
China Mobile Communications Group Co Ltd
China Mobile Communications Ltd Research Institute
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by China Mobile Communications Group Co Ltd, China Mobile Communications Ltd Research Institute filed Critical China Mobile Communications Group Co Ltd
Priority to CN201910000979.5A priority Critical patent/CN111400548B/en
Publication of CN111400548A publication Critical patent/CN111400548A/en
Application granted granted Critical
Publication of CN111400548B publication Critical patent/CN111400548B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/29—Graphical models, e.g. Bayesian networks
    • G06F18/295—Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention provides a deep learning and Markov chain-based recommendation method and device, and relates to the technical field of machine learning. The method comprises the following steps: determining the transition probability of a Markov chain corresponding to a target user according to the historical behavior information of the target user watching the video; calculating a target embedding vector according to the transition probability; and obtaining a recommended target video according to the target embedded vector and the attribute information of the target user. According to the scheme of the invention, the accuracy of user recommendation and the convergence speed during training are improved.

Description

Deep learning and Markov chain-based recommendation method and device
Technical Field
The invention relates to the technical field of machine learning, in particular to a deep learning and Markov chain based recommendation method and device.
Background
With the increasing of internet data, multi-element heterogeneous data including videos, audios, images and texts contains rich user behavior information and personalized demand information, an original traditional method cannot process the data, and with the development of deep learning, people begin to introduce deep learning into a recommendation field, wherein the deep learning can learn the capability of intrinsic characteristics of a data set from a sample, and on the other hand, the deep learning performs automatic characteristic learning from the multi-element heterogeneous data, so that different data are mapped to a same hidden space, and uniform representation of the data can be obtained; the combination of deep learning and the traditional method and the complete deep learning all achieve unusual performances.
However, in the current video recommendation mode, the personalized preference of the user to the video is mainly modeled by using the historical behavior data, the user characteristics and the contextual information of the user, but the accuracy of recommendation and the convergence speed of model training are greatly reduced due to simple mechanical addition.
Disclosure of Invention
The invention aims to provide a recommendation method and device based on deep learning and a Markov chain, so as to improve the accuracy of user recommendation and the convergence speed during model training.
To achieve the above object, an embodiment of the present invention provides a deep learning and markov chain based recommendation method, including:
determining the transition probability of a Markov chain corresponding to a target user according to the historical behavior information of the target user watching the video;
calculating a target embedding vector according to the transition probability;
and obtaining a recommended target video according to the target embedded vector and the attribute information of the target user.
Wherein the historical behavior information includes at least: an identification of a video and viewing time information corresponding to the video.
The method for determining the transition probability of the Markov chain corresponding to the target user according to the historical behavior information of the target user watching the video comprises the following steps:
by the formula
Figure BDA0001933579900000021
Obtaining a transition probability pij； wherein ,
Figure BDA0001933579900000022
an identification representing the video i viewed at time t,
Figure BDA0001933579900000023
an identifier, i ∈ [1, 2.,. n ], representing the video j viewed at the next instant in time t]，j∈[1，2，...，n]N is a positive integer; if it is
Figure BDA0001933579900000024
Exist, then
Figure BDA0001933579900000025
If it is
Figure BDA0001933579900000026
Is absent, then
Figure BDA0001933579900000027
The method for determining the transition probability of the Markov chain corresponding to the target user according to the historical behavior information of the target user watching the video comprises the following steps:
by the formula
Figure BDA0001933579900000028
Obtaining the similarity sijAnd is formed by pij＝sijTo obtain a transition probability pij； wherein ,
eian embedded vector of a video i watched at the current moment; e.g. of the typejAn embedded vector of video j viewed for a time next to the current time.
The method for determining the transition probability of the Markov chain corresponding to the target user according to the historical behavior information of the target user watching the video comprises the following steps:
by the formula
Figure BDA0001933579900000029
Obtaining the Euclidean distance dijAnd is formed by pij＝dijTo obtain a transition probability pij； wherein ,
eikembedding vector e for video i viewed at the current momentiA scalar in dimension k; e.g. of the typejkAn embedded vector e of a video j viewed for a time next to the current timejScalar in dimension k.
Wherein calculating a target embedding vector according to the transition probability comprises:
by the formula
Figure BDA0001933579900000031
Obtaining a transition probability pijNormalized value S ofijAnd is formed by Pij＝SijObtaining a target transition probability Pij；
By the formula esum＝∑eiPijObtaining an embedded vector sum esum；
By the formula eavg＝esumN, obtaining a target embedded vector eavg； wherein ,
eiis the embedded vector of video i viewed at the current moment.
Obtaining a recommended target video according to the target embedded vector and the attribute information of the target user, wherein the obtaining of the recommended target video comprises the following steps:
training a neural network model through the target embedded vector and the attribute information of the target user;
predicting the watching probability and the watching duration of the target user to the video of the video library based on the trained neural network model;
and screening out the target video according to the watching probability and the watching duration.
To achieve the above object, an embodiment of the present invention provides a deep learning and markov chain based recommendation apparatus, including:
the first processing module is used for determining the transition probability of the Markov chain corresponding to a target user according to the historical behavior information of the target user watching the video;
the second processing module is used for calculating a target embedded vector according to the transition probability;
and the third processing module is used for obtaining the recommended target video according to the target embedded vector and the attribute information of the target user.
Wherein the historical behavior information includes at least: an identification of a video and viewing time information corresponding to the video.
Wherein the first processing module is further configured to:
by the formula
Figure BDA0001933579900000032
Obtaining a transition probability pij； wherein ,
Figure BDA0001933579900000033
an identification representing the video i viewed at time t,
Figure BDA0001933579900000034
an identifier, i ∈ [1, 2.,. n ], representing the video j viewed at the next instant in time t]，j∈[1，2，...，n]N is a positive integer; if it is
Figure BDA0001933579900000041
Exist, then
Figure BDA0001933579900000042
If it is
Figure BDA0001933579900000043
Is absent, then
Figure BDA0001933579900000044
Wherein the first processing module is further configured to:
by the formula
Figure BDA0001933579900000045
Obtaining the similarity sijAnd is formed by pij＝sijTo obtain a transition probability pij； wherein ,
eian embedded vector of a video i watched at the current moment; e.g. of the typejAn embedded vector of video j viewed for a time next to the current time.
Wherein the first processing module is further configured to:
by the formula
Figure BDA0001933579900000046
Obtaining the Euclidean distance dijAnd is formed by pij＝dijTo obtain a transition probability pij； wherein ,
eikembedding vector e for video i viewed at the current momentiA scalar in dimension k; e.g. of the typejkAn embedded vector e of a video j viewed for a time next to the current timejScalar in dimension k.
Wherein the second processing module comprises:
a first processing submodule for passing through a formula
Figure BDA0001933579900000047
Obtaining a transition probability pijNormalized value S ofijAnd is formed by Pij＝SijObtaining a target transition probability Pij；
A second processing submodule for passing through the formula esum＝∑eiPijObtaining an embedded vector sum esum；
A third processing submodule forBy the formula eavg＝esumN, obtaining a target embedded vector eavg； wherein ,
eiis the embedded vector of video i viewed at the current moment.
Wherein the third processing module comprises:
the training submodule is used for training a neural network model through the target embedded vector and the attribute information of the target user;
the prediction submodule is used for predicting the watching probability and the watching duration of the target user to the video of the video library based on the trained neural network model;
and the screening submodule is used for screening out the target video according to the watching probability and the watching duration.
To achieve the above object, an embodiment of the present invention provides a recommendation device, including a transceiver, a memory, a processor, and a computer program stored on the memory and executable on the processor; the processor, when executing the computer program, implements the deep learning and markov chain based recommendation method as described above.
To achieve the above object, an embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored, which, when being executed by a processor, implements the steps in the deep learning and markov chain based recommendation method as described above.
The technical scheme of the invention has the following beneficial effects:
according to the method, a target user watches the historical behavior information of the video, the transition probability of the Markov chain corresponding to the target user is determined, then a target embedding vector is further obtained according to the determined transition probability, and then the recommended target video is obtained by combining the target embedding vector and the attribute information of the target user. Therefore, the historical behavior of the user watching the video is regarded as a Markov chain, the target embedding vector is calculated according to the transition probability, noise interference can be reduced in the screening process, the pertinence of the user is enhanced, and the accuracy of recommending the target video and the convergence speed of the model during training are improved.
Drawings
Fig. 1 is a flowchart of a deep learning and markov chain based recommendation method according to an embodiment of the present invention;
FIG. 2 is a second flowchart of a deep learning and Markov chain based recommendation method according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of an embodiment of the present invention;
FIG. 4 is a schematic partial detail view of FIG. 3;
FIG. 5 is a block diagram of a deep learning and Markov chain based recommendation device in accordance with an embodiment of the present invention;
fig. 6 is a block diagram of a recommendation device according to another embodiment of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantages of the present invention more apparent, the following detailed description is given with reference to the accompanying drawings and specific embodiments.
As shown in fig. 1, a deep learning and markov chain based recommendation method according to an embodiment of the present invention includes:
step 101, determining the transition probability of a Markov chain corresponding to a target user according to the historical behavior information of the target user watching a video;
102, calculating a target embedded vector according to the transition probability;
and 103, obtaining a recommended target video according to the target embedded vector and the attribute information of the target user.
According to the step 101-103, in the method of the embodiment of the present invention, firstly, the target user watches the historical behavior information of the video, the transition probability of the markov chain corresponding to the target user is determined, then, the target embedding vector is further obtained according to the determined transition probability, and then, the recommended target video is obtained by combining the target embedding vector and the attribute information of the target user. Therefore, the historical behavior of the user watching the video is regarded as a Markov chain, the target embedding vector is calculated according to the transition probability, noise interference can be reduced in the screening process, the pertinence of the user is enhanced, and the accuracy of recommending the target video and the convergence speed of the model during training are improved.
Wherein the attribute information includes but is not limited to: user gender, user age, type of video the user is interested in, etc.
Optionally, the historical behavior information includes at least: an identification of a video and viewing time information corresponding to the video.
Here, the identifier of the video is a unique identifier corresponding to the video in the video library, and the viewing time information corresponding to the video includes at least one of the following information: viewing start time, viewing end time, viewing duration, and ordering among all viewed videos (i.e., the ones viewed).
In this embodiment, optionally, step 101 includes:
by the formula
Figure BDA0001933579900000061
Obtaining a transition probability pij； wherein ,
Figure BDA0001933579900000062
an identification representing the video i viewed at time t,
Figure BDA0001933579900000063
an identifier, i ∈ [1, 2.,. n ], representing the video j viewed at the next instant in time t]，j∈[1，2，...，n]N is a positive integer; if it is
Figure BDA0001933579900000064
Exist, then
Figure BDA0001933579900000065
If it is
Figure BDA0001933579900000066
Is absent, then
Figure BDA0001933579900000067
Here, the probability p of transition will be passedijIs calculated by
Figure BDA0001933579900000071
Computing p in conjunction with historical behavior information of a target user's viewed videoij. For example, the historical behavior information indicates that the target user viewed video 1 at time t and viewed video 2 at the time next to time t (i.e., time t + 1), and therefore
Figure BDA0001933579900000072
Figure BDA0001933579900000073
Still alternatively, step 101 includes:
by the formula
Figure BDA0001933579900000074
Obtaining the similarity sijAnd is formed by pij＝sijTo obtain a transition probability pij； wherein ,
eian embedded vector of a video i watched at the current moment; e.g. of the typejAn embedded vector for a video viewed at a time next to the current time.
Here, the video is embedded vectorized based on the Markov chain, and the embedded vector e of the video i viewed at the current time isiEmbedded vector e of video viewed at the next instant of the current instantjBy the formula
Figure BDA0001933579900000075
Calculating the similarity sijAnd the similarity is taken as the transition probability.
Still alternatively, step 101 includes:
by the formula
Figure BDA0001933579900000076
Obtaining the Euclidean distance dijAnd is formed by pij＝dijTo obtainTransition probability pij； wherein ,
eikembedding vector e for video i viewed at the current momentiA scalar in dimension k; e.g. of the typejkAn embedded vector e of a video j viewed for a time next to the current timejScalar in dimension k.
Thus, the embedded vector e of the video i viewed at the current momentiScalar e in dimension kikAnd an embedded vector e of a video j viewed at a time next to the previous timejScalar e in dimension kjkBy the formula
Figure BDA0001933579900000077
Calculated Euclidean distance dijAnd euclidean distance as the transition probability.
Thereafter, to further ensure the accuracy of the recommendation, step 102 includes:
by the formula
Figure BDA0001933579900000081
Obtaining a transition probability pijNormalized value S ofijAnd is formed by Pij＝SijObtaining a target transition probability Pij；
By the formula esum＝∑eiPijObtaining an embedded vector sum esum；
By the formula eavg＝esumN, obtaining a target embedded vector eavg； wherein ,
eiis the embedded vector of video i viewed at the current moment.
Here, the transition probabilities obtained in step 101 are first normalized to obtain target transition probabilities, then embedded vector sums are calculated, and then average values are taken to obtain target embedded vectors capable of better removing interference. E.g. esum＝∑eiPij＝e1P12+e2P23+…+eiPij+3+enWhen i is 1,2, …, n-1; j is 1,2, …, n.
Optionally, as shown in fig. 2, step 103 includes:
step 201, training a neural network model through the target embedded vector and the attribute information of the target user;
step 202, predicting the watching probability and the watching duration of the target user to the video of the video library based on the trained neural network model;
and step 203, screening out the target video according to the watching probability and the watching duration.
Thus, after the target embedded vector is obtained in step 102, the neural network model can be trained by combining the target embedded vector and the attribute information of the target user, the viewing probability and the viewing duration of each video in the video library by the target user are predicted, and then the target video is screened out according to the viewing probability and the viewing duration. The target video screening according to the watching probability and the watching duration is executed based on a preset rule, and the preset rule can be set as follows:
and sorting the videos in the candidate set from long to short according to the watching probability, and selecting the first n videos as target videos.
Certainly, the preset rule is not limited to the above manner, and may be defined by a system or a user, or may be sorted according to the viewing time from long to short, the first n videos are selected as a candidate set, then the videos in the candidate set are sorted according to the viewing probability from large to small, and the first m videos are selected as target videos, which are not listed one by one. Wherein the setting of m and n is done by the system or the user.
In addition, in this embodiment, the training is performed in a Batch mode, the prediction is input according to the user context, and some skills may be used in the training, such as initialization of weights by Xavier/he, dropout, gradient pruning, BN (Batch Normalization).
Thus, the method of the embodiment of the present invention, as shown in fig. 3 and 4, is to video I at the embedding layern，In-1，…，Im+1After embedding vectorization, Markov chains are utilizedAfter the vector is calculated, it can be associated with other user attribute information such as I1，I2，…，ImAnd connecting, performing neural network model training on a neural network RE L U layer, and finally screening out the target video through prediction.
In addition, for the recommendation of search, after the first y searched videos are embedded into vectorization, the Markov chain is used for averaging the videos to be used as the last video watching behavior of the video watching behaviors, then the Markov chain is used for calculating vectors, and then the vectors are connected with other user attribute information to carry out neural network model training, and finally the target video is screened out through prediction.
In summary, in the method according to the embodiment of the present invention, a target user first watches historical behavior information of a video, a transition probability of a markov chain corresponding to the target user is determined, then a target embedding vector is further obtained according to the determined transition probability, and then a recommended target video is obtained by combining the target embedding vector and attribute information of the target user. Therefore, the historical behavior of the user watching the video is regarded as a Markov chain, the target embedding vector is calculated according to the transition probability, noise interference can be reduced in the screening process, the pertinence of the user is enhanced, and the accuracy of recommending the target video and the convergence speed of the model during training are improved.
As shown in fig. 5, a deep learning and markov chain based recommendation apparatus according to an embodiment of the present invention includes:
the first processing module 501 is configured to determine, according to historical behavior information of a target user watching a video, a transition probability of a markov chain corresponding to the target user;
a second processing module 502, configured to calculate a target embedding vector according to the transition probability;
and a third processing module 503, configured to obtain a recommended target video according to the target embedded vector and the attribute information of the target user.
Wherein the historical behavior information includes at least: an identification of a video and viewing time information corresponding to the video.
Wherein the first processing module is further configured to:
by the formula
Figure BDA0001933579900000101
Obtaining a transition probability pij； wherein ,
Figure BDA0001933579900000102
an identification representing the video i viewed at time t,
Figure BDA0001933579900000103
an identifier, i ∈ [1, 2.,. n ], representing the video j viewed at the next instant in time t]，j∈[1，2，...，n]N is a positive integer; if it is
Figure BDA0001933579900000104
Exist, then
Figure BDA0001933579900000105
If it is
Figure BDA0001933579900000106
Is absent, then
Figure BDA0001933579900000107
Wherein the first processing module is further configured to:
by the formula
Figure BDA0001933579900000108
Obtaining the similarity sijAnd is formed by pij＝sijTo obtain a transition probability pij； wherein ,
eian embedded vector of a video i watched at the current moment; e.g. of the typejAn embedded vector of video j viewed for a time next to the current time.
Wherein the first processing module is further configured to:
by the formula
Figure BDA0001933579900000109
Obtaining the Euclidean distance dijAnd is formed by pij＝dijTo obtain a transition probability pij； wherein ,
eikembedding vector e for video i viewed at the current momentiA scalar in dimension k; e.g. of the typejkAn embedded vector e of a video j viewed for a time next to the current timejScalar in dimension k.
Wherein the second processing module comprises:
a first processing submodule for passing through a formula
Figure BDA00019335799000001010
Obtaining a transition probability pijNormalized value S ofijAnd is formed by Pij＝SijObtaining a target transition probability Pij；
A second processing submodule for passing through the formula esum＝∑eiPijObtaining an embedded vector sum esum；
A third processing submodule for passing through the formula eavg＝esumN, obtaining a target embedded vector eavg； wherein ,eiIs the embedded vector of video i viewed at the current moment.
Wherein the third processing module comprises:
the training submodule is used for training a neural network model through the target embedded vector and the attribute information of the target user;
the prediction submodule is used for predicting the watching probability and the watching duration of the target user to the video of the video library based on the trained neural network model;
and the screening submodule is used for screening out the target video according to the watching probability and the watching duration.
According to the recommendation device, firstly, historical behavior information of videos watched by a target user is used, the transition probability of the Markov chain corresponding to the target user is determined, then a target embedding vector is further obtained according to the determined transition probability, and then the recommended target videos are obtained by combining the target embedding vector and the attribute information of the target user. Therefore, the historical behavior of the user watching the video is regarded as a Markov chain, the target embedding vector is calculated according to the transition probability, noise interference can be reduced in the screening process, the pertinence of the user is enhanced, and the accuracy of recommending the target video and the convergence speed of the model during training are improved.
It should be noted that the device is a device to which the deep learning and markov chain based recommendation method is applied, and the implementation manner of the method embodiment is applicable to the device and can achieve the same technical effect.
A recommendation device according to another embodiment of the present invention, as shown in fig. 6, includes a transceiver 610, a memory 620, a processor 600, and a computer program stored on the memory 620 and operable on the processor 600; the processor 600, when executing the computer program, implements the above deep learning and markov chain based recommendation method applied to the recommendation device.
The transceiver 610 is used for receiving and transmitting data under the control of the processor 600.
Where in fig. 6, the bus architecture may include any number of interconnected buses and bridges, with various circuits being linked together, particularly one or more processors represented by processor 600 and memory represented by memory 620. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, power management circuits, and the like, which are well known in the art, and therefore, will not be described any further herein. The bus interface provides an interface. The transceiver 610 may be a number of elements including a transmitter and a receiver that provide a means for communicating with various other apparatus over a transmission medium. The processor 600 is responsible for managing the bus architecture and general processing, and the memory 620 may store data used by the processor 600 in performing operations.
A computer-readable storage medium according to an embodiment of the present invention stores thereon a computer program, and when the computer program is executed by a processor, the steps in the deep learning and markov chain based recommendation method described above are implemented, and the same technical effects can be achieved. The computer-readable storage medium may be a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk.
It is further noted that many of the functional units described in this specification have been labeled as modules, in order to more particularly emphasize their implementation independence.
In embodiments of the present invention, modules may be implemented in software for execution by various types of processors. An identified module of executable code may, for instance, comprise one or more physical or logical blocks of computer instructions which may, for instance, be constructed as an object, procedure, or function. Nevertheless, the executables of an identified module need not be physically located together, but may comprise disparate instructions stored in different bits which, when joined logically together, comprise the module and achieve the stated purpose for the module.
Indeed, a module of executable code may be a single instruction, or many instructions, and may even be distributed over several different code segments, among different programs, and across several memory devices. Likewise, operational data may be identified within the modules and may be embodied in any suitable form and organized within any suitable type of data structure. The operational data may be collected as a single data set, or may be distributed over different locations including over different storage devices, and may exist, at least partially, merely as electronic signals on a system or network.
When a module can be implemented by software, considering the level of existing hardware technology, a module that can be implemented by software can build corresponding hardware circuits including conventional very large scale integration (V L SI) circuits or gate arrays and existing semiconductors such as logic chips, transistors, or other discrete components to implement corresponding functions, without considering the cost.
The exemplary embodiments described above are described with reference to the drawings, and many different forms and embodiments of the invention may be made without departing from the spirit and teaching of the invention, therefore, the invention is not to be construed as limited to the exemplary embodiments set forth herein. Rather, these exemplary embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art. In the drawings, the size and relative sizes of elements may be exaggerated for clarity. The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. Unless otherwise indicated, a range of values, when stated, includes the upper and lower limits of the range and any subranges therebetween.
While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims (16)

1. A deep learning and Markov chain based recommendation method is characterized by comprising the following steps:
determining the transition probability of a Markov chain corresponding to a target user according to the historical behavior information of the target user watching the video;
calculating a target embedding vector according to the transition probability;
and obtaining a recommended target video according to the target embedded vector and the attribute information of the target user.
2. The method of claim 1, wherein the historical behavior information comprises at least: an identification of a video and viewing time information corresponding to the video.
3. The method of claim 1, wherein determining the transition probability of the Markov chain corresponding to the target user according to the historical behavior information of the target user watching the video comprises:
by the formula
Figure FDA0001933579890000011
Obtaining a transition probability pij； wherein ,
Figure FDA0001933579890000012
an identification representing the video i viewed at time t,
Figure FDA0001933579890000013
an identifier, i ∈ [1, 2.,. n ], representing the video j viewed at the next instant in time t]，j∈[1，2，...，n]N is a positive integer; if it is
Figure FDA0001933579890000014
Exist, then
Figure FDA0001933579890000015
If it is
Figure FDA0001933579890000016
Is absent, then
Figure FDA0001933579890000017
4. The method of claim 1, wherein determining the transition probability of the Markov chain corresponding to the target user according to the historical behavior information of the target user watching the video comprises:
by the formula
Figure FDA0001933579890000018
Obtaining the similarity sijAnd is formed by pij＝sijTo obtain a transition probability pij； wherein ,
eian embedded vector of a video i watched at the current moment; e.g. of the typejAn embedded vector of video j viewed for a time next to the current time.
5. The method of claim 1, wherein determining the transition probability of the Markov chain corresponding to the target user according to the historical behavior information of the target user watching the video comprises:
by the formula
Figure FDA0001933579890000021
Obtaining the Euclidean distance dijAnd is formed by pij＝dijTo obtain a transition probability pij； wherein ,
eikembedding vector e for video i viewed at the current momentiA scalar in dimension k; e.g. of the typejkAn embedded vector e of a video j viewed for a time next to the current timejScalar in dimension k.
6. The method of claim 1, wherein computing a target embedding vector based on the transition probabilities comprises:
by the formula
Figure FDA0001933579890000022
Obtaining a transition probability pijNormalized value S ofijAnd is formed by Pij＝SijObtaining a target transition probability Pij；
By the formula esum＝∑eiPijObtaining an embedded vector sum esum；
By the formula eavg＝esumN, obtaining a target embedded vector eavg； wherein ,
eiis the embedded vector of video i viewed at the current moment.
7. The method of claim 1, wherein obtaining the recommended target video according to the target embedding vector and the attribute information of the target user comprises:
training a neural network model through the target embedded vector and the attribute information of the target user;
predicting the watching probability and the watching duration of the target user to the video of the video library based on the trained neural network model;
and screening out the target video according to the watching probability and the watching duration.
8. A deep learning and markov chain based recommendation device, comprising:
the first processing module is used for determining the transition probability of the Markov chain corresponding to a target user according to the historical behavior information of the target user watching the video;
the second processing module is used for calculating a target embedded vector according to the transition probability;
and the third processing module is used for obtaining the recommended target video according to the target embedded vector and the attribute information of the target user.
9. The recommendation device of claim 8, wherein the historical behavior information comprises at least: an identification of a video and viewing time information corresponding to the video.
10. The recommendation device of claim 8, wherein the first processing module is further configured to:
by the formula
Figure FDA0001933579890000031
Obtaining a transition probability pij； wherein ,
Figure FDA0001933579890000032
an identification representing the video i viewed at time t,
Figure FDA0001933579890000033
an identifier, i ∈ [1, 2.,. n ], representing the video j viewed at the next instant in time t]，j∈[1，2，...，n]N is a positive integer; if it is
Figure FDA0001933579890000034
Exist, then
Figure FDA0001933579890000035
If it is
Figure FDA0001933579890000036
Is absent, then
Figure FDA0001933579890000037
11. The recommendation device of claim 8, wherein the first processing module is further configured to:
by the formula
Figure FDA0001933579890000038
Obtaining the similarity sijAnd is formed by pij＝sijTo obtain a transition probability pij； wherein ,
eian embedded vector of a video i watched at the current moment; e.g. of the typejAn embedded vector of video j viewed for a time next to the current time.
12. The recommendation device of claim 8, wherein the first processing module is further configured to:
by the formula
Figure FDA0001933579890000039
Obtaining the Euclidean distance dijAnd is formed by pij＝dijTo obtain a transition probability pij； wherein ,
eikembedding vector e for video i viewed at the current momentiA scalar in dimension k; e.g. of the typejkAn embedded vector e of a video j viewed for a time next to the current timejScalar in dimension k.
13. The recommendation device of claim 8, wherein the second processing module comprises:
a first processing submodule for passing through a formula
Figure FDA00019335798900000310
Obtaining a transition probability pijNormalized value S ofijAnd is formed by Pij＝SijObtaining a target transition probability Pij；
A second processing submodule for passing through the formula esum＝∑eiPij(ii) a (i-1, 2, n-1; j-1, 2, … n) to obtain an embedded vector sum esum；
A third processing submodule for passing through the formula eavg＝esumN, obtaining a target embedded vector eavg； wherein ,
eiis the embedded vector of video i viewed at the current moment.
14. The recommendation device of claim 8, wherein the third processing module comprises:
the training submodule is used for training a neural network model through the target embedded vector and the attribute information of the target user;
the prediction submodule is used for predicting the watching probability and the watching duration of the target user to the video of the video library based on the trained neural network model;
and the screening submodule is used for screening out the target video according to the watching probability and the watching duration.
15. A recommendation device comprising a transceiver, a memory, a processor and a computer program stored on the memory and executable on the processor; characterized in that the processor, when executing the computer program, implements the deep learning and markov chain based recommendation method according to any one of claims 1 to 7.
16. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the deep learning and markov chain based recommendation method according to any one of claims 1 to 7.
CN201910000979.5A 2019-01-02 2019-01-02 Recommendation method and device based on deep learning and Markov chain Active CN111400548B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910000979.5A CN111400548B (en) 2019-01-02 2019-01-02 Recommendation method and device based on deep learning and Markov chain

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910000979.5A CN111400548B (en) 2019-01-02 2019-01-02 Recommendation method and device based on deep learning and Markov chain

Publications (2)

Publication Number Publication Date
CN111400548A true CN111400548A (en) 2020-07-10
CN111400548B CN111400548B (en) 2023-09-22

Family

ID=71433853

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910000979.5A Active CN111400548B (en) 2019-01-02 2019-01-02 Recommendation method and device based on deep learning and Markov chain

Country Status (1)

Country Link
CN (1) CN111400548B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111159473A (en) * 2019-12-25 2020-05-15 天津中科智能识别产业技术研究院有限公司 Deep learning and Markov chain based connection recommendation method
CN111881355A (en) * 2020-07-28 2020-11-03 北京深演智能科技股份有限公司 Object recommendation method and device, storage medium and processor
CN112259193A (en) * 2020-10-09 2021-01-22 江苏大学 Drug-dropping state prediction method based on Markov chain
CN112801700A (en) * 2021-01-29 2021-05-14 北京达佳互联信息技术有限公司 Virtual object changing method and device, electronic device and storage medium
CN113312512A (en) * 2021-06-10 2021-08-27 北京百度网讯科技有限公司 Training method, recommendation device, electronic equipment and storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140267301A1 (en) * 2013-03-14 2014-09-18 Canon Kabushiki Kaisha Systems and methods for feature fusion
CN107633100A (en) * 2017-10-23 2018-01-26 苏州大学 A kind of point of interest based on incorporation model recommends method and device
WO2018112696A1 (en) * 2016-12-19 2018-06-28 深圳大学 Content pushing method and content pushing system

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140267301A1 (en) * 2013-03-14 2014-09-18 Canon Kabushiki Kaisha Systems and methods for feature fusion
WO2018112696A1 (en) * 2016-12-19 2018-06-28 深圳大学 Content pushing method and content pushing system
CN107633100A (en) * 2017-10-23 2018-01-26 苏州大学 A kind of point of interest based on incorporation model recommends method and device

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
杨安驹;杨云;周嫒嫒;闵玉涓;秦怡;: "基于隐马尔科夫模型的融合推荐算法" *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111159473A (en) * 2019-12-25 2020-05-15 天津中科智能识别产业技术研究院有限公司 Deep learning and Markov chain based connection recommendation method
CN111881355A (en) * 2020-07-28 2020-11-03 北京深演智能科技股份有限公司 Object recommendation method and device, storage medium and processor
CN111881355B (en) * 2020-07-28 2023-03-10 北京深演智能科技股份有限公司 Object recommendation method and device, storage medium and processor
CN112259193A (en) * 2020-10-09 2021-01-22 江苏大学 Drug-dropping state prediction method based on Markov chain
CN112801700A (en) * 2021-01-29 2021-05-14 北京达佳互联信息技术有限公司 Virtual object changing method and device, electronic device and storage medium
CN113312512A (en) * 2021-06-10 2021-08-27 北京百度网讯科技有限公司 Training method, recommendation device, electronic equipment and storage medium
CN113312512B (en) * 2021-06-10 2023-10-31 北京百度网讯科技有限公司 Training method, recommending device, electronic equipment and storage medium

Also Published As

Publication number Publication date
CN111400548B (en) 2023-09-22

Similar Documents

Publication Publication Date Title
CN111400548A (en) Deep learning and Markov chain-based recommendation method and device
US9990558B2 (en) Generating image features based on robust feature-learning
CN103299324B (en) Potential son is used to mark the mark learnt for video annotation
US9965717B2 (en) Learning image representation by distilling from multi-task networks
CN110717099B (en) Method and terminal for recommending film
CN110390033A (en) Training method, device, electronic equipment and the storage medium of image classification model
CN111046275A (en) User label determining method and device based on artificial intelligence and storage medium
CN113011172B (en) Text processing method, device, computer equipment and storage medium
CN110990600A (en) Multimedia file recommendation method, multimedia file recommendation device, multimedia file parameter adjustment device, multimedia file recommendation medium and electronic equipment
CN108959550B (en) User focus mining method, device, equipment and computer readable medium
CN111159473A (en) Deep learning and Markov chain based connection recommendation method
CN116955591A (en) Recommendation language generation method, related device and medium for content recommendation
CN112100509B (en) Information recommendation method, device, server and storage medium
CN112328881A (en) Article recommendation method and device, terminal device and storage medium
CN110019809B (en) Classification determination method and device and network equipment
CN115878891A (en) Live content generation method, device, equipment and computer storage medium
CN117216361A (en) Recommendation method, recommendation device, electronic equipment and computer readable storage medium
CN109063934B (en) Artificial intelligence-based combined optimization result obtaining method and device and readable medium
CN118132733B (en) Test question retrieval method, system, storage medium and electronic equipment
CN118230224B (en) Label scoring method, label scoring model training method and device
CN118171208B (en) Multi-mode multi-label association classification method, system, storage medium and computer
CN118799633A (en) Image classification method, device and readable storage medium
CN114764457A (en) Video processing method and device, electronic equipment and readable storage medium
CN114443965A (en) Model training and content recall methods, media, apparatuses, and computing devices
CN117951333A (en) Training method of value model of video category, video searching method and device

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