CN110442788A - A kind of information recommendation method and device - Google Patents
A kind of information recommendation method and device Download PDFInfo
- Publication number
- CN110442788A CN110442788A CN201910665176.1A CN201910665176A CN110442788A CN 110442788 A CN110442788 A CN 110442788A CN 201910665176 A CN201910665176 A CN 201910665176A CN 110442788 A CN110442788 A CN 110442788A
- Authority
- CN
- China
- Prior art keywords
- sample
- information
- user
- users
- historical behavior
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- 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
- G06N3/047—Probabilistic or stochastic 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)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A kind of information recommendation method and device provided in an embodiment of the present invention, according to the historical behavior of the first user, define the user identifier of the first user, according to the user identifier of the first user, information to be recommended and probabilistic forecasting network model trained in advance, determine that the first user treats the interested probability of recommendation information, according to obtained probability and information to be recommended, to first user's recommendation information.Based on above-mentioned processing, the validity of information recommendation can be improved.
Description
Technical field
The present invention relates to field of computer technology, more particularly to a kind of information recommendation method and device.
Background technique
With the development of computer technology, today's society is in the epoch of information outburst, is facing massive information
When, user is difficult from massive information quickly to select oneself interested information.Recommendation side based on neural collaborative filtering
Method can be filtered massive information, determine that user can may be felt emerging in turn by the possible interested information of user
The information recommendation of interest is to user.
In the prior art, neural collaborative filtering network model can be based on to user's recommendation information.For example, can be by sample
The user identifier (user identifier can be the preset mark for indicating user's uniqueness) of user and a sample information (sample
The corresponding information of the historical behavior of user) it is used as NCF (NeuralCollaborative Filtering, neural collaborative filtering)
The input parameter of network model mixes the sample with family to the interested probability of the sample information as corresponding output parameter, to NCF
Network model is trained.In turn, to predict that a certain user treats recommendation information using trained NCF network model interested
When probability, NCF network model can be according to the user identifier of the user, information to be recommended, the user identifier of sample of users and sample
Relationship between this corresponding information of historical behavior, determines that the user treats the interested probability of recommendation information, in turn, can be with
To the information of the interested maximum probability of user recommended user.
Inventor has found that at least there are the following problems for the prior art in the implementation of the present invention:
In the prior art, if a certain user is not sample of users, that is, not including in the user identifier of sample of users has this
The user identifier of user, then NCF network model is marked according to the user of the user identifier of the user, information to be recommended, sample of users
Know the relationship between the corresponding information of historical behavior of sample, the user determined treats the interested probability of recommendation information
Accuracy it is lower.In turn, cause the validity of information recommendation in the prior art lower.
Summary of the invention
The embodiment of the present invention is designed to provide a kind of information recommendation method, to solve information recommendation in the prior art
The lower problem of validity.Specific technical solution is as follows:
In a first aspect, in order to achieve the above object, the embodiment of the invention provides a kind of information recommendation method, the methods
Include:
According to the historical behavior of the first user, the user identifier of first user is defined;
According to the user identifier of first user, information to be recommended and probabilistic forecasting network model trained in advance, really
Fixed first user is to the interested probability of information to be recommended;
According to obtained probability and the information to be recommended, Xiang Suoshu the first user recommendation information;
Wherein, the probabilistic forecasting network model is trained to obtain based on default training set, the default training
Collection includes multiple training samples, and the training sample includes the user identifier, sample information and the sample of users of sample of users
To the interested probability of the sample information, the sample information is the corresponding information of historical behavior of the sample of users, institute
The user identifier for stating sample of users is to be defined according to the historical behavior of the sample of users.
Optionally, the training sample is generated, comprising:
For each sample of users, it is ranked up according to historical behavior of the chronological order to the sample of users;
According to the first preset number historical behavior of the sample of users, the user identifier of the sample of users is defined;
According to the historical behavior after the first preset number historical behavior of the sample of users, the sample is determined
It is general to set first to the interested probability of first sample information for the sample of users for the corresponding first sample information of user
Rate;
According to the historical behavior before the first preset number historical behavior of the sample of users, the sample is determined
It is general to set second to the interested probability of the second sample information for the sample of users for corresponding second sample information of user
Rate, wherein second probability is lower than first probability;
Generation includes the training sample of the corresponding positive sample of the sample of users and negative sample, wherein the positive sample
User identifier, the first sample information and the sample of users in this including the sample of users believe the first sample
Interested probability is ceased, includes user identifier, second sample information and the sample of the sample of users in the negative sample
This user is to the interested probability of the second sample information.
Optionally, the historical behavior according to the first user defines the user identifier of first user, comprising:
It is ranked up according to historical behavior of the chronological order to first user;
According in the historical behavior of first user apart from the second preset number historical behavior that current time is nearest,
Define the user identifier of first user, wherein second preset number is no more than the user for defining the sample of users
The number of historical behavior used in identifying.
Optionally, the user identifier of first user is fixed according to the corresponding information of historical behavior of first user
Justice;
It is described according to the user identifier of first user, information to be recommended and probabilistic forecasting network mould trained in advance
Type determines first user to the interested probability of information to be recommended, comprising:
Based on the first preset algorithm in probabilistic forecasting network model trained in advance, the user of first user is marked
Knowledge is handled, and obtains the first linear characterization vector, and handle the information to be recommended, obtain second linearly characterize to
Amount;
Based on to the described first linear characterization vector carry out the processing result of process of convolution and described second linearly characterize to
Amount obtains the linear spy of the common trait of the corresponding information of historical behavior for indicating first user and the information to be recommended
Levy vector;
Based on the second preset algorithm in the probabilistic forecasting network model, the user identifier of first user is carried out
Processing, obtain the first non-linear characterization vector, and handle the information to be recommended, obtain second it is non-linear characterize to
Amount;
Based on the knot for carrying out Fusion Features to the second non-linear characterization vector described in the described first non-linear characterization vector sum
Fruit obtains indicating the non-linear of the corresponding information of historical behavior of first user and the common trait of the information to be recommended
Feature vector;
Fusion Features are carried out to nonlinear characteristic vector described in the linear character vector sum, obtain target feature vector;
Recurrence processing is carried out to the target feature vector, determines that first user is interested in the information to be recommended
Probability.
Optionally, pre- according to the user identifier of first user, information to be recommended and probability trained in advance described
Network model is surveyed, before determining first user to the interested probability of information to be recommended, the method also includes:
Based on the historical behavior of user collaborative filtering UserCF algorithm and first user, from preset multiple information
Third preset number information is determined, as the information to be recommended.
Second aspect, in order to achieve the above object, the embodiment of the invention provides a kind of information recommending apparatus, described devices
Include:
First determining module defines the user identifier of first user for the historical behavior according to the first user;
Second determining module, for according to the general of the user identifier of first user, information to be recommended and training in advance
Rate predicts network model, determines first user to the interested probability of information to be recommended;
Recommending module, for according to obtained probability and the information to be recommended, Xiang Suoshu the first user recommendation information;
Wherein, the probabilistic forecasting network model is trained to obtain based on default training set, the default training
Collection includes multiple training samples, and the training sample includes the user identifier, sample information and the sample of users of sample of users
To the interested probability of the sample information, the sample information is the corresponding information of historical behavior of the sample of users, institute
The user identifier for stating sample of users is to be defined according to the historical behavior of the sample of users.
Optionally, described device further include:
Generation module, for being directed to each sample of users, according to chronological order to the historical behavior of the sample of users
It is ranked up;
According to the first preset number historical behavior of the sample of users, the user identifier of the sample of users is defined;
According to the historical behavior after the first preset number historical behavior of the sample of users, the sample is determined
It is general to set first to the interested probability of first sample information for the sample of users for the corresponding first sample information of user
Rate;
According to the historical behavior before the first preset number historical behavior of the sample of users, the sample is determined
It is general to set second to the interested probability of the second sample information for the sample of users for corresponding second sample information of user
Rate, wherein second probability is lower than first probability;
Generation includes the training sample of the corresponding positive sample of the sample of users and negative sample, wherein the positive sample
User identifier, the first sample information and the sample of users in this including the sample of users believe the first sample
Interested probability is ceased, includes user identifier, second sample information and the sample of the sample of users in the negative sample
This user is to the interested probability of the second sample information.
Optionally, first determining module, specifically for the history according to chronological order to first user
Behavior is ranked up;
According in the historical behavior of first user apart from the second preset number historical behavior that current time is nearest,
Define the user identifier of first user, wherein second preset number is no more than the user for defining the sample of users
The number of historical behavior used in identifying.
Optionally, the user identifier of first user is fixed according to the corresponding information of historical behavior of first user
Justice;
Second determining module, specifically for based on the first pre- imputation in probabilistic forecasting network model trained in advance
Method handles the user identifier of first user, obtains the first linear characterization vector, and to the information to be recommended into
Row processing obtains the second linear characterization vector;
Based on to the described first linear characterization vector carry out the processing result of process of convolution and described second linearly characterize to
Amount obtains the linear spy of the common trait of the corresponding information of historical behavior for indicating first user and the information to be recommended
Levy vector;
Based on the second preset algorithm in the probabilistic forecasting network model, the user identifier of first user is carried out
Processing, obtain the first non-linear characterization vector, and handle the information to be recommended, obtain second it is non-linear characterize to
Amount;
Based on the knot for carrying out Fusion Features to the second non-linear characterization vector described in the described first non-linear characterization vector sum
Fruit obtains indicating the non-linear of the corresponding information of historical behavior of first user and the common trait of the information to be recommended
Feature vector;
Fusion Features are carried out to nonlinear characteristic vector described in the linear character vector sum, obtain target feature vector;
Recurrence processing is carried out to the target feature vector, determines that first user is interested in the information to be recommended
Probability.
Optionally, described device further include:
Screening module, for the historical behavior based on user collaborative filtering UserCF algorithm and first user, from pre-
If multiple information in determine third preset number information, as the information to be recommended.
In the another aspect that the present invention is implemented, in order to achieve the above object, the embodiment of the invention also provides a kind of electronics
Equipment, which is characterized in that including processor, communication interface, memory and communication bus, wherein processor, communication interface are deposited
Reservoir completes mutual communication by communication bus;
Memory, for storing computer program;
Processor, when for executing the program stored on memory, the step of realizing any of the above-described information recommendation method.
At the another aspect that the present invention is implemented, in order to achieve the above object, present invention implementation additionally provides a kind of computer
Readable storage medium storing program for executing is stored with computer program in the computer readable storage medium, and the computer program is by processor
The step of any of the above-described information recommendation method is realized when execution.
At the another aspect that the present invention is implemented, in order to achieve the above object, the embodiment of the invention also provides one kind to include
The computer program product of instruction, when run on a computer, so that computer executes any of the above-described information recommendation method.
A kind of information recommendation method and device provided in an embodiment of the present invention, according to the historical behavior of the first user, definition
The user identifier of first user, according to the user identifier of the first user, information to be recommended and probabilistic forecasting network trained in advance
Model determines that the first user treats the interested probability of recommendation information, according to obtained probability and information to be recommended, uses to first
Family recommendation information.
Based on above-mentioned processing, even if a certain user is not sample of users, due to the use according to the user identifier of the user
What the historical behavior at family determined, it is known that, it is the user identifier (i.e. the corresponding information of the historical behavior of the user) of the user, to be recommended
Correlation between information and the user identifier (i.e. the corresponding information of the historical behavior of sample of users) of sample of users is stronger, into
And probabilistic forecasting network model is determined according to the user identifier of the user, the user identifier of information to be recommended and sample of users
The user treat the interested probability of recommendation information accuracy it is higher, according to obtained probability and information to be recommended, to this
User's recommendation information can be improved the validity of information recommendation.
Certainly, implement any of the products of the present invention or method it is not absolutely required at the same reach all the above excellent
Point.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of flow chart of information recommendation method provided in an embodiment of the present invention;
Fig. 2 is a kind of exemplary flow chart of information recommendation method provided in an embodiment of the present invention;
Fig. 3 is a kind of structure chart of probabilistic forecasting network model provided in an embodiment of the present invention;
Fig. 4 is a kind of structure chart of information recommending apparatus provided in an embodiment of the present invention;
Fig. 5 is the structure chart of a kind of electronic equipment provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Below by specific embodiment, information recommendation method provided in an embodiment of the present invention is described in detail.
Referring to Fig. 1, Fig. 1 is a kind of flow chart of information recommendation method provided in an embodiment of the present invention, and this method can answer
For electronic equipment, which can be server, or terminal, the electronic equipment are used for user's recommendation
Breath.
This method may include steps of:
S101: according to the historical behavior of the first user, the user identifier of the first user is defined.
Wherein, the historical behavior of user refers to behavior of the user in a certain historical time section, for example, the historical behavior of user
It can be search behavior of the user within past three days, be also possible to watching behavior or user of the user within past two days
In past intraday splitting glass opaque, but it is not limited to this.
Illustratively, if a certain user has viewed video A, video B, the history row of the user in past three days
To include: viewing video A, watching video B.
In addition, the historical behavior of the user is corresponding if the historical behavior of a certain user is the search behavior of the user
Information can for the user search the relevant information of keyword, for example, the user has searched for star A, then star A searches for this
The keyword of Suo Hangwei, the corresponding information of the historical behavior of the user can be the relevant information of star A, the relevant letter of star A
Breath can be the movie and television play that star A is performed, or the relevant topic of star A, but it is not limited to this.
In inventive embodiments, the historical behavior of the available user of electronic equipment, according to the history row of the user of acquisition
For corresponding information, the user identifier of user is determined.
In a kind of implementation, if the historical behavior of a certain user is one, electronic equipment can directly be gone through this
User identifier of the corresponding information of history behavior as the user.
Illustratively, the historical behavior of user can be the watching behavior of user, if a certain user was at past one day
In have viewed video A, electronic equipment, can be directly by the history after the historical behavior (i.e. viewing video A) for determining the user
The corresponding information of behavior (i.e. video A), the user identifier as the user.
In another implementation, if the historical behavior of a certain user be it is multiple, electronic equipment can be from the user's
In multiple historical behaviors, the second preset number historical behavior of selected distance current time recently is preset the second of selection
The corresponding information of number historical behavior, the user identifier as the user.
Wherein, the second preset number can be rule of thumb arranged by technical staff, and the second preset number is no more than definition
The number of historical behavior used in the user identifier of sample of users, the second preset number can be 4-10, and but it is not limited to this.
The historical behavior of the available user of electronic equipment goes through the user according to the sequencing of corresponding time
History behavior is ranked up, and obtains the sequence of the historical behavior of the user.In turn, electronic equipment can be from the historical behavior of the user
Sequence in, selected distance current time nearest the second preset number historical behavior is a by the second preset number of selection
The corresponding information of historical behavior, the user identifier as the user.
Illustratively, the second preset number can be 5, and the historical behavior of a certain user can be the viewing row of the user
If the user has viewed video A in 8:00, to have viewed video S in 15:00, having viewed video D in 9:00, seen in 10:00
It has seen video F, has had viewed video G in 11:00, had viewed video H in 12:00, had viewed video J in 13:00, seen in 14:00
Video K is seen.
Electronic equipment can be after determining the historical behavior of the user, according to the sequencing of corresponding time to the use
The historical behavior at family is ranked up, and obtains the sequence of the historical behavior of the user are as follows: viewing video A, viewing video D, viewing view
Frequency F, viewing video G, viewing video H, viewing video J, viewing video K, viewing video S, then electronic equipment can be from the user
Historical behavior sequence in, selected distance current time nearest 5 historical behaviors, the corresponding information of 5 historical behaviors
Are as follows: video G, video H, video J, video K, video S, then electronic equipment can will comprising video G, video H, video J, video K,
User identifier of the sequence of video S as the user.
S102: according to the user identifier of the first user, information to be recommended and probabilistic forecasting network model trained in advance, really
Fixed first user treats the interested probability of recommendation information.
Wherein, probabilistic forecasting network model is trained to obtain based on default training set, and it includes more for presetting training set
A training sample, training sample include that the user identifier, sample information and sample of users of sample of users are interested in sample information
Probability, sample information is the corresponding information of historical behavior of sample of users, and the user identifier of sample of users is to be used according to sample
What the historical behavior at family defined.
It is understood that information to be recommended can be the information that user does not browse when to user's recommendation information, it can also
To be information that user browsed.It is emerging can to determine that user treats recommendation information sense for the method introduced according to embodiments of the present invention
The probability of interest, in turn, to user's recommendation information.
Electronic equipment can be according to preparatory trained probabilistic forecasting network model to the user identifier of the first user (i.e. the
The corresponding information of the historical behavior of one user) and information to be recommended handle, in turn, determine that the first user treats recommendation information
Interested probability.
Correspondingly, electronic equipment can also generate the training sample in default training set according to the historical behavior of sample of users
This in turn can be according to the training probabilistic forecasting network model of multiple training samples in default training set.
The method for generating the training sample in default training set, may comprise steps of:
Step 1: it is directed to each sample of users, is arranged according to historical behavior of the chronological order to the sample of users
Sequence.
For each sample of users, the historical behavior of the available sample of users of electronic equipment, according to the corresponding time
Sequencing is ranked up the historical behavior of the sample of users, in turn, carries out respective handling according to ranking results.
Step 2: according to the first preset number historical behavior of the sample of users, user's mark of the sample of users is defined
Know.
Wherein, the first preset number can be rule of thumb arranged by technical staff, for example, the first preset number can be 4-
10, but it is not limited to this.
In inventive embodiments, electronic equipment can choose the first preset number from the historical behavior of the sample of users
A historical behavior.In turn, according to the corresponding information of the first preset number historical behavior of selection, the user of sample of users is determined
Mark.
Electronic equipment can be from the historical behavior of the sample of users of this after sequence, selected distance current time closer
One preset number historical behavior, using the corresponding information of the first preset number historical behavior as the user of the sample of users
Mark.
Step 3: according to the historical behavior after the first preset number historical behavior of the sample of users, determining should
It is general to set first to the interested probability of first sample information for the sample of users for the corresponding first sample information of sample of users
Rate.
Wherein, the first probability can be arranged according to the historical behavior of first sample information is determined.
The first preset number historical behavior in the historical behavior of the sample of users is closer apart from current time, according to this
The first sample information that historical behavior in the historical behavior of sample of users after the first preset number historical behavior is determined,
It can accurately reflect the hobby of the sample of users nearest a period of time to a certain extent, it therefore, can be by the sample
This user is set as biggish numerical value to the interested probability of first sample information (i.e. the first probability), for example, the first probability can
Think 1, but it is not limited to this.
Electronic equipment can be chosen at after the first preset number historical behavior from the historical behavior of the sample of users
A historical behavior, using the corresponding information of the historical behavior as first sample information.And according to the historical behavior, by the sample
This user is set as the first probability to the interested probability of first sample information.
Step 4: according to the historical behavior before the first preset number historical behavior of the sample of users, determining should
It is general to set second to the interested probability of the second sample information for the sample of users for corresponding second sample information of sample of users
Rate.
Wherein, the second probability is lower than the first probability, and the second probability can be according to the history row for determining the second sample information
For setting.
Historical behavior in the historical behavior of the sample of users before the first preset number historical behavior apart from it is current when
It carves farther out, user may be no longer to the second sample determined according to the historical behavior before the first preset number historical behavior
Information is interested, therefore, the sample of users can be set and be lower than to the interested probability of the second sample information (i.e. the second probability)
First probability.For example, the first probability is 1, the second probability is 0, and but it is not limited to this.
Electronic equipment can be chosen at before the first preset number historical behavior from the historical behavior of the sample of users
A historical behavior, using the corresponding information of the historical behavior as the second sample information.And according to the historical behavior, by the sample
This user is set as the second probability to the interested probability of the second sample information.
Step 5: generation includes the training sample of the sample of users corresponding positive sample and negative sample.
Wherein, user identifier, first sample information and sample of users in positive sample including the sample of users are to the
One sample information interested probability includes user identifier, the second sample information and the sample of the sample of users in negative sample
This user is to the interested probability of the second sample information.
In the user identifier for determining the sample of users, the corresponding sample information of the sample of users and the sample of users
After the interested probability of the sample information determined, electronic equipment can be by the user identifier of the sample of users, the first sample
This information and the sample of users are to the interested probability of first sample information, as trained positive sample;By the sample of users
User identifier, the second sample information and the sample of users are to the interested probability of the second sample information, as trained negative sample.
In turn, the available default training set including multiple training samples.
In a kind of implementation, the first preset number can be 5, and the first probability can be 1, and the second probability can be 0.
If the historical behavior of a certain sample of users includes: 8:00 search video A, 9:00 watches video B, 10:00 viewing
Video A, 11:00 search for video C, and 12:00 searches for video D, and 13:00 watches video E, and 14:00 searches for video F, and electronic equipment can
To be ranked up according to historical behavior of the corresponding chronological order to the user, the sequence of the historical behavior of the user is obtained
Are as follows: historical behavior 1 (i.e. search video A), historical behavior 2 (i.e. viewing video B), historical behavior 3 (i.e. viewing video A), history
Behavior 4 (i.e. search video C), historical behavior 5 (i.e. search video D), historical behavior 6 (i.e. viewing video E), historical behavior 7 is (i.e.
Watch video F).
In turn, electronic equipment can choose 5 historical behaviors of historical behavior 2 to historical behavior 6, will go through comprising this 5 times
The sequence of the corresponding information of history behavior (i.e. video B, video A, video C, video D, video E), the user as the sample of users
Mark.
Then, electronic equipment can choose historical behavior 7 (i.e. viewing video F), (i.e. by the corresponding information of the historical behavior
Video F), 1 is set as to the interested probability of first sample information as first sample information, and by the sample of users.
In addition, electronic equipment can choose historical behavior 1 (i.e. search video A), (i.e. by the corresponding information of the historical behavior
Video A), 0 is set as to the interested probability of the second sample information as the second sample information, and by the sample of users.
The user identifier of the sample of users (can be included video B, video A, video C, video D, video by electronic equipment
The sequence of E), first sample information (i.e. video F) and the sample of users to the interested probability of first sample information (i.e. 1),
As trained positive sample;By the user identifier of the sample of users (include video B, video A, video C, video D, video E's
Sequence), the second sample information (i.e. video A) and the sample of users to the interested probability of the second sample information (i.e. 0), make
For trained negative sample.
In turn, electronic equipment can be trained probabilistic forecasting network model according to default training set, specifically, electronics
Equipment can mix the sample with the input parameter of the user identifier and sample information at family as probabilistic forecasting network model, mix the sample with
Family as corresponding output parameter, is trained the interested probability of sample information to probabilistic forecasting network model, until general
Rate prediction network model reaches the condition of convergence, obtains trained probabilistic forecasting network model.
Illustratively, if the positive sample of training includes: the user identifier (sequence comprising video A, video B of sample of users 1
Column), first sample information (video C) and sample of users 1 are to the interested probability of video C (that is, 1).Trained negative sample includes:
The user identifier (sequence comprising video D, video E) of sample of users 2, the second sample information (video F) and 2 pairs of sample of users views
The interested probability of frequency F (that is, 0).
Electronic equipment can mix the sample with the user identifier (that is, comprising video A, the sequence of video B) and first sample at family 1
Information (that is, video C) mixes the sample with family 1 to the interested probability of video C as the input parameter of probabilistic forecasting network model
(that is, 1) is used as corresponding output parameter;Mix the sample with the user identifier (that is, comprising video D, the sequence of video E) and at family 2
It is interested in video F to mix the sample with family 2 as the input parameter of probabilistic forecasting network model for two sample informations (that is, video F)
Probability (that is, 0) be used as corresponding output parameter, probabilistic forecasting network model is trained, until probabilistic forecasting network mould
Type reaches the condition of convergence, obtains trained probabilistic forecasting network model.
In addition, in order to improve recommendation efficiency, electronic equipment can be to preset more when the number of preset information is more
A information is screened, using the information filtered out as information to be recommended.
Preset multiple information refer to information all in information bank, and the information in information bank is technical staff according to current energy
What the Internet resources enough provided determined.
It is understood that may include the corresponding information of historical behavior of the user in information bank, can not also include
The corresponding information of the historical behavior of the user.
When preset information is multiple, determine a certain user to preset multiple information according to probabilistic forecasting network model
When interested probability, need to calculate for each information primary.For example, preset information has 1000000, then need using
Probabilistic forecasting network model calculates 1000000 times, to determine that the user is interested general to 1000000 preset information difference
Rate takes a long time.
User is calculated to the time required for preset multiple interested probability of information in order to shorten, and is improved and is recommended effect
Rate.Electronic equipment can screen preset multiple information, and third preset number information is determined from multiple information,
As information to be recommended.
In a kind of implementation, electronic equipment can (User Collaborative Filtering be used based on UserCF
Family collaborative filtering) algorithm and the first user historical behavior, determined from preset multiple information third preset number believe
Breath, as information to be recommended.
Wherein, third preset number can be rule of thumb arranged by technical staff, and third preset number can be 100, but
It is not limited to this.
Due to the structure of UserCF algorithm, the structure relative to probabilistic forecasting network model is relatively simple, according to UserCF
Algorithm calculating user is lower to the computation complexity of preset multiple interested probability of information, therefore, according to UserCF algorithm
User is calculated to the lasting duration of preset multiple interested probability of information, is less than according to probabilistic forecasting network model meter
User is calculated to the lasting duration of preset multiple interested probability of information.Then electronic equipment can be according to UserCF algorithm
Preset multiple information are screened.
Electronic equipment can be corresponding according to the corresponding information of historical behavior, the sample of users of UserCF algorithm, the first user
Information and preset multiple information, determine the first user to preset multiple interested probability of information.In turn, electronics
Equipment can be ranked up preset multiple information according to the probability being calculated, and user in preset multiple information is felt
The biggish third preset number information of the probability of interest, as information to be recommended.
Illustratively, third preset number can be 100, if the number of preset information is 500.When it needs to be determined that certain
When one user distinguishes interested probability to 500 information, electronic equipment can be according to the corresponding letter of historical behavior of the user
Breath and UserCF algorithm determine that the user distinguishes interested probability to 500 information, according to the size of obtained probability, from
In multiple information, biggish 100 information of the interested probability of the user is chosen, as information to be recommended.
In turn, electronic equipment can be according to the user identifier of the user, information to be recommended and probabilistic forecasting trained in advance
Network model determines that the user treats the interested probability of recommendation information.
Correspondingly, step S102 may comprise steps of:
Step 1: based on the first preset algorithm in probabilistic forecasting network model trained in advance, to the use of the first user
Family mark is handled, and obtains the first linear characterization vector, and treat recommendation information and handled, obtain second linearly characterize to
Amount.
In a kind of implementation, the first preset algorithm can for ALS (alternating least squares, alternately most
Small two multiply) algorithm.
It is pre- in probability when it needs to be determined that a certain user treats recommendation information interested probability in inventive embodiments
The linear transformation part of network model is surveyed, the can be obtained according to ALS algorithm, to handling for the user identifier of the user
One linear characterization vector Xua;It treats recommendation information to be handled, obtains the second linear characterization vector Xia。
Step 2: based on to the first linear characterization vector carry out the processing result of process of convolution and second linearly characterize to
Amount obtains the linear character vector of the common trait of the corresponding information of historical behavior for indicating the first user and information to be recommended.
Obtaining the first linear characterization vector XuaWith the second linear characterization vector XiaLater, probabilistic forecasting net can be passed through
The convolutional layer of network model, to the first linear characterization vector XuaProcess of convolution is carried out, obtains indicating that the historical behavior of the user is corresponding
Information common trait linear character vector Cua.In turn, by the point lamination of probabilistic forecasting network model, to linear character
Vector CuaWith the second linear characterization vector XiaCarry out dot product operation, obtain indicating the corresponding information of the historical behavior of the user with
The linear character vector C of the common trait of information to be recommendedlinear。
Step 3: based on the second preset algorithm in probabilistic forecasting network model, the user identifier of the first user is carried out
Processing, obtains the first non-linear characterization vector, and treat recommendation information and handled, obtains the second non-linear characterization vector.
In a kind of implementation, the second preset algorithm can be Fasttext (Fast Text Classification device) algorithm.
In the nonlinear transformation part of probabilistic forecasting network model, by Fasttext, to the user identifier of the user into
Row processing, obtains the first non-linear characterization vector Xuf;It treats recommendation information to be handled, obtains the second non-linear characterization vector
Xif。
Step 4: based on the knot for carrying out Fusion Features to the first non-linear non-linear characterization vector of characterization vector sum second
Fruit, obtain indicate the first user the corresponding information of historical behavior and information to be recommended common trait nonlinear characteristic to
Amount.
Obtaining the first non-linear characterization vector XufWith the second non-linear characterization vector XifIt later, can be pre- by probability
The first preset formula in the full articulamentum of the nonlinear transformation part of network model is surveyed, to the first non-linear characterization vector XufWith
Second non-linear characterization vector XifFusion Features are carried out, obtain indicating the corresponding information of the historical behavior of the user and letter to be recommended
The non-linear characterization vector C of breathconcat.Wherein, the first preset formula can be with are as follows:
Cconcat=concat (Xuf, Xif)
Wherein, concat () indicates to carry out splicing to the array in bracket.
In turn, according to the second preset formula, to non-linear characterization vector CconcatFeature extraction is carried out, obtains indicating the use
The nonlinear characteristic vector C of the common trait of the corresponding information of the historical behavior at family and information to be recommendednonlinear.Wherein, second
Preset formula can be with are as follows:
Cnonlinear=ReLU (W × Cconcat+B)
Wherein, ReLU (Rectified Linear Unit, linear amending unit) is commonly used in a kind of artificial neural network
Activation primitive, W indicate probabilistic forecasting network model transformation parameter, B indicate probabilistic forecasting network model offset parameter.
Step 5: Fusion Features are carried out to linear character vector sum nonlinear characteristic vector, obtain target feature vector.
In the linear character of the common trait for the corresponding information of historical behavior and information to be recommended for obtaining indicating the user
Vector ClinearWith nonlinear characteristic vector CnonlinearIt later, can be pre- by the third in the hidden layer of probabilistic forecasting network model
If formula, to linear character vector ClinearWith nonlinear characteristic vector CnonlinearFusion Features are carried out, obtain indicating the user
The corresponding information of historical behavior and information to be recommended common trait target feature vector Chidden.Wherein, the default public affairs of third
Formula are as follows:
Chidden=concat (Clinear, Cnonlinear)
Wherein, concat () indicates to carry out splicing to the array in bracket.
Step 6: carrying out recurrence processing to target feature vector, and it is interested general to determine that the first user treats recommendation information
Rate.
By the 4th preset formula in the output layer of probabilistic forecasting network model, to the historical behavior pair for indicating the user
The target feature vector C of the common trait of the information and information to be recommended answeredhiddenRecurrence processing is carried out, the user is obtained and treats
The interested probability of recommendation information.Wherein, the 4th preset formula can be with are as follows:
P=soffmax (W × Chidden+B)
Wherein, P indicates that the user treats the interested probability of recommendation information, and softmax () is indicated to the numerical value in bracket
Recurrence processing is carried out, W indicates the transformation parameter of probabilistic forecasting network model, and B indicates the offset parameter of probabilistic forecasting network model.
Illustratively, when it needs to be determined that a certain user treats recommendation information interested probability, if information to be recommended
It include: video C, video D, video E, the user identifier of the user determined are as follows: the sequence comprising video A, video B, then electronics
Equipment can be using the user identifier (that is, sequence comprising video A, video B) and video C of the user as probabilistic forecasting network mould
The input parameter of type.Then the output parameter of corresponding probabilistic forecasting network model is the user to the interested probability of video C
Pc。
Similarly, the available user is to the interested probability P of video DdAnd the user is to the interested probability of video E
Pe。
In a kind of implementation, in order to ensure to treat recommendation information interested by the user of probabilistic forecasting network model prediction
The accuracy of probability, probabilistic forecasting network model can be NCF network model.
Correspondingly, electronic equipment can believe sample according to the user identifier, sample information and sample of users of sample of users
Interested probability is ceased, NCF network model is trained.In turn, when needing to predict that it is emerging that a certain user treats recommendation information sense
Interest probability when, electronic equipment can according to the user identifier of the user, information to be recommended and trained NCF network model,
Determine that the user treats the interested probability of recommendation information.
In addition, in order to improve the accuracy that the user of NCF network model prediction treats the interested probability of recommendation information, it can
Periodically to update NCF network model, the update cycle of NCF network model can be rule of thumb configured by technical staff.
In a kind of implementation, the update cycle can be 1 day, if electronic equipment is according to user in first 60 days of acquisition
The corresponding information of historical behavior, training NCF network model.At the 61st day, reach the NCF network model update cycle, electronics is set
It is standby NCF network model to be trained according to the corresponding information of historical behavior of user in the 2nd day to the 61st day;At the 62nd day,
Reach the NCF network model update cycle, electronic equipment can be according to the corresponding letter of historical behavior of user in the 3rd day to the 62nd day
Breath, training NCF network model.
S103: according to obtained probability and information to be recommended, to first user's recommendation information.
In inventive embodiments, when needing to a certain user's recommendation information, electronic equipment can be treated according to the user
The interested probability of recommendation information, to user's recommendation information.
When information to be recommended is one, electronic equipment may determine that the user to the interested probability of information to be recommended
Whether preset threshold is reached, if the user reaches preset threshold to the interested probability of information to be recommended, to the user
Recommend the information to be recommended;If the user is not up to preset threshold to the interested probability of information to be recommended, not to this
User's recommendation information.Wherein, preset threshold can be rule of thumb arranged by technical staff.
Illustratively, preset threshold can be 0.5, and information to be recommended can be video A, if NCF network model determines
The user to the interested probability of video A be P1, when P1 be greater than or equal to 0.5 when, electronic equipment can be recommended to the user
Video A;When P1 is less than 0.5, then electronic equipment can not be to user's recommendation information.
When information to be recommended is multiple, electronic equipment can treat the interested probability of recommendation information according to the user
Size, to user's recommendation information.
In a kind of implementation, the sequence that electronic equipment can be descending according to probability is treated recommendation information and is arranged
Sequence obtains information sequence to be recommended, in turn, can recommend to the user more forward wait push away in the information sequence to be recommended
Recommend information.
Illustratively, if information to be recommended are as follows: video A, video B, video C and video D, it is true according to NCF network model
Fixed a certain user is respectively as follows: 0.5,0.3,0.7,0.9 to video interested probability to be recommended.Electronic equipment is according to probability
Descending sequence treats the information sequence to be recommended after recommendation information is ranked up, obtained are as follows: video D, video C, view
Frequency A, video B, then video D can be recommended the user by electronic equipment, alternatively, electronic equipment can push away video D and video C
It recommends and gives the user.
Based on above-mentioned processing, even if a certain user is not sample of users, due to the use according to the user identifier of the user
The corresponding information of the history at family determines, it is known that, the user identifier (i.e. the corresponding information of the historical behavior of the user) of the user,
Correlation between the user identifier (i.e. the corresponding information of the historical behavior of sample of users) of information and sample of users to be recommended compared with
By force, in turn, probabilistic forecasting network model is according to the user identifier of the user, the user identifier of information to be recommended and sample of users,
The accuracy that the user determined treats the interested probability of recommendation information is higher, according to obtained probability and letter to be recommended
Breath, to user's recommendation information, can be improved the validity of information recommendation.
In addition, define user identifier according to the corresponding information of the historical behavior of user, by user identifier and user
The corresponding information of historical behavior stores when local, can only construct depositing for the corresponding information of historical behavior for storing user
It stores up space (being properly termed as information storage space), according to the available user identifier of data in information storage space, does not have to again
The memory space (being properly termed as user identifier memory space) for storing user identifier is individually constructed, compared with the existing technology
In, it needs to construct two memory spaces (that is, information storage space and user identifier memory space), reduces memory use.
Optionally, information to be recommended can be video to be recommended, and referring to fig. 2, Fig. 2 is one kind provided in an embodiment of the present invention
The exemplary flow chart of information recommendation method, may comprise steps of:
S201: according to the historical behavior of sample of users, the default training set comprising multiple training samples is generated.
Wherein, presetting includes multiple training samples in training set, includes the user identifier of sample of users, sample in training sample
This information, for sample of users to the interested probability of sample information, sample information is the corresponding information of historical behavior of sample of users,
The user identifier of sample of users is to be defined according to the historical behavior of sample of users.
S202: according to default training set, NCF network model is trained.
S203: by the second preset number historical behavior pair in the historical behavior of the first user apart from current time recently
The information answered, the user identifier as the first user.
Wherein, the second preset number can be rule of thumb arranged by technical staff, and the second preset number can be 4-10, but
It is not limited to this.
S204: according to the historical behavior of the first user and UserCF algorithm, third is determined from preset multiple videos
Preset number video, as video to be recommended.
Wherein, third preset number can be rule of thumb arranged by technical staff, and third preset number can be 100, but
It is not limited to this.
S205: it according to the user identifier of the first user, video to be recommended and preparatory trained NCF network model, determines
First user is to video interested probability to be recommended.
S206: recommend the video of the interested maximum probability of the first user in video to be recommended to the first user.
Referring to Fig. 3, Fig. 3 is a kind of structure chart of probabilistic forecasting network model provided in an embodiment of the present invention, and the probability is pre-
Surveying network model may include: linear segment, non-linear partial, hidden layer and output layer.
Linear transformation part may include: input layer, convolutional layer and point lamination.
Input layer can be by the first preset algorithm, and user identifier and information to be recommended to user are respectively processed,
Obtain respective linear characterization vector.
Convolutional layer can linear characterization vector corresponding to user identifier and information to be recommended carry out feature extraction, obtain energy
Enough indicate the feature vector of the common trait of the corresponding information of historical behavior of user.
Point lamination can to indicate user the corresponding information of historical behavior common trait feature vector, and indicate to
The linear characterization vector of recommendation information carries out dot product operation, obtains user identifier (the i.e. history row of user that can indicate user
For corresponding information) linear character vector with the common trait of information to be recommended.
Nonlinear transformation part includes: input layer and full articulamentum.
Input layer can be by the second preset algorithm, and user identifier and information to be recommended to user are respectively processed,
Obtain respective non-linear characterization vector.
The articulamentum that articulamentum can be identical by multiple structures and different parameter entirely is constituted.
Full articulamentum can non-linear characterization vector progress feature extraction corresponding to user identifier and information to be recommended, obtain
To the common trait of the user identifier (i.e. the corresponding information of the historical behavior of user) and information to be recommended that can indicate user
Nonlinear characteristic vector.
Linear character vector sum nonlinear characteristic vector can be carried out Fusion Features by hidden layer, obtain to indicate user's
The target feature vector of the common trait of the corresponding information of historical behavior and information to be recommended.
Output layer can carry out recurrence processing to target feature vector, and obtaining user, to treat recommendation information interested general
Rate, in turn, probabilistic forecasting network model can export user and treat the interested probability of recommendation information.
Referring to fig. 4, based on the same inventive concept, corresponding with information recommendation method provided by the above embodiment, the present invention
Embodiment additionally provides a kind of information recommending apparatus, and described device includes:
First determining module 401 defines user's mark of first user for the historical behavior according to the first user
Know;
Second determining module 402, for according to the user identifier of first user, information to be recommended and training in advance
Probabilistic forecasting network model determines first user to the interested probability of information to be recommended;
Recommending module 403, for according to obtained probability and the information to be recommended, Xiang Suoshu the first user recommendation
Breath;
Wherein, the probabilistic forecasting network model is trained to obtain based on default training set, the default training
Collection includes multiple training samples, and the training sample includes the user identifier, sample information and the sample of users of sample of users
To the interested probability of the sample information, the sample information is the corresponding information of historical behavior of the sample of users, institute
The user identifier for stating sample of users is to be defined according to the historical behavior of the sample of users.
Optionally, described device further include:
Generation module, for being directed to each sample of users, according to chronological order to the historical behavior of the sample of users
It is ranked up;
According to the first preset number historical behavior of the sample of users, the user identifier of the sample of users is defined;
According to the historical behavior after the first preset number historical behavior of the sample of users, the sample is determined
It is general to set first to the interested probability of first sample information for the sample of users for the corresponding first sample information of user
Rate;
According to the historical behavior before the first preset number historical behavior of the sample of users, the sample is determined
It is general to set second to the interested probability of the second sample information for the sample of users for corresponding second sample information of user
Rate, wherein second probability is lower than first probability;
Generation includes the training sample of the corresponding positive sample of the sample of users and negative sample, wherein the positive sample
User identifier, the first sample information and the sample of users in this including the sample of users believe the first sample
Interested probability is ceased, includes user identifier, second sample information and the sample of the sample of users in the negative sample
This user is to the interested probability of the second sample information.
Optionally, first determining module 401, specifically for being gone through according to chronological order to first user
History behavior is ranked up;
According in the historical behavior of first user apart from the second preset number historical behavior that current time is nearest,
Define the user identifier of first user, wherein second preset number is no more than the user for defining the sample of users
The number of historical behavior used in identifying.
Optionally, the user identifier of first user is fixed according to the corresponding information of historical behavior of first user
Justice;
Second determining module 402, specifically for pre- based on first in probabilistic forecasting network model trained in advance
Imputation method handles the user identifier of first user, obtains the first linear characterization vector, and to the letter to be recommended
Breath is handled, and the second linear characterization vector is obtained;
Based on to the described first linear characterization vector carry out the processing result of process of convolution and described second linearly characterize to
Amount obtains the linear spy of the common trait of the corresponding information of historical behavior for indicating first user and the information to be recommended
Levy vector;
Based on the second preset algorithm in the probabilistic forecasting network model, the user identifier of first user is carried out
Processing, obtain the first non-linear characterization vector, and handle the information to be recommended, obtain second it is non-linear characterize to
Amount;
Based on the knot for carrying out Fusion Features to the second non-linear characterization vector described in the described first non-linear characterization vector sum
Fruit obtains indicating the non-linear of the corresponding information of historical behavior of first user and the common trait of the information to be recommended
Feature vector;
Fusion Features are carried out to nonlinear characteristic vector described in the linear character vector sum, obtain target feature vector;
Recurrence processing is carried out to the target feature vector, determines that first user is interested in the information to be recommended
Probability.
Optionally, described device further include:
Screening module, for the historical behavior based on user collaborative filtering UserCF algorithm and first user, from pre-
If multiple information in determine third preset number information, as the information to be recommended.
Based on above-mentioned processing, even if a certain user is not sample of users, due to the use according to the user identifier of the user
The corresponding information of the historical behavior at family determines, it is known that, user identifier (the i.e. corresponding letter of the historical behavior of the user of the user
Breath), the correlation between information to be recommended and the user identifier (i.e. the corresponding information of the historical behavior of sample of users) of sample of users
Property it is stronger, in turn, probabilistic forecasting network model according to the user of the user identifier of the user, information to be recommended and sample of users mark
Know, the accuracy that the user determined treats the interested probability of recommendation information is higher, according to obtained probability and to be recommended
Information can be improved the validity of information recommendation to user's recommendation information.
The embodiment of the invention also provides a kind of electronic equipment, as shown in figure 5, include processor 501, communication interface 502,
Memory 503 and communication bus 504, wherein processor 501, communication interface 502, memory 503 are complete by communication bus 504
At mutual communication,
Memory 503, for storing computer program;
Processor 501 when for executing the program stored on memory 503, realizes following steps:
According to the historical behavior of the first user, the user identifier of first user is defined;
According to the user identifier of first user, information to be recommended and probabilistic forecasting network model trained in advance, really
Fixed first user is to the interested probability of information to be recommended;
According to obtained probability and the information to be recommended, Xiang Suoshu the first user recommendation information;
Wherein, the probabilistic forecasting network model is trained to obtain based on default training set, the default training
Collection includes multiple training samples, and the training sample includes the user identifier, sample information and the sample of users of sample of users
To the interested probability of the sample information, the sample information is the corresponding information of historical behavior of the sample of users, institute
The user identifier for stating sample of users is to be defined according to the historical behavior of the sample of users.
The communication bus that above-mentioned electronic equipment is mentioned can be Peripheral Component Interconnect standard (Peripheral Component
Interconnect, PCI) bus or expanding the industrial standard structure (Extended Industry Standard
Architecture, EISA) bus etc..The communication bus can be divided into address bus, data/address bus, control bus etc..For just
It is only indicated with a thick line in expression, figure, it is not intended that an only bus or a type of bus.
Communication interface is for the communication between above-mentioned electronic equipment and other equipment.
Memory may include random access memory (Random Access Memory, RAM), also may include non-easy
The property lost memory (Non-Volatile Memory, NVM), for example, at least a magnetic disk storage.Optionally, memory may be used also
To be storage device that at least one is located remotely from aforementioned processor.
Above-mentioned processor can be general processor, including central processing unit (Central Processing Unit,
CPU), network processing unit (Network Processor, NP) etc.;It can also be digital signal processor (Digital Signal
Processing, DSP), it is specific integrated circuit (Application Specific Integrated Circuit, ASIC), existing
It is field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device, discrete
Door or transistor logic, discrete hardware components.
Electronic equipment provided in an embodiment of the present invention, when carrying out information recommendation, even if a certain user is not sample of users,
It is determined due to the corresponding information of the historical behavior of the user according to the user identifier of the user, it is known that, the user of the user
Identify (i.e. the corresponding information of the historical behavior of the user), user identifier (the i.e. sample of users of information to be recommended and sample of users
The corresponding information of historical behavior) between correlation it is stronger, in turn, probabilistic forecasting network model according to the user of the user mark
Know, the user identifier of information to be recommended and sample of users, the user determined treats the standard of the interested probability of recommendation information
Exactness is higher, can be improved the effective of information recommendation to user's recommendation information according to obtained probability and information to be recommended
Property.
In another embodiment provided by the invention, a kind of computer readable storage medium is additionally provided, which can
It reads to be stored with computer program in storage medium, any information in embodiment is realized when the computer program is executed by processor
The step of recommended method.
Specifically, the above method includes:
According to the historical behavior of the first user, the user identifier of first user is defined;
According to the user identifier of first user, information to be recommended and probabilistic forecasting network model trained in advance, really
Fixed first user is to the interested probability of information to be recommended;
According to obtained probability and the information to be recommended, Xiang Suoshu the first user recommendation information;
Wherein, the probabilistic forecasting network model is trained to obtain based on default training set, the default training
Collection includes multiple training samples, and the training sample includes the user identifier, sample information and the sample of users of sample of users
To the interested probability of the sample information, the sample information is the corresponding information of historical behavior of the sample of users, institute
The user identifier for stating sample of users is to be defined according to the historical behavior of the sample of users.
It should be noted that other implementations of above- mentioned information recommended method are identical as preceding method embodiment part,
Which is not described herein again.
By running the instruction stored in computer readable storage medium provided in an embodiment of the present invention, pushed away carrying out information
When recommending, even if a certain user is not sample of users, since the historical behavior of the user according to the user identifier of the user is corresponding
Information determine, it is known that, the user identifier (i.e. the corresponding information of the historical behavior of the user) of the user, information to be recommended and
Correlation between the user identifier (i.e. the corresponding information of the historical behavior of sample of users) of sample of users is stronger, in turn, probability
Prediction network model is according to the user identifier of the user, the user identifier of information to be recommended and sample of users, the use determined
The accuracy that the interested probability of recommendation information is treated at family is higher, according to obtained probability and information to be recommended, pushes away to the user
Information is recommended, can be improved the validity of information recommendation.
In another embodiment provided by the invention, a kind of computer program product comprising instruction is additionally provided, when it
When running on computers, so that the step of computer executes any information recommended method in above-described embodiment.
Specifically, the above method includes:
According to the historical behavior of the first user, the user identifier of first user is defined;
According to the user identifier of first user, information to be recommended and probabilistic forecasting network model trained in advance, really
Fixed first user is to the interested probability of information to be recommended;
According to obtained probability and the information to be recommended, Xiang Suoshu the first user recommendation information;
Wherein, the probabilistic forecasting network model is trained to obtain based on default training set, the default training
Collection includes multiple training samples, and the training sample includes the user identifier, sample information and the sample of users of sample of users
To the interested probability of the sample information, the sample information is the corresponding information of historical behavior of the sample of users, institute
The user identifier for stating sample of users is to be defined according to the historical behavior of the sample of users.
It should be noted that other implementations of above- mentioned information recommended method are identical as preceding method embodiment part,
Which is not described herein again.
By running computer program product provided in an embodiment of the present invention, when carrying out information recommendation, even if a certain use
Family is not sample of users, is determined due to the corresponding information of the historical behavior of the user according to the user identifier of the user, can
Know, the user of the user identifier (i.e. the corresponding information of the historical behavior of the user) of the user, information to be recommended and sample of users
The correlation identified between (i.e. the corresponding information of the historical behavior of sample of users) is stronger, in turn, probabilistic forecasting network model root
According to the user identifier of the user, the user identifier of information to be recommended and sample of users, the user determined treats recommendation information
The accuracy of interested probability is higher, can be improved according to obtained probability and information to be recommended to user's recommendation information
The validity of information recommendation.
In the above-described embodiments, can come wholly or partly by software, hardware, firmware or any combination thereof real
It is existing.When implemented in software, it can entirely or partly realize in the form of a computer program product.The computer program
Product includes one or more computer instructions.When loading on computers and executing the computer program instructions, all or
It partly generates according to process or function described in the embodiment of the present invention.The computer can be general purpose computer, dedicated meter
Calculation machine, computer network or other programmable devices.The computer instruction can store in computer readable storage medium
In, or from a computer readable storage medium to the transmission of another computer readable storage medium, for example, the computer
Instruction can pass through wired (such as coaxial cable, optical fiber, number from a web-site, computer, server or data center
User's line (DSL)) or wireless (such as infrared, wireless, microwave etc.) mode to another web-site, computer, server or
Data center is transmitted.The computer readable storage medium can be any usable medium that computer can access or
It is comprising data storage devices such as one or more usable mediums integrated server, data centers.The usable medium can be with
It is magnetic medium, (for example, floppy disk, hard disk, tape), optical medium (for example, DVD) or semiconductor medium (such as solid state hard disk
SolidState Disk (SSD)) etc..
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in process, method, article or equipment including the element.
Each embodiment in this specification is all made of relevant mode and describes, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for device,
For electronic equipment, computer readable storage medium and computer program product embodiments, since it is substantially similar to method reality
Example is applied, so being described relatively simple, the relevent part can refer to the partial explaination of embodiments of method.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the scope of the present invention.It is all
Any modification, equivalent replacement, improvement and so within the spirit and principles in the present invention, are all contained in protection scope of the present invention
It is interior.
Claims (12)
1. a kind of information recommendation method, which is characterized in that the described method includes:
According to the historical behavior of the first user, the user identifier of first user is defined;
According to the user identifier of first user, information to be recommended and probabilistic forecasting network model trained in advance, institute is determined
The first user is stated to the interested probability of information to be recommended;
According to obtained probability and the information to be recommended, Xiang Suoshu the first user recommendation information;
Wherein, the probabilistic forecasting network model is trained to obtain based on default training set, the default training set packet
Multiple training samples are included, the training sample includes the user identifier, sample information and the sample of users of sample of users to institute
The interested probability of sample information is stated, the sample information is the corresponding information of historical behavior of the sample of users, the sample
The user identifier of this user is to be defined according to the historical behavior of the sample of users.
2. the method according to claim 1, wherein generating the training sample, comprising:
For each sample of users, it is ranked up according to historical behavior of the chronological order to the sample of users;
According to the first preset number historical behavior of the sample of users, the user identifier of the sample of users is defined;
According to the historical behavior after the first preset number historical behavior of the sample of users, the sample of users is determined
Corresponding first sample information sets the first probability to the interested probability of first sample information for the sample of users;
According to the historical behavior before the first preset number historical behavior of the sample of users, the sample of users is determined
Corresponding second sample information sets the second probability to the interested probability of the second sample information for the sample of users,
Wherein, second probability is lower than first probability;
Generation includes the training sample of the corresponding positive sample of the sample of users and negative sample, wherein in the positive sample
User identifier, the first sample information and the sample of users including the sample of users are to the first sample information sense
The probability of interest includes that the user identifier, second sample information and the sample of the sample of users is used in the negative sample
Family is to the interested probability of the second sample information.
3. the method according to claim 1, wherein the historical behavior according to the first user, described in definition
The user identifier of first user, comprising:
It is ranked up according to historical behavior of the chronological order to first user;
According to the second preset number historical behavior in the historical behavior of first user apart from current time recently, definition
The user identifier of first user, wherein second preset number is no more than the user identifier for defining the sample of users
The number of used historical behavior.
4. the method according to claim 1, wherein the user identifier of first user is according to described first
The corresponding information of the historical behavior of user defines;
It is described according to the user identifier of first user, information to be recommended and probabilistic forecasting network model trained in advance, really
Fixed first user is to the interested probability of information to be recommended, comprising:
Based on the first preset algorithm in probabilistic forecasting network model trained in advance, to the user identifier of first user into
Row processing obtains the first linear characterization vector, and handles the information to be recommended, obtains the second linear characterization vector;
Based on the processing result and the second linear characterization vector for carrying out process of convolution to the described first linear characterization vector, obtain
To the corresponding information of historical behavior and the information to be recommended for indicating first user common trait linear character to
Amount;
Based on the second preset algorithm in the probabilistic forecasting network model, at the user identifier of first user
Reason, obtains the first non-linear characterization vector, and handle the information to be recommended, obtains the second non-linear characterization vector;
Based on to described first it is non-linear characterization vector sum described in second it is non-linear characterization vector carry out Fusion Features as a result,
To the nonlinear characteristic of the common trait for the corresponding information of historical behavior and the information to be recommended for indicating first user
Vector;
Fusion Features are carried out to nonlinear characteristic vector described in the linear character vector sum, obtain target feature vector;
Recurrence processing is carried out to the target feature vector, determines that first user is interested to the information to be recommended general
Rate.
5. the method according to claim 1, wherein the user identifier according to first user, to
Recommendation information and probabilistic forecasting network model trained in advance, determine that first user is interested to the information to be recommended
Before probability, the method also includes:
Based on the historical behavior of user collaborative filtering UserCF algorithm and first user, determined from preset multiple information
Third preset number information out, as the information to be recommended.
6. a kind of information recommending apparatus, which is characterized in that described device includes:
First determining module defines the user identifier of first user for the historical behavior according to the first user;
Second determining module, for pre- according to the user identifier of first user, information to be recommended and probability trained in advance
Network model is surveyed, determines first user to the interested probability of information to be recommended;
Recommending module, for according to obtained probability and the information to be recommended, Xiang Suoshu the first user recommendation information;
Wherein, the probabilistic forecasting network model is trained to obtain based on default training set, the default training set packet
Multiple training samples are included, the training sample includes the user identifier, sample information and the sample of users of sample of users to institute
The interested probability of sample information is stated, the sample information is the corresponding information of historical behavior of the sample of users, the sample
The user identifier of this user is to be defined according to the historical behavior of the sample of users.
7. device according to claim 6, which is characterized in that described device further include:
Generation module is carried out for being directed to each sample of users according to historical behavior of the chronological order to the sample of users
Sequence;
According to the first preset number historical behavior of the sample of users, the user identifier of the sample of users is defined;
According to the historical behavior after the first preset number historical behavior of the sample of users, the sample of users is determined
Corresponding first sample information sets the first probability to the interested probability of first sample information for the sample of users;
According to the historical behavior before the first preset number historical behavior of the sample of users, the sample of users is determined
Corresponding second sample information sets the second probability to the interested probability of the second sample information for the sample of users,
Wherein, second probability is lower than first probability;
Generation includes the training sample of the corresponding positive sample of the sample of users and negative sample, wherein in the positive sample
User identifier, the first sample information and the sample of users including the sample of users are to the first sample information sense
The probability of interest includes that the user identifier, second sample information and the sample of the sample of users is used in the negative sample
Family is to the interested probability of the second sample information.
8. device according to claim 6, which is characterized in that first determining module is specifically used for according to time elder generation
Sequence is ranked up the historical behavior of first user afterwards;
According to the second preset number historical behavior in the historical behavior of first user apart from current time recently, definition
The user identifier of first user, wherein second preset number is no more than the user identifier for defining the sample of users
The number of used historical behavior.
9. device according to claim 6, which is characterized in that the user identifier of first user is according to described first
The corresponding information of the historical behavior of user defines;
Second determining module, specifically for based on the first preset algorithm in probabilistic forecasting network model trained in advance,
The user identifier of first user is handled, obtains the first linear characterization vector, and carry out to the information to be recommended
Processing obtains the second linear characterization vector;
Based on the processing result and the second linear characterization vector for carrying out process of convolution to the described first linear characterization vector, obtain
To the corresponding information of historical behavior and the information to be recommended for indicating first user common trait linear character to
Amount;
Based on the second preset algorithm in the probabilistic forecasting network model, at the user identifier of first user
Reason, obtains the first non-linear characterization vector, and handle the information to be recommended, obtains the second non-linear characterization vector;
Based on to described first it is non-linear characterization vector sum described in second it is non-linear characterization vector carry out Fusion Features as a result,
To the nonlinear characteristic of the common trait for the corresponding information of historical behavior and the information to be recommended for indicating first user
Vector;
Fusion Features are carried out to nonlinear characteristic vector described in the linear character vector sum, obtain target feature vector;
Recurrence processing is carried out to the target feature vector, determines that first user is interested to the information to be recommended general
Rate.
10. device according to claim 6, which is characterized in that described device further include:
Screening module, for the historical behavior based on user collaborative filtering UserCF algorithm and first user, from preset
Third preset number information is determined in multiple information, as the information to be recommended.
11. a kind of electronic equipment, which is characterized in that including processor, communication interface, memory and communication bus, wherein processing
Device, communication interface, memory complete mutual communication by communication bus;
Memory, for storing computer program;
Processor when for executing the program stored on memory, realizes any method and step of claim 1-5.
12. a kind of computer readable storage medium, which is characterized in that be stored with computer in the computer readable storage medium
Program, the computer program realize method and step as claimed in any one of claims 1 to 5 when being executed by processor.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910665176.1A CN110442788A (en) | 2019-07-23 | 2019-07-23 | A kind of information recommendation method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910665176.1A CN110442788A (en) | 2019-07-23 | 2019-07-23 | A kind of information recommendation method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110442788A true CN110442788A (en) | 2019-11-12 |
Family
ID=68431229
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910665176.1A Pending CN110442788A (en) | 2019-07-23 | 2019-07-23 | A kind of information recommendation method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110442788A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111008667A (en) * | 2019-12-06 | 2020-04-14 | 北京爱奇艺科技有限公司 | Feature extraction method and device and electronic equipment |
CN111191059A (en) * | 2019-12-31 | 2020-05-22 | 腾讯科技(深圳)有限公司 | Image processing method, image processing device, computer storage medium and electronic equipment |
CN111797318A (en) * | 2020-07-01 | 2020-10-20 | 喜大(上海)网络科技有限公司 | Information recommendation method, device, equipment and storage medium |
CN111814048A (en) * | 2020-07-03 | 2020-10-23 | 北京邮电大学 | Information recommendation method and device |
CN111814044A (en) * | 2020-06-30 | 2020-10-23 | 广州视源电子科技股份有限公司 | Recommendation method and device, terminal equipment and storage medium |
CN112905839A (en) * | 2021-02-10 | 2021-06-04 | 北京有竹居网络技术有限公司 | Model training method, model using device, storage medium and equipment |
CN113032589A (en) * | 2021-03-29 | 2021-06-25 | 北京奇艺世纪科技有限公司 | Multimedia file recommendation method and device, electronic equipment and readable storage medium |
CN113672801A (en) * | 2021-07-30 | 2021-11-19 | 北京三快在线科技有限公司 | Information processing method and device, storage medium and electronic equipment |
CN115225702A (en) * | 2021-04-19 | 2022-10-21 | 北京达佳互联信息技术有限公司 | Information pushing method and device, electronic equipment and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106485562A (en) * | 2015-09-01 | 2017-03-08 | 苏宁云商集团股份有限公司 | A kind of commodity information recommendation method based on user's history behavior and system |
US20180240158A1 (en) * | 2017-02-17 | 2018-08-23 | Kasatria Analytics Sdn Bhd | Computer implemented system and method for customer profiling using micro-conversions via machine learning |
CN109087178A (en) * | 2018-08-28 | 2018-12-25 | 清华大学 | Method of Commodity Recommendation and device |
CN109685631A (en) * | 2019-01-10 | 2019-04-26 | 博拉网络股份有限公司 | A kind of personalized recommendation method based on big data user behavior analysis |
-
2019
- 2019-07-23 CN CN201910665176.1A patent/CN110442788A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106485562A (en) * | 2015-09-01 | 2017-03-08 | 苏宁云商集团股份有限公司 | A kind of commodity information recommendation method based on user's history behavior and system |
US20180240158A1 (en) * | 2017-02-17 | 2018-08-23 | Kasatria Analytics Sdn Bhd | Computer implemented system and method for customer profiling using micro-conversions via machine learning |
CN109087178A (en) * | 2018-08-28 | 2018-12-25 | 清华大学 | Method of Commodity Recommendation and device |
CN109685631A (en) * | 2019-01-10 | 2019-04-26 | 博拉网络股份有限公司 | A kind of personalized recommendation method based on big data user behavior analysis |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111008667A (en) * | 2019-12-06 | 2020-04-14 | 北京爱奇艺科技有限公司 | Feature extraction method and device and electronic equipment |
CN111008667B (en) * | 2019-12-06 | 2023-06-02 | 北京爱奇艺科技有限公司 | Feature extraction method and device and electronic equipment |
CN111191059A (en) * | 2019-12-31 | 2020-05-22 | 腾讯科技(深圳)有限公司 | Image processing method, image processing device, computer storage medium and electronic equipment |
CN111191059B (en) * | 2019-12-31 | 2023-05-05 | 腾讯科技(深圳)有限公司 | Image processing method, device, computer storage medium and electronic equipment |
CN111814044A (en) * | 2020-06-30 | 2020-10-23 | 广州视源电子科技股份有限公司 | Recommendation method and device, terminal equipment and storage medium |
CN111797318A (en) * | 2020-07-01 | 2020-10-20 | 喜大(上海)网络科技有限公司 | Information recommendation method, device, equipment and storage medium |
CN111797318B (en) * | 2020-07-01 | 2024-02-23 | 喜大(上海)网络科技有限公司 | Information recommendation method, device, equipment and storage medium |
CN111814048B (en) * | 2020-07-03 | 2023-01-17 | 北京邮电大学 | Information recommendation method and device |
CN111814048A (en) * | 2020-07-03 | 2020-10-23 | 北京邮电大学 | Information recommendation method and device |
CN112905839A (en) * | 2021-02-10 | 2021-06-04 | 北京有竹居网络技术有限公司 | Model training method, model using device, storage medium and equipment |
CN113032589A (en) * | 2021-03-29 | 2021-06-25 | 北京奇艺世纪科技有限公司 | Multimedia file recommendation method and device, electronic equipment and readable storage medium |
CN113032589B (en) * | 2021-03-29 | 2024-01-23 | 北京奇艺世纪科技有限公司 | Multimedia file recommendation method and device, electronic equipment and readable storage medium |
CN115225702A (en) * | 2021-04-19 | 2022-10-21 | 北京达佳互联信息技术有限公司 | Information pushing method and device, electronic equipment and storage medium |
CN115225702B (en) * | 2021-04-19 | 2023-08-08 | 北京达佳互联信息技术有限公司 | Information pushing method and device, electronic equipment and storage medium |
CN113672801A (en) * | 2021-07-30 | 2021-11-19 | 北京三快在线科技有限公司 | Information processing method and device, storage medium and electronic equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110442788A (en) | A kind of information recommendation method and device | |
US20180158097A1 (en) | Network-aware Product Rollout in Online Social Networks | |
US20180068371A1 (en) | Learning Vector-Space Representations of Items for Recommendations using Word Embedding Models | |
US10445699B2 (en) | Social electronic design automation | |
CN105975641A (en) | Video recommendation method ad device | |
CN106294830A (en) | The recommendation method and device of multimedia resource | |
CN104243590B (en) | Resource object recommends method and apparatus | |
CN111506820B (en) | Recommendation model, recommendation method, recommendation device, recommendation equipment and recommendation storage medium | |
CN110413867B (en) | Method and system for content recommendation | |
CN103744916B (en) | A kind of method and apparatus for sharing temperature information for being used to determine target video | |
CN110971659A (en) | Recommendation message pushing method and device and storage medium | |
CN112100221B (en) | Information recommendation method and device, recommendation server and storage medium | |
CN103455522A (en) | Recommendation method and system of application extension tools | |
JP2010204966A (en) | Sampling device, sampling method, sampling program, class distinction device and class distinction system | |
CN106168980A (en) | Multimedia resource recommends sort method and device | |
Rabiee et al. | A biogeography-based optimisation algorithm for a realistic no-wait hybrid flow shop with unrelated parallel machines to minimise mean tardiness | |
CN110990624B (en) | Video recommendation method, device, equipment and storage medium | |
CN110555172A (en) | user relationship mining method and device, electronic equipment and storage medium | |
CN115455280A (en) | Recommendation list determining method and server | |
CN106257507A (en) | The methods of risk assessment of user behavior and device | |
CN112883275B (en) | Live broadcast room recommendation method, device, server and medium | |
CN116955808A (en) | Game recommendation method, device, electronic equipment and medium | |
CN110874639A (en) | Method and device for acquiring operation information | |
CN111507471B (en) | Model training method, device, equipment and storage medium | |
US20170004511A1 (en) | Identifying Drivers for a Metric-of-Interest |
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 |