CN110362753A - A kind of personalized neural network recommendation method and system based on user concealed feedback - Google Patents
A kind of personalized neural network recommendation method and system based on user concealed feedback Download PDFInfo
- Publication number
- CN110362753A CN110362753A CN201910283969.7A CN201910283969A CN110362753A CN 110362753 A CN110362753 A CN 110362753A CN 201910283969 A CN201910283969 A CN 201910283969A CN 110362753 A CN110362753 A CN 110362753A
- Authority
- CN
- China
- Prior art keywords
- user
- model
- ifnnrm
- recommendation information
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
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
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The personalized neural network recommendation method and system based on user concealed feedback that the invention discloses a kind of, the present invention is based on the hidden datas of user to establish implicit feedback neural network recommendation (IFNNRM) model, the user identifier that will acquire and the multiple message identifications to be recommended, it is input in established IFNNRM model, output obtains recommendation information, and user concealed data include the characteristic information of user, associated with the user want recommendation information and user and the associated behavioural information wanted between recommendation information of user.Since hidden data of the embodiment of the present invention when establishing IFNNRM model based on user is established, so being recommended using IFNNRM when recommending, so as to the hidden data based on user, being embodied as user's recommendation information.
Description
Technical field
The present invention relates to field of computer technology, in particular to a kind of personalized neural network based on user concealed feedback
Recommended method and system.
Background technique
With the fast development of Internet technology, Internet lateral root is that user's progress personalization pushes away according to user data
It recommends.During carrying out personalized recommendation, the history preference according to user and the user's history data of behavior are needed, is mentioned to user
For its interested recommendation information.In order to be embodied as user's recommendation information, there are content-based recommendation method, collaborative filtering at present
Recommended method, the recommended method based on correlation rule and combined recommendation method, are sketched below.
Content-based recommendation method is used to excavate user's Item Information for liking in the past, so as to excavate and in the past
The similar Item Information of the Item Information liked, recommends user as recommendation information.This method by commenting user in the past
The aspect of model for dividing description and the Item Information of article to establish user for user, in recommendation process, by the aspect of model of user
Match with the feature of new article information, so that scoring of the user to new article information is exported, as user's recommendation information.
Collaborative filtering recommending method is the score data by user to project, is got and target user or Item Information
Similar object is recommended as Candidate Recommendation information.There are two types of modes for collaborative filtering recommending method: the collaboration based on user
Filtered recommendation method and project-based collaborative filtering recommending method.Wherein based on the collaborative filtering recommending method of user are as follows: first
Scoring of the user for article is first obtained, the nearest-neighbors set of user, last benefit are then found by the similitude between user
It is completed to be Item Information that user recommends with the scoring of anticipation function and neighbor user set；Collaborative filtering based on article pushes away
Recommend method are as follows: the similar terms set of destination item is obtained according to similitude between project, passes through anticipation function and similar article collection
Close the Item Information list for being produced as user's recommendation.
Recommended method based on correlation rule is the user that analysis uses affairs X in affairs set of the user using article
In how many is also interested in affairs Y, to, based on analysis, be user's recommendation information during subsequent recommendation.This method
It is user's recommendation information according to the historical data of the correlation rule and user that generate in advance.
Combined recommendation method is a kind of method for combining a variety of recommended methods, and every kind that setting is respectively adopted pushes away
Recommend method, according to the historical data of user be user's recommendation information after, then organically combine, the recommendation information in conjunction with after,
As the final information recommended for user.
The input information of above-mentioned every kind of recommended method is the historical data of user respectively, is then based on the historical data of user
Export recommendation information.Wherein, above-mentioned every kind of recommended method is based on the historical data of user, and the historical data of user is all
What explicit way acquired, that is, history acquisition user is actively to the grading information of commenting of Item Information, and this explicitly obtain
The user's history data got can not be got in the case where there is many scenes, for example user does not actively carry out Item Information
Scoring etc..At this moment, using above-mentioned recommended method just cannot achieve as user's recommendation information.
Summary of the invention
In view of this, the embodiment of the present invention provides a kind of personalized neural network recommendation side based on user concealed feedback
Method, this method can be based on user concealed data, be embodied as user's recommendation information.
The embodiment of the present invention also provides a kind of personalized neural network recommendation system based on user concealed feedback, the system
User concealed data can be based on, user's recommendation information is embodied as.
The embodiments of the present invention are implemented as follows:
A kind of personalized neural network recommendation method based on user concealed feedback, this method comprises:
Hidden data based on user establishes implicit feedback neural network recommendation IFNNRM model；
The user identifier that will acquire and the multiple message identifications to be recommended, are input in established IFNNRM model,
Output obtains recommendation information.
The hidden data of the user includes the characteristic information of user, associated with the user wants recommendation information and user
And the associated behavioural information wanted between recommendation information of user.
The acquisition of the hidden data of the user includes:
Based on user identifier, the internet access log of user is obtained, is obtained from the internet access log of user
To user concealed data.
The IFNNRM model of establishing includes:
Training sample is constructed according to the hidden data of the user, is input in the input layer of IFNNRM model, the instruction
Practice sample to include the user's characteristic information indicated using vector, associated with the user want recommendation information and user and user's phase
The associated behavioural information wanted between recommendation information；
It is implicit that the embeding layer of IFNNRM model from the user's characteristic information vector of the input layer extracts to obtain user characteristics
Vector, and associated with the user recommendation information vector is wanted to extract to obtain and associated with the user implicit want recommendation information from described
Vector wants in the progress of recommendation information vector user's hidden feature information vector and associated with the user imply
Product；
The user characteristics of the embeding layer are implied into vector and associated with the user imply is wanted between recommendation information vector
Apposition is input in the convolutional neural networks of IFNNRM model setting and carries out convolutional calculation and Chi Huahou, and extraction obtains user's row
For hidden feature vector；
The user characteristics of the embeding layer are implied into vector, associated with the user imply wants recommendation information and the user
After hidden feature information vector and the implicit inner product for wanting recommendation information vector associated with the user carry out random initializtion,
It is input to the neural network for the degression type being arranged in IFNNRM model, extraction obtains user behavior hidden feature vector；
The user behavior hidden feature vector that output layer in IFNNRM model extracts convolutional neural networks with
And the user behavior hidden feature vector that extracts of degression type neural network spliced after be mapped as scoring, obtain recommending
Information whether recommended.
What the weight and the IFNNRM model used in the convolutional neural networks of the IFNNRM model setting was arranged passs
Subtract the weight that the neural network of formula uses, initial value is random, then passes through back-propagation algorithm renewal learning weight.
It constructs verifying sample and test sample respectively according to the hidden data of the user, is input to the IFNNRM model
In verified and tested, it is described respectively verifying sample and test sample in user concealed data in the training sample
At least occur primary.
In the user concealed data that will acquire, it is input to before established IFNNRM model, further includes:
According to the query intention of user, the recommendation information of wanting of matching user identifier, judgement are found from the database established
What the user generated in systems wants whether the behavior of recommendation information data is less than or equal to the recommendation threshold value set, if so, by institute
It states and wants recommendation information as the recommendation information for user；
If not, wanting recommendation information mark to be input to be established by what user identifier was respectively obtained with database matching
In IFNNRM model.
The database of the foundation is using word frequency/inverse document frequency TF/IDF algorithm setting；
It is described using it is described want recommendation information as the recommendation information for user before, further includes:
Using setting proposed algorithm to it is described want recommendation information to be ranked up after, then recommendation will be wanted described in after sequence
Breath is as the recommendation information for user.
A kind of personalized neural network recommendation system based on user concealed feedback, the system include: input unit, output
Unit, model foundation unit and model treatment unit, wherein
Model foundation unit establishes IFNNRM model for the hidden data based on user, is sent to model treatment unit；
Input unit, user identifier and the multiple message identifications to be recommended for will acquire, inputs to model treatment
Unit；
Model treatment unit is received for receiving to obtain established IFNNRM model and store from model foundation unit
The user identifier of input unit input and the multiple message identifications to be recommended, are input in the IFNNRM model of preservation and handle, defeated
The information recommended for user is obtained out；
Output unit is that the information that user recommends exports for that will obtain.
The model treatment unit, is also used to before being input to the IFNNRM model of preservation, comprising:
According to the query intention of user, the recommendation information of wanting of matching user identifier, judgement are found from the database established
What the user generated in systems wants whether the behavior of recommendation information data is less than or equal to the recommendation threshold value set, if so, by institute
It states and wants recommendation information as the recommendation information for user；If not, to be pushed away what user identifier was obtained with database matching respectively
It recommends message identification and is input in established IFNNRM model and handle.
As above as it can be seen that the present invention is based on the hidden datas of user to establish implicit feedback neural network recommendation (IFNNRM) mould
Type, the user identifier that will acquire and the multiple message identifications to be recommended, are input in established IFNNRM model, export
To recommendation information, user concealed data include the characteristic information of user, associated with the user want recommendation information and user and use
The associated behavioural information wanted between recommendation information in family.Since the embodiment of the present invention is based on user when establishing IFNNRM model
Hidden data establish, it is real so as to the hidden data based on user so recommended when recommending using IFNNRM
It is now user's recommendation information.
Detailed description of the invention
Fig. 1 is the personalized neural network recommendation method flow provided in an embodiment of the present invention based on user concealed feedback
Figure；
Fig. 2 is by the structural schematic diagram provided in an embodiment of the present invention for establishing IFNNRM model；
Fig. 3 is the personalized neural network recommendation method concrete example provided in an embodiment of the present invention based on user concealed feedback
Sub-process figure；
Fig. 4 is that the personalized neural network recommendation system structure provided in an embodiment of the present invention based on user concealed feedback is shown
It is intended to.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention more comprehensible, right hereinafter, referring to the drawings and the embodiments,
The present invention is further described.
From background technique as can be seen that existing recommended method is all based on the explicit data progress of user, user's
Explicit data is the collected user of history actively to the score information etc. of Item Information, and this user explicitly got goes through
History data can not be got in the case where there is many scenes.Therefore, the information that the explicit data when recommending based on user is recommended
Also inaccurate, or success can not be recommended.
Therefore, in order to solve this problem, implicit feedback neural network is established the present invention is based on the hidden data of user to push away
It recommends (IFNNRM) model, the user identifier that will acquire and the multiple message identifications to be recommended, is input to and is established
In IFNNRM model, output obtains recommendation information, and user concealed data include the characteristic information of user, associated with the user want
Recommendation information and the associated behavioural information wanted between recommendation information of user and user.
Since hidden data of the embodiment of the present invention when establishing IFNNRM model based on user is established, so recommending
When, recommended using IFNNRM, so as to the hidden data based on user, is embodied as user's recommendation information.
Specifically, the group of the hidden data of user becomes the characteristic information of user, associated with the user wants recommendation
Breath and user and the associated behavioural information wanted between recommendation information of user.Wherein, the characteristic information of user is going through for user
The information such as history preference and behavioural characteristic, the information etc. associated with the user for wanting recommendation information to select for user's history.User
And the associated behavioural information wanted between recommendation information of user may include the information that browses web sites of user, viewing video data
Duration thumbs up information and forwarding information etc., and the information that user's history selected can be Item Information or use that user bought
The knowledge information etc. that some selected application field of family provides, this does not need user initiatively to Item Information or user institute
The knowledge information that some application field for selecting provides scores, it is only necessary to which user goes to produce according to oneself interest active
Raw behavior, is then directly analyzed from generated behavior by network side and is obtained, and this mode will not allow user to generate dislike, is mentioned
User experience is risen.
In embodiments of the present invention, the hidden data process of user is obtained are as follows: arrange to the internet access log of user
And obtain.
Fig. 1 is the personalized neural network recommendation method flow provided in an embodiment of the present invention based on user concealed feedback
Figure, the specific steps are that:
Step 101, the hidden data based on user establish IFNNRM model；
Step 102, the user identifier that will acquire and the multiple message identifications to be recommended, are input to established IFNNRM
In model, output obtains recommendation information.
In the method, the hidden data of the user includes the characteristic information of user, associated with the user to recommend
Information and the associated behavioural information wanted between recommendation information of user and user.
In the method, the process of the user concealed data got includes:
Based on user identifier, the internet access log of user is obtained, is obtained from the internet access log of user
To user concealed data.
In embodiments of the present invention, the process for establishing IFNNRM model is as follows.
As shown in Fig. 2, Fig. 2 is by the structural schematic diagram provided in an embodiment of the present invention for establishing IFNNRM model.
Firstly, building is input to the training sample of IFNNRM model, it include acquired user characteristics in training sample
Information, it is associated with the user want recommendation information and user and the associated behavioural information wanted between recommendation information of user, use
Family characteristic information and associated with the user recommendation information is wanted to be input to training sample and established using vectorial
The input layer of IFNNRM model；
Herein, can using one-hot encoding (one-hot) coding mode generate user's characteristic information vector and with user's phase
It is associated to want recommendation information vector；
Secondly, extracting to obtain user characteristics from user's characteristic information vector in the embeding layer of established IFNNRM model
Implicit vector, from it is associated with the user want to extract in recommendation information vector obtain associated with the user implicit wanting recommendation information
The two is carried out inner product by vector, obtains the implicit vector of user characteristics and associated with the user imply wants recommendation information vector
Between inner product vector；
In this step, extract be exactly to user's characteristic information and the arrangement associated with the user for wanting recommendation information,
For example wanting recommendation information is menu information, it is determined that the menu information that user did and the menu information that do not did, respectively as
Positive sample and negative sample handle positive sample and negative sample for vector data form, and be divided into training set, verifying set and
Test combines, and verifies set and test the user concealed data in set and must occur once in training set；
Again, user behavior hidden feature vector extracts to obtain from two parts, a part using convolutional neural networks,
A part is to extract to obtain by the neural network of degression type；
Wherein, the characteristic procedure of user is extracted using convolutional neural networks are as follows: user characteristics imply vector sum and user's phase
Associated imply wants recommendation information vector to construct Interactive matrix by apposition.Convolutional layer is used to extract user characteristics and wants recommendation
Interaction feature between breath.One interaction featurePass through j-th of shared weightIt extracts, D is Interactive matrix, ws
Indicate window size, wherein * is convolution operator,It isBiasing, f is nonlinear activation function, such as formula
(1) shown in:
After have passed through convolution, pond layer extracts representative feature from convolutional layer, using maximum pond (max-
Pooling) mode,To indicate；
Using degression type neural network extract user behavior hidden feature vector process in, by user characteristics imply to
Amount associated with the user implicit wants recommendation information vector and user characteristics to imply vector and associated with the user implicit
After wanting the inner product vector between recommendation information vector to carry out random initializtion using Gaussian Profile mode, by these three part vectors
It is spliced into a vector, after being inputted, the calculating of the feature extraction of at least two layers neural network is carried out, is output to output layer；
Finally, in output layer, user behavior hidden feature vector and degression type that convolutional neural networks are extracted
The user behavior hidden feature vector that neural network extracts obtains splicing vector after being spliced, pass through the transformation of output layer
It is prediction scoring by splicing DUAL PROBLEMS OF VECTOR MAPPINGSuch as formula (2)
Wherein zL-1Indicate user behavior hidden feature vector and degression type nerve net that convolutional neural networks extract
The user behavior hidden feature vector that network extracts carries out spliced splicing vector,It indicates to do linear change to splicing vector
It changes, σ indicates sigmoid function, for extracting deeper feature, to obtain whether single recommendation information is recommended.
In this way, training sample is input in the IFNNRM model of above-mentioned foundation, in the training process, the setting of weighted value
It can be by functions such as Gaussian Profiles come random initializtion, by back-propagation algorithm come the weight of progressive updating IFNNRM model
Value so that IFNNRM model can gradually learn to user feature and user-association recommendation information between interaction
Relationship and the associated behavioural information wanted between recommendation information of user and user, obtain trained IFNNRM model.
After IFNNRM model is trained to, construction verifying sample and test sample, the verifying sample and the test specimens
Originally the user's characteristic information that includes, with user-association want recommendation information and user and user it is associated want recommendation information it
Between behavioural information, at least occurred in training sample primary, the IFNNRM model established tested.
The embodiment of the present invention disposes the IFNNRM model that test passes through, and IFNNRM model can will be recommended multiple
Information each confidence level (between 0 to 1) for wanting recommendation information is obtained by last sigmoid function.It is set by this
Reliability is ranked up the multiple information to be recommended according to confidence level, just can obtain the information for providing the recommendation of setting number.
In embodiments of the present invention, using IFNNRM model, the IFNNRM model is had the advantage that
First, to user's characteristic information vector and it is associated with the user want recommendation information vector carry out apposition processing, shape
At Interactive matrix between the two, convolutional neural networks are recycled to extract to obtain interaction feature on Interactive matrix；
Second, by obtained user concealed characteristic information vector, it is associated with the user implicitly want recommendation information vector, with
And user concealed characteristic information vector and it is associated with the user implicitly want this three parts of inner product between recommendation information vector into
It after row splicing, then is input in degression type neural network and is calculated, extraction obtains the interaction feature of these three parts；
Third spells the user behavior hidden feature vector that convolutional neural networks and degression type neural network are exported
It connects, the apposition and inner product etc. of implicit user feature and recommendation information associated with the user, to optimize IFNNRM model
Weight.
The user identifier that will acquire in step 102 described in Fig. 1 and multiple message identifications for wanting recommendation information, it is defeated
Enter into the IFNNRM model established, output obtains recommending the process of obtained Item Information are as follows:
From the user identifier of acquisition and the multiple message identifications to be recommended, obtain user characteristic information vector and with
User is associated to want recommendation information vector, is input in established IFNNRM model, after IFNNRM model is handled, pushes away
It recommends to obtain Item Information.
Herein, IFNNRM model it is associated with the user to the characteristic information vector of the user of input want recommendation information to
Amount is handled, and is handled in above-mentioned training process with IFNNRM model identical.
Fig. 3 is the personalized neural network recommendation method concrete example provided in an embodiment of the present invention based on user concealed feedback
Sub-process figure, the specific steps are that:
Step 301, the intent query recommendation request for receiving user；
Step 302 matches according to intent query recommendation request from the database established and to match with query intention
Want recommendation information；
Step 303 judges that the user generated in systems wants whether the behavior of recommendation information data is less than or equal to setting
Recommend threshold value, if it is, being transferred to step 304 execution；If it is not, then being transferred to step 305 execution；
Step 304 wants recommendation information as the recommendation information for user for described, is sent to user；
In this step, can also by the screening that recommend Item Information to carry out other set Generalization bounds and
After sequence, as the recommendation information for user, it is sent to user；
User identifier is input to trained IFNNRM model with multiple marks for wanting recommendation information respectively by step 305
Middle processing；
In this step, IFNNRM model can obtain the multiple information to be recommended by last sigmoid function
Each wants the confidence level (between 0 to 1) of recommendation information, by this confidence level to the multiple information to be recommended according to confidence level
It is ranked up, just can obtain the information for providing the recommendation of setting number；
After step 306, the information that the IFNNRM model established output is user's recommendation, it is sent to user.
During described in Fig. 3, the database established is using word frequency/inverse document frequency (TF/IDF) algorithm
It is arranged, wherein TF is frequency of each word in this inquiry in the inquiry of user, and the frequency of appearance is bigger, more related；
IDF is each word in the inquiry of user all frequencies for wanting to occur in recommendation information text in entire database, frequency occurs
Rate is bigger, more uncorrelated.And the degree of correlation is also related to the text size of recommendation information is wanted, and sentence is longer, and the degree of correlation is weaker.
It lifts a specific example and illustrates the embodiment of the present invention, in this example, it is assumed that the information that recommend user is dish
Spectrum information, process are as described below.
Firstly, receiving the recommendation request for carrying user identifier, based on the user identifier of the recommendation request, user is obtained
Hidden data, such as acquire user inquire cuisines if art, to the words art carry out semantic translation.Such as obtain user's
Art is talked about as " I wants to eat Guangdong snack ", then semantic translation is carried out by words art, the hidden data for obtaining user is that user wants to eat extensively
Eastern snack；
Secondly, the user is wanted the information input for eating Guangdong snack into set database, database carry out
Match, obtain multiple menu informations of matching Guangdong snack, returns；
Again, judge whether menu data behavior that the user generates in systems is greater than the recommendation threshold value of setting, if
It is the IFNNRM model then needed using being established；If it is not, then the multiple menus for the Guangdong snack for directly obtaining matching are believed
Breath output；
It, can also be using the recommendation of setting before the multiple menu informations output for the Guangdong snack for directly obtaining matching
Algorithm, for example random recommended method is used, the multiple menu informations for the Guangdong snack that matching obtains are ranked up, then the row of output
The multiple menu informations for the Guangdong snack that the matching after sequence obtains；
Finally, acquiring the user's characteristic information in the hidden data of user, and from database according to user identifier
Multiple menu informations of obtained matching Guangdong snack, are input in established IFNNRM model and handle, by what is established
IFNNRM model reorders to multiple menu informations of the matching Guangdong snack according to scoring height, returns in sequence
User wants multiple menu informations of recommendation information.
Fig. 4 is that the personalized neural network recommendation system structure provided in an embodiment of the present invention based on user concealed feedback is shown
It is intended to, comprising: input unit, output unit, model foundation unit and model treatment unit, wherein
Model foundation unit establishes IFNNRM model for the hidden data based on user, is sent to model treatment unit；
Input unit, user identifier and the multiple message identifications to be recommended for will acquire, inputs to model treatment
Unit；
Model treatment unit is received for receiving to obtain established IFNNRM model and store from model foundation unit
The user identifier of input unit input and the multiple message identifications to be recommended, are input in the IFNNRM model of preservation and handle, defeated
The information recommended for user is obtained out；
Output unit is that the information that user recommends exports for that will obtain.
The model treatment unit, is also used to before being input to the IFNNRM model of preservation, comprising:
According to the query intention of user, the recommendation information of wanting of matching user identifier, judgement are found from the database established
What the user generated in systems wants whether the behavior of recommendation information data is less than or equal to the recommendation threshold value set, if so, by institute
It states and wants recommendation information as the recommendation information for user；If not, to be pushed away what user identifier was obtained with database matching respectively
It recommends message identification and is input in established IFNNRM model and handle.
Scheme provided in an embodiment of the present invention is used to can be adapted for multiple application fields, such as sound for user's recommendation information
The recommendation of the information in fields such as happy, books or film.Using the embodiment of the present invention, it is only necessary to get the use of some application field
Family hidden data trains IFNNRM model, then user identifier is input to training with multiple marks for wanting recommendation information respectively
In good IFNNRM model, so that it may obtain the recommendation information in some application field.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent substitution, improvement and etc. done be should be included within the scope of the present invention.
Claims (10)
1. a kind of personalized neural network recommendation method based on user concealed feedback, which is characterized in that this method comprises:
Hidden data based on user establishes implicit feedback neural network recommendation IFNNRM model；
The user identifier that will acquire and the multiple message identifications to be recommended, are input in established IFNNRM model, output
Obtain recommendation information.
2. recommended method as described in claim 1, which is characterized in that the hidden data of the user includes the feature letter of user
Breath associated with the user wants recommendation information and user and the associated behavioural information wanted between recommendation information of user.
3. recommended method as claimed in claim 1 or 2, which is characterized in that the acquisition of the hidden data of the user includes:
Based on user identifier, the internet access log of user is obtained, acquires use from the internet access log of user
Family hidden data.
4. recommended method as claimed in claim 2, which is characterized in that the IFNNRM model of establishing includes:
Training sample is constructed according to the hidden data of the user, is input in the input layer of IFNNRM model, the trained sample
Originally include the user's characteristic information indicated using vector, associated with the user want recommendation information and user and user associated
The behavioural information wanted between recommendation information；
The embeding layer of IFNNRM model extracts to obtain the implicit vector of user characteristics from the user's characteristic information vector of the input layer,
And from it is described it is associated with the user want recommendation information vector extract to obtain it is associated with the user it is implicit want recommendation information vector,
Recommendation information vector is wanted to carry out inner product user's hidden feature information vector and associated with the user imply；
The user characteristics of the embeding layer are implied into vector and associated with the user imply wants apposition between recommendation information vector,
It is input in the convolutional neural networks of IFNNRM model setting and carries out convolutional calculation and Chi Huahou, it is implicit that extraction obtains user behavior
Feature vector；
The user characteristics of the embeding layer are implied into vector, associated with the user imply wants recommendation information and the user to imply
After characteristic information vector and the implicit inner product for wanting recommendation information vector associated with the user carry out random initializtion, input
The neural network for the degression type being arranged into IFNNRM model, extraction obtain user behavior hidden feature vector；
Output layer in IFNNRM model is by the user behavior hidden feature vector that convolutional neural networks extract and passs
Subtract after the user behavior hidden feature vector that formula neural network extracts is spliced and be mapped as scoring, obtains the letter to be recommended
Whether breath is recommended.
5. recommended method as claimed in claim 4, which is characterized in that in the convolutional neural networks of the IFNNRM model setting
The weight that the weight of use and the neural network of the degression type of IFNNRM model setting use, initial value is random, then leads to
Cross back-propagation algorithm renewal learning weight.
6. recommended method as claimed in claim 4, which is characterized in that construct verifying respectively according to the hidden data of the user
Sample and test sample are input in the IFNNRM model and are verified and tested, described respectively in verifying sample and test
User concealed data in sample at least occur primary in the training sample.
7. recommended method as claimed in claim 2, which is characterized in that in the user concealed data that will acquire, input
To before the IFNNRM model established, further includes:
According to the query intention of user, that finds matching user identifier from the database established wants recommendation information, judges the use
What family generated in systems wants the behavior of recommendation information data whether to be less than or equal to the recommendation threshold value of setting, if so, wanting by described in
Recommendation information is as the recommendation information for user；
If not, wanting recommendation information mark to be input to established IFNNRM for what user identifier was obtained with database matching respectively
In model.
8. recommended method as claimed in claim 7, which is characterized in that the database of the foundation is using word frequency/inverse text frequency
The setting of rate index TF/IDF algorithm；
It is described using it is described want recommendation information as the recommendation information for user before, further includes:
Using setting proposed algorithm to it is described want recommendation information to be ranked up after, then recommendation information will be wanted to make described in after sequence
For the recommendation information for user.
9. a kind of personalized neural network recommendation system based on user concealed feedback, which is characterized in that the system includes: input
Unit, output unit, model foundation unit and model treatment unit, wherein
Model foundation unit establishes IFNNRM model for the hidden data based on user, is sent to model treatment unit；
Input unit, user identifier and the multiple message identifications to be recommended for will acquire, inputs to model treatment unit；
Model treatment unit receives input for receiving to obtain established IFNNRM model and store from model foundation unit
The user identifier of unit input and the multiple message identifications to be recommended, are input in the IFNNRM model of preservation and handle, export
To the information recommended for user；
Output unit is that the information that user recommends exports for that will obtain.
10. system as claimed in claim 9, which is characterized in that the model treatment unit is also used to be input to preservation
Before IFNNRM model, comprising:
According to the query intention of user, that finds matching user identifier from the database established wants recommendation information, judges the use
What family generated in systems wants the behavior of recommendation information data whether to be less than or equal to the recommendation threshold value of setting, if so, wanting by described in
Recommendation information is as the recommendation information for user；If not, wanting recommendation for what user identifier was obtained with database matching respectively
Breath mark, which is input in established IFNNRM model, to be handled.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910283969.7A CN110362753B (en) | 2019-04-10 | 2019-04-10 | Personalized neural network recommendation method and system based on user implicit feedback |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910283969.7A CN110362753B (en) | 2019-04-10 | 2019-04-10 | Personalized neural network recommendation method and system based on user implicit feedback |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110362753A true CN110362753A (en) | 2019-10-22 |
| CN110362753B CN110362753B (en) | 2021-12-17 |
Family
ID=68214865
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910283969.7A Active CN110362753B (en) | 2019-04-10 | 2019-04-10 | Personalized neural network recommendation method and system based on user implicit feedback |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110362753B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110955775A (en) * | 2019-11-11 | 2020-04-03 | 南通大学 | Drawing book recommendation method based on implicit inquiry |
| CN111143686A (en) * | 2019-12-30 | 2020-05-12 | 北京百度网讯科技有限公司 | Resource 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 |
| CN112115378A (en) * | 2020-09-16 | 2020-12-22 | 长沙理工大学 | Recommendation prediction system and recommendation prediction method based on graph convolution collaborative filtering |
| CN112541131A (en) * | 2020-12-07 | 2021-03-23 | 东北大学 | Recommendation method based on multiple interest influences of neighbor users |
| CN113139120A (en) * | 2020-01-20 | 2021-07-20 | 佛山市顺德区美的电热电器制造有限公司 | Electronic equipment and recipe recommendation method and apparatus |
| CN113836439A (en) * | 2021-09-14 | 2021-12-24 | 上海任意门科技有限公司 | User matching method, computing device, and computer-readable storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180075353A1 (en) * | 2016-09-13 | 2018-03-15 | Sap Se | Method and system for cold start video recommendation |
| CN108536856A (en) * | 2018-04-17 | 2018-09-14 | 重庆邮电大学 | Mixing collaborative filtering film recommended models based on two aside network structure |
| CN108595527A (en) * | 2018-03-28 | 2018-09-28 | 中山大学 | A kind of personalized recommendation method and system of the multi-source heterogeneous information of fusion |
| CN109190030A (en) * | 2018-08-22 | 2019-01-11 | 南京工业大学 | Implicit feedback recommendation method fusing node2vec and deep neural network |
| CN109299370A (en) * | 2018-10-09 | 2019-02-01 | 中国科学技术大学 | Multipair grade personalized recommendation method |
-
2019
- 2019-04-10 CN CN201910283969.7A patent/CN110362753B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180075353A1 (en) * | 2016-09-13 | 2018-03-15 | Sap Se | Method and system for cold start video recommendation |
| CN108595527A (en) * | 2018-03-28 | 2018-09-28 | 中山大学 | A kind of personalized recommendation method and system of the multi-source heterogeneous information of fusion |
| CN108536856A (en) * | 2018-04-17 | 2018-09-14 | 重庆邮电大学 | Mixing collaborative filtering film recommended models based on two aside network structure |
| CN109190030A (en) * | 2018-08-22 | 2019-01-11 | 南京工业大学 | Implicit feedback recommendation method fusing node2vec and deep neural network |
| CN109299370A (en) * | 2018-10-09 | 2019-02-01 | 中国科学技术大学 | Multipair grade personalized recommendation method |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110955775A (en) * | 2019-11-11 | 2020-04-03 | 南通大学 | Drawing book recommendation method based on implicit inquiry |
| CN111143686A (en) * | 2019-12-30 | 2020-05-12 | 北京百度网讯科技有限公司 | Resource recommendation method and device |
| CN113139120A (en) * | 2020-01-20 | 2021-07-20 | 佛山市顺德区美的电热电器制造有限公司 | Electronic equipment and recipe recommendation method and apparatus |
| 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 |
| CN111814044B (en) * | 2020-06-30 | 2024-06-18 | 广州视源电子科技股份有限公司 | Recommendation method, recommendation device, terminal equipment and storage medium |
| CN112115378A (en) * | 2020-09-16 | 2020-12-22 | 长沙理工大学 | Recommendation prediction system and recommendation prediction method based on graph convolution collaborative filtering |
| CN112115378B (en) * | 2020-09-16 | 2022-04-19 | 长沙理工大学 | Recommendation prediction system and recommendation prediction method based on graph convolution collaborative filtering |
| CN112541131A (en) * | 2020-12-07 | 2021-03-23 | 东北大学 | Recommendation method based on multiple interest influences of neighbor users |
| CN112541131B (en) * | 2020-12-07 | 2021-10-29 | 东北大学 | Recommendation method based on multiple interest influences of neighbor users |
| CN113836439A (en) * | 2021-09-14 | 2021-12-24 | 上海任意门科技有限公司 | User matching method, computing device, and computer-readable storage medium |
| CN113836439B (en) * | 2021-09-14 | 2024-01-30 | 上海任意门科技有限公司 | User matching method, computing device, and computer-readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110362753B (en) | 2021-12-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111931062B (en) | Training method and related device of information recommendation model | |
| CN110362753A (en) | A kind of personalized neural network recommendation method and system based on user concealed feedback | |
| RU2745632C1 (en) | Automated response server device, terminal device, response system, response method and program | |
| Rana et al. | A study of the dynamic features of recommender systems | |
| CN111061946B (en) | Method, device, electronic equipment and storage medium for recommending scenerized content | |
| US10536580B2 (en) | Recommendations based on feature usage in applications | |
| CN111553754B (en) | Updating method and device of behavior prediction system | |
| CN105893609B (en) | A kind of mobile APP recommended method based on weighted blend | |
| CN101779180A (en) | Method and apparatus for context-based content recommendation | |
| US8645292B2 (en) | Serendipitous recommendations system and method | |
| CN111488524B (en) | Attention-oriented semantic-sensitive label recommendation method | |
| CN112131472A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN113051468B (en) | Movie recommendation method and system based on knowledge graph and reinforcement learning | |
| CN112257841A (en) | Data processing method, device and equipment in graph neural network and storage medium | |
| CN114201516B (en) | User portrait construction method, information recommendation method and related devices | |
| CN110795640B (en) | Self-adaptive group recommendation method for compensating group member difference | |
| CN113792212A (en) | Multimedia resource recommendation method, device, equipment and storage medium | |
| CN117216281A (en) | Knowledge graph-based user interest diffusion recommendation method and system | |
| US20170109411A1 (en) | Assisted creation of a search query | |
| CN116205700A (en) | Recommendation method and device for target product, computer equipment and storage medium | |
| US20130332440A1 (en) | Refinements in Document Analysis | |
| Ashraf et al. | Personalized news recommendation based on multi-agent framework using social media preferences | |
| CN116956183A (en) | Multimedia resource recommendation method, model training method, device and storage medium | |
| CN112989177A (en) | Information processing method, information processing device, electronic equipment and computer storage medium | |
| Deenadayalan et al. | User Feature Similarity Supported Collaborative Filtering for Page Recommendation Using Hybrid Shuffled Frog Leaping Algorithm. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |