CN109408729A - Material is recommended to determine method, apparatus, storage medium and computer equipment - Google Patents
Material is recommended to determine method, apparatus, storage medium and computer equipment Download PDFInfo
- Publication number
- CN109408729A CN109408729A CN201811482337.5A CN201811482337A CN109408729A CN 109408729 A CN109408729 A CN 109408729A CN 201811482337 A CN201811482337 A CN 201811482337A CN 109408729 A CN109408729 A CN 109408729A
- Authority
- CN
- China
- Prior art keywords
- user
- recommended
- characteristics vector
- factorization machine
- machine model
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention provides a kind of recommendation material and determines method, comprising: obtains the user characteristics vector of target user；The user characteristics vector is inputted to the index structure constructed in advance, material characteristics vector and the shortest preceding preset quantity of the user characteristics vector distance material to be recommended are determined based on nearest neighbor search；By the material characteristics vector input of the user characteristics vector, material to be recommended Factorization machine model trained in advance, the prediction result of Factorization machine model is obtained；According to the prediction result, the material recommended to the target user is determined from the material to be recommended.This method can reduce the calculation force request of Factorization machine, effectively mitigate computing relay and storage pressure, so that the sphere of action of Factorization machine increases, it is no longer limited to operate in relatively small content library, and the accuracy for guaranteeing that material is recommended is considered by multidimensional, it realizes from lesser content library and precisely recommends to the leap that huge volumes of content library is precisely recommended.
Description
Technical field
The present invention relates to proposed algorithm technical fields, specifically, the present invention relates to a kind of recommendation materials to determine method, dress
It sets, computer readable storage medium and computer equipment.
Background technique
With the fast development of Internet technology, internet makes the acquisition of information, the speed of propagation and scale reach empty
Preceding level realizes the information sharing and interaction in the whole world.In the age of information explosion, examined by proposed algorithm or information
Rope algorithm takes out content associated with the user in more data contents that can quickly comform.
Currently, existing some recommended methods can be in conjunction with the feature of user and material, to retrieve user preference or pass
The content of connection, but these methods are often limited to the speed of service and commercial cost of computer, can only operate in relatively small
In content library, it is difficult in the practical application scene applied to multiplicity；In addition, there are also some conventional methods, such as collaborative filtering, square
Battle array decomposes scheduling algorithm, may operate on huge volumes of content library, but cannot carry out personalization in conjunction with the feature of user and material
Screening, thus reduce the accuracy of recommendation.Therefore, it needs at present a kind of both in combination with the feature of user and material to guarantee to push away
The accuracy recommended, and have the recommendation side of lower calculating cost and the quick speed of service to operate on huge volumes of content library
Method.
Summary of the invention
It is that at least can solve above-mentioned one of technological deficiency, the present invention provides the recommendation materials of following technical scheme to determine
Method and corresponding device, computer readable storage medium and computer equipment.
The embodiment of the present invention provides a kind of recommendation material and determines method, include the following steps: according on one side
Obtain the user characteristics vector of target user；
The user characteristics vector is inputted into the index structure that constructs in advance, based on nearest neighbor search determine material characteristics to
Amount and the shortest preceding preset quantity of the user characteristics vector distance material to be recommended；
By the material characteristics vector input of the user characteristics vector, material to be recommended Factorization machine mould trained in advance
Type obtains the prediction result of Factorization machine model；
According to the prediction result, the material recommended to the target user is determined from the material to be recommended.
In one embodiment, by following steps, training obtains the Factorization machine model in advance:
The behavioral data that acquisition historical user executes material；
The behavioral data is analyzed, user characteristics and material characteristics are extracted；
The user characteristics and the material characteristics are input to Factorization machine model to be trained to be trained, are obtained
Factorization machine model after training.
In one embodiment, described that the user characteristics and the material characteristics are input to Factorization to be trained
Machine model is trained, the Factorization machine model after being trained, comprising:
The user characteristics and the material characteristics are input to Factorization machine model to be trained；
Based on stochastic gradient descent method, when solving the loss function of the Factorization machine model and reaching predetermined convergence condition
Corresponding model parameter；
Factorization machine model according to the model parameter, after being trained.
In one embodiment, described that the user characteristics and the material characteristics are input to Factorization to be trained
Machine model is trained, after the Factorization machine model after being trained, further includes:
The Factorization machine model is parsed, it will be according to the user characteristics and the corresponding user generated of the material characteristics
Feature vector and the separation of material characteristics vector, respectively obtain user characteristics vector sum material characteristics vector；
Save material characteristics vector described in the user characteristics vector sum.
In one embodiment, after material characteristics vector described in the preservation user characteristics vector sum, further includes:
Obtain the material characteristics vector；
Index structure is constructed in advance according to the material characteristics vector.
In one embodiment, the cross term of the Factorization machine model is the cross term of simplified processing, the letter
The cross term for changing processing omits user characteristics cross term and material characteristics cross term；
The formula of the cross term of the simplified processing are as follows:
Wherein, V ∈ Rn×k, < vi,vj> indicate that two sizes are the vector V of kiAnd VjInner product of vectors, xixjIt is characterized xiWith
Feature xjCombination, Σ V (user) × Σ V (material) be user characteristics vector sum material characteristics vector inner product of vectors.
In one embodiment, described according to the prediction result, it determines from the material to be recommended to the target
The material that user recommends, comprising:
Determine the value of the material characteristics cross term of the preset quantity material to be recommended；
According to the prediction result and described value, the predicted value of the preset quantity material to be recommended is calculated separately；
In the preset quantity material to be recommended, the maximum material to be recommended of the predicted value is determined as to described
The material that target user recommends.
In one embodiment, the material to be recommended includes: video, information, music, advertisement.
In one embodiment, it according to the prediction result, determines from the material to be recommended to the target user
After the material of recommendation, further includes:
Recommend the material to the target user.
In addition, the embodiment of the present invention provides a kind of recommendation material determining device according on the other hand, comprising:
Feature vector obtains module, for obtaining the user characteristics vector of target user；
Nearest neighbor search module, for the user characteristics vector to be inputted the index structure constructed in advance, based on nearest
Neighbour, which searches for, determines material characteristics vector and the shortest preceding preset quantity of the user characteristics vector distance material to be recommended；
Prediction module, for training the material characteristics vector input of the user characteristics vector, material to be recommended in advance
Factorization machine model, obtain the prediction result of Factorization machine model；
Material determining module, for determining from the material to be recommended and being used to the target according to the prediction result
The material that family is recommended.
The embodiment of the present invention provides a kind of computer readable storage medium, the computer according to another aspect
Computer program is stored on readable storage medium storing program for executing, the computer program realizes above-mentioned recommendation material when being executed by processor
Determine method.
The embodiment of the present invention provides a kind of computer equipment according to another aspect, and the computer includes one
Or multiple processors；Memory；One or more computer programs, wherein one or more of computer programs are stored in
It in the memory and is configured as being executed by one or more of processors, one or more of computer program configurations
For: it executes above-mentioned recommendation material and determines method.
Compared with the prior art, the present invention has the following beneficial effects:
Recommendation material provided by the invention determines method, apparatus, computer readable storage medium and computer equipment, passes through
In advance building index structure, determine material recommended to the user in conjunction with nearest neighbor search and Factorization machine, greatly reduce because
The calculation force request of sub- disassembler effectively mitigates computing relay and storage pressure, so that the sphere of action of Factorization machine increases, no
It is confined to operate in relatively small content library again, and the feature of Factorization machine combination user and material is predicted, lead to
It crosses multidimensional and considers the accuracy for guaranteeing that material is recommended, realize and precisely recommend to huge volumes of content library from lesser content library and precisely push away
The leap recommended.
In addition, the present invention is also carried out by the user characteristics vector sum material characteristics vector for generating Factorization machine model
It separates and is stored in advance to line server, material can be recommended to provide convenience for subsequent creation index structure and for target user,
To recommend efficiency to provide strong technical support to further increase material.
In addition, the present invention also passes through constructs index structure in advance, can recommend to provide fast quick checking when material for user to be subsequent
The condition of inquiry, when by can also significantly mitigate in off-line calculation server construction index structure and when being uploaded to line server
The computational burden of line server recommends efficiency to provide strong technical support to further increase the material of line server.
In addition, the present invention is also by the cross term of simplified Factorization machine model, and by the cross term meter of Factorization machine
Calculation is converted to the space length computational problem based on nearest neighbor search, further reduced the calculation force request of Factorization machine, is
The material for further increasing line server recommends efficiency to provide strong technical support.
The additional aspect of the present invention and advantage will be set forth in part in the description, these will become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Above-mentioned and/or additional aspect and advantage of the invention will become from the following description of the accompanying drawings of embodiments
Obviously and it is readily appreciated that, in which:
Fig. 1 is a kind of method flow diagram for recommending material to determine method provided in an embodiment of the present invention；
Fig. 2 is the method flow diagram of Factorization machine training method provided in an embodiment of the present invention；
Fig. 3 is the structure flow chart that recommendation material provided in an embodiment of the present invention determines system；
Fig. 4 is another method flow diagram for recommending material to determine method provided in an embodiment of the present invention；
Fig. 5 is a kind of structural schematic diagram for recommending material determining device provided in an embodiment of the present invention；
Fig. 6 is another structural schematic diagram for recommending material determining device provided in an embodiment of the present invention；
Fig. 7 is the structural schematic diagram of computer equipment provided in an embodiment of the present invention.
Specific embodiment
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, and for explaining only the invention, and is not construed as limiting the claims.
Those skilled in the art of the present technique are appreciated that unless expressly stated, singular " one " used herein, " one
It is a ", " described " and "the" may also comprise plural form.It is to be further understood that being arranged used in specification of the invention
Diction " comprising " refer to that there are the feature, integer, step, operation, element and/or component, but it is not excluded that in the presence of or addition
Other one or more features, integer, step, operation, element, component and/or their group.Wording used herein " and/
Or " it include one or more associated wholes for listing item or any cell and all combinations.
Those skilled in the art of the present technique are appreciated that unless otherwise defined, all terms used herein (including technology art
Language and scientific term), there is meaning identical with the general understanding of those of ordinary skill in fields of the present invention.Should also
Understand, those terms such as defined in the general dictionary, it should be understood that have in the context of the prior art
The consistent meaning of meaning, and unless idealization or meaning too formal otherwise will not be used by specific definitions as here
To explain.
The embodiment of the invention provides a kind of recommendation materials to determine method, as shown in Figure 1, this method comprises:
Step S110: the user characteristics vector of target user is obtained.
For the present embodiment, the target user is the online user currently recommended in the interior request material of application.The use
Family feature vector be user characteristics lower dimensional space indicate, the user characteristics include but is not limited to: user using interior ID, year
Age, gender, place city, using interior moon liveness, apart from it is previous log in application time interval.
For the present embodiment, line server pre-saves the user characteristics vector using interior each historical user, when
When target user requests material to be recommended, target user's client sends material recommendation request, line server to line server
The material recommendation request is received and responds, by preset user characteristics vector localization method, such as according to target user's
ID is quickly positioned from the user characteristics vector of mass users and is obtained the user characteristics vector of the target user.
In practical application scene, user can be by opening application, characterizing display using interior refresh page or click
The modes of operation such as control of more materials request material to be recommended.Wherein, the material can be video, information, music, advertisement
Etc. contents, it is not limited in the embodiment of the present invention.
The user characteristics vector: being inputted the index structure constructed in advance by step S120, is determined based on nearest neighbor search
Material characteristics vector and the shortest preceding preset quantity of the user characteristics vector distance material to be recommended.
For the present embodiment, line server also pre-saves the material characteristics vector using interior each historical user.
The material characteristics vector is that the lower dimensional space of material characteristics indicates that the material characteristics include but is not limited to: the application of material
Interior ID, stylistic category label, source.The index structure is preparatory according to the material characteristics vector of the interior each historical user of application
Built-up spatial data structure, the index structure can provide the online query efficiency of high speed.
For the present embodiment, nearest neighbor search (NNS, Nearest Neighbor Search) is used to tie in the index
The user characteristics vector arest neighbors searched and inputted in structure, the i.e. shortest material characteristics vector of space length.
When requiring to look up the preceding K data item with target data arest neighbors, using K nearest neighbor search (K-NN).Therefore make
For a preferred embodiment, the nearest neighbor search is specially K nearest neighbor search, described to determine material spy based on nearest neighbor search
Vector and the shortest preceding preset quantity of the user characteristics vector distance material to be recommended are levied, is specially searched based on K arest neighbors
Rope determines material characteristics vector and the shortest preceding preset quantity of the user characteristics vector distance material to be recommended, i.e., from described
The preset quantity material characteristics vector with the user characteristics vector arest neighbors, the preset number searched are searched in index structure
Measure the material to be recommended that a material characteristics vector corresponds to preset quantity.
For the present embodiment, the preset quantity can be any positive integers greater than 1 such as 5,10,20, art technology
Personnel can determine the specific value of the preset quantity according to practical application request, and the present embodiment does not limit this.
For nearest neighbor search, for inputting the user characteristics vector of index structure and the material searched in index structure
Correlation between space length between feature vector, with user characteristics vector sum material characteristics vector there are incidence relation,
The incidence relation specifically: space length between the two is closer, then correlation is bigger, also is understood as, if correlation is got over
Greatly, then between the two distance is closer.Based on the principle, it can be achieved that passing through nearest neighbor search quick obtaining from huge volumes of content library
With user characteristics vector correlation biggish preset quantity material to be recommended, may be pushed away to the target user with effectively reducing
The sample of material recommended reduces the workload of operation of Factorization machine model in subsequent step.
Step S130: by the input of the material characteristics vector of the user characteristics vector, material to be recommended it is trained in advance because
Sub- disassembler model, obtains the prediction result of Factorization machine model.
Traditional Factorization machine model is by by the continuous space of the Feature Mapping of user and material to low-dimensional, the factor
Disassembler portrays the correlation degree between user characteristics and material characteristics using the size of inner product, then the pass between all features
The summation of connection degree indicates user for the overall preference of material.
The formula for the Factorization machine model that binary is intersected is as follows:
Wherein, parameter x is vector, indicates the feature vector of a user and corresponding material, one feature of every one-dimensional representation
Numerical value；Y is real number, indicates the correlation of the user He the material, the more big then correlation maximum of y value；w0It is overall biasing,
W is vector, indicates the significance level of each feature；V is matrix, and every a line indicates the continuous low-dimensional mapping an of feature；<i,
vj> indicate that two sizes are the vector V of kiAnd VjInner product of vectors, xixjIt is characterized xiWith feature xjCombination.
By the prediction principle of analysis conventional Factorization machine model, the limitation of traditional Factorization machine can be summed up
Be, the correlation for needing successively to calculate target user with each material, so as to cause computing relay and storage pressure, it is difficult
To realize the correlation for calculating target user and all materials in huge volumes of content library.And in above-mentioned steps S120, based on nearest
Neighbour's search reduces the sample of material that need to calculate correlation in advance for Factorization machine model.
For the present embodiment, by determining preset quantity by the user characteristics vector of target user, based on nearest neighbor search
The material characteristics vector input of a material to be recommended Factorization machine model trained in advance, i.e., according to Factorization machine model pair
Correlation between the target user that the correlation calculations formula answered calculates separately and each material to be recommended calculates
The obtained each correlation i.e. prediction result of Factorization machine model.
Step S140: according to the prediction result, determination is recommended to the target user from the material to be recommended
Material.
For the present embodiment, according to the prediction result, to the counted target user of meter and each material to be recommended
Between correlation be ranked up by sequence from big to small, according to the ranking results of correlation from the material to be recommended really
Orient the material that the target user recommends.
Wherein, the material that Xiang Suoshu target user recommends can be one or more.
When recommending a material to the target user, foremost, i.e. correlation maximum will be come in ranking results
Corresponding material is determined as the material recommended to target user；
When recommending at least two materials to the target user, the material of front respective numbers will be come in ranking results
It is determined as the material recommended to target user.
Recommendation material provided in an embodiment of the present invention determines method, by constructing index structure in advance, searches in conjunction with arest neighbors
Rope and Factorization machine determine material recommended to the user, greatly reduce the calculation force request of Factorization machine, effectively mitigate meter
It calculates delay and storage pressure is no longer limited to operate in relatively small content so that the sphere of action of Factorization machine increases
On library, and the feature of Factorization machine combination user and material is predicted, is considered by multidimensional and is guaranteed that material is recommended accurate
Degree, realizes from lesser content library and precisely recommends to the leap that huge volumes of content library is precisely recommended.
In one embodiment, it according to the prediction result, determines from the material to be recommended to the target user
After the material of recommendation, further includes:
Recommend the material to the target user.
For the present embodiment, after line server determines the material recommended to the target user, Xiang Suoshu target
User client sends the identified material, so that material is shown in the respective application of target user's client.
In the present embodiment, recommend to determine material determined by method based on recommendation material provided by the invention to user,
It family can be used to be quickly obtained the material of its preference in huge volumes of content library, and then improve user to the user satisfaction of application.
In one embodiment, as shown in Fig. 2, the Factorization machine model is by following steps, training is obtained in advance:
Step S210: the behavioral data that acquisition historical user executes material.
For the present embodiment, the historical user indicates once to log in application in preset time and executes corresponding operating to material
User.Wherein, the preset time can be times, the present embodiment such as one hour, one day, one week, 30 days and not limit this
It is fixed.
The behavioral data that the historical user executes material includes but is not limited to: clicking, plays, forwards, thumbs up, comments on
Equal behaviors.
For example, when the material be short-sighted frequency when, then the behavioral data include: historical user click, play, forward it is short
Video such as is thumbed up, is commented on to short-sighted frequency at the behaviors.
In another example then the behavioral data includes: that historical user clicks, browsing, turns when the material is Domestic News
The behaviors such as hair, comment Domestic News.
Step S220: analyzing the behavioral data, extracts user characteristics and material characteristics.
For the present embodiment, the behavioral data of historical user collected is stored in historical user's behavior day by log system
Will rejects the hash in behavioral data by analyzing historical user's user behaviors log, and the behavioral data of acquisition is whole
Reason is user characteristics and material characteristics.
Step S230: the user characteristics and the material characteristics are input to Factorization machine model to be trained and are carried out
Training, the Factorization machine model after being trained.
For the present embodiment, in training Factorization machine model, obtained user characteristics will be arranged and material characteristics are defeated
Enter into Factorization machine model to be trained, by successively repetitive exercise can be trained the Factorization machine model is suitable
Model parameter, thus the Factorization machine model after being trained.
In other embodiments, behavioral data that can also be nearest by timing acquiring historical user, to Factorization machine
The model parameter of model is adjusted, timing updating factor disassembler.
For example, can answered per acquiring in historical user nearly one hour every other hour in the application scenarios of short video recommendations
With the click of interior execution, broadcasting, the short-sighted frequency of forwarding, the behavioral datas such as thumbed up, commented on to short-sighted frequency, to the behavior number of acquisition
According to being analyzed to extract user characteristics and short video features, and the user characteristics and short video features are input to Factor minute
It in solution machine model, is adjusted with the model parameter to Factorization machine model, obtains the Factorization machine model that timing updates.
In one embodiment, step S230 " by the user characteristics and the material characteristics be input to it is to be trained because
Sub- disassembler model is trained, the Factorization machine model after being trained ", comprising:
Step S231: the user characteristics and the material characteristics are input to Factorization machine model to be trained；
Step S232: being based on stochastic gradient descent method, and the loss function for solving the Factorization machine model reaches predetermined
Corresponding model parameter when the condition of convergence；
Step S233: the Factorization machine model according to the model parameter, after being trained.
For the present embodiment, by the loss function of defining factor disassembler model, constantly by the user characteristics and institute
It states material characteristics iteration and is input to Factorization machine model to be trained, and changed step by step based on stochastic gradient descent method SGD
The loss function that generation solves the Factorization machine model reaches predetermined convergence condition, that is, minimizes corresponding mould when loss function
Shape parameter completes the training of Factorization machine model, the Factorization machine model after being trained.
In one embodiment, described that the user characteristics and the material characteristics are input to Factorization to be trained
Machine model is trained, after the Factorization machine model after being trained, further includes:
The Factorization machine model is parsed, it will be according to the user characteristics and the corresponding user generated of the material characteristics
Feature vector and the separation of material characteristics vector, respectively obtain user characteristics vector sum material characteristics vector；
Save material characteristics vector described in the user characteristics vector sum.
For the present embodiment, during training Factorization machine model, Factorization machine model can learning model it is defeated
The low-dimensional continuous representation for entering feature automatically generates corresponding use according to the user characteristics of the historical user inputted and material characteristics
Family feature vector and material characteristics vector, by parsing Factorization machine model, can by user characteristics vector and material characteristics to
Amount separation respectively obtains user characteristics vector sum material characteristics vector, and user characteristics vector sum material characteristics vector is stored
To line server.
Specifically, the separation that user characteristics vector sum material characteristics vector can be realized based on the mode manually marked, is passed through
To in behavioral data collected user characteristics and material characteristics be labeled by feature ownership, such as gender is labeled as using
Stylistic category is labeled as material label by family feature, so that Factorization machine model is generating user characteristics vector sum material spy
When levying vector, user characteristics vector sum material characteristics vector can be automatically separated based on the mark of input feature vector.
In the present embodiment, it is carried out by the user characteristics vector sum material characteristics vector for generating Factorization machine model
It separates and is stored in advance to line server, material can be recommended to provide convenience for subsequent creation index structure and for target user,
To recommend efficiency to provide strong technical support to further increase material.
In one embodiment, after material characteristics vector described in the preservation user characteristics vector sum, further includes:
Obtain the material characteristics vector；
Index structure is constructed in advance according to the material characteristics vector.
For the present embodiment, index structure supports quick search, therefore line server can obtain the history that it is preserved and use
The material characteristics vector at family, and the material characteristics vector that will acquire is ranked up by preset putting in order, building is indexed
Structure.
In the practical application scene for recommending material to target user, the index structure constructed in advance can be used for institute
The user characteristics vector of the target user of input carries out quick search, based on nearest neighbor search determine material characteristics vector with it is described
User characteristics vector distance shortest preceding preset quantity material to be recommended.
In the present embodiment, by constructing index structure in advance, can recommend to provide fast quick checking when material for user to be subsequent
The condition of inquiry recommends efficiency to provide strong technical support to further increase material.
As a preferred embodiment, described the step of index structure is constructed according to the material characteristics vector in advance from
It is completed in line computation server.
As shown in figure 3, determining the structural schematic diagram of system for the recommendation material of one embodiment, the recommendation material is determined
System includes user client 31, server 32, wherein the server 32 specifically includes online server 321 and offline meter
Server 322 is calculated, which is not necessarily referring to the equipment that can be seen in objective world, and refers to recommendation provided in an embodiment of the present invention
Expect to determine that realizing for method is by what user client 31, line server 321 and off-line calculation server 322 were constituted
System.Specified application is installed in user client 321, line server 322 be used for based on the determination of Factorization machine model to
The material that user recommends, in the present embodiment, off-line calculation server from line server for obtaining what it was pre-saved
Material characteristics vector, and according to the material characteristics vector off-line calculation index structure, after index structure building is completed on
Reach line server.
In the present embodiment, it by off-line calculation server construction index structure and being uploaded to line server, can show
The computational burden for mitigating line server is write, recommends efficiency to provide strong technology to further increase the material of line server
It supports.
In one embodiment, the cross term of the Factorization machine model is the cross term of simplified processing, the letter
The cross term for changing processing omits user characteristics cross term and material characteristics cross term；
The formula of the cross term of the simplified processing are as follows:
Wherein, V ∈ Rn×k, < vi,vj> indicate that two sizes are the vector V of kiAnd VjInner product of vectors, xixjIt is characterized xiWith
Feature xjCombination, Σ V (user) × Σ V (material) be user characteristics vector sum material characteristics vector inner product of vectors.
In the formula for the Factorization machine model that binary is intersected,It is ripe for those skilled in the art
The linear model known,For the cross term of Factorization machine model.
For the present embodiment, for the calculation force request for further decreasing Factorization machine model, to Factorization machine model
Cross term is simplified.
Wherein, the cross term of Factorization machine model simplifies process specifically:
A, according to being characterized in that belonging to user still belongs to material, by the cross term of Factorization machine model be organized into as
Lower form:
Wherein, V ∈ Rn×k, < vi,vj> indicate that two sizes are the vector V of kiAnd VjInner product of vectors, xixjIt is characterized xiWith
Feature xjCombination, Σ V (user) × Σ V (material) be user characteristics vector sum material characteristics vector inner product of vectors.
For example, when the material is short-sighted frequency, and x is five dimensional vectors (User ID, user country, video ID, video
Author, if dancing), then corresponding cross term can arrange are as follows:
Cross term=(V (User ID)+V (user country)) × (V (video ID)+V (video author)+V (whether dancing))
+ video features inner product+user characteristics inner product two-by-two two-by-two
B, in view of practical material is recommended in application scenarios, the target user only recommended online request material, which determines, to be recommended
Material can add the same value to the predictor calculation of any material, substantial user characteristics are handed over that is, for the same user
It pitches and effect is had no to the prediction result for calculating recommendation material, therefore negligible user characteristics cross term, cross term further arrange are as follows:
It c, is that can be determined from the material to be recommended to the mesh described according to the prediction result convenient for calculating
Ignore article cross term before the step of material that mark user recommends, and in the step for determining the material recommended to the target user
The article cross term is rethought to retrieve loss of significance in rapid, and cross term can be arranged further as the simplified processing
Cross term:
D, the cross term of the simplified processing as obtained in step c is it is found that Σ V (user) × Σ V (article) is indicated
The inner product of vectors of user characteristics vector sum material characteristics vector, for the cross term of the simplified processing, when the inner product of vectors
When maximum, the value of the cross term of the Factorization machine model is maximum, i.e., in huge volumes of content library, Factorization machine model is to working as
The predicted value of preceding material is maximum.Therefore the completion of the cross term of Factorization machine model is converted into inner product maximum search, by further
Define distance function:
Distance (X, Y)=1-InnerProduct (X, Y)
Therefore, the nearest neighbor search problem that inner product maximum search can be converted to nonmetric space, by by user characteristics
Vector is input in the index structure constructed in advance based on material characteristics vector and carries out nearest neighbor search, when sky between the two
Between distance it is shorter, then user and the correlation of material are bigger.In the step s 120, material characteristics are determined based on nearest neighbor search
While vector and the shortest preceding preset quantity of the user characteristics vector distance material to be recommended, also it can be obtained described preset
The distance between material characteristics vector and the user characteristics vector of quantity material to be recommended are worth, can be by distance function
One step calculate the cross term of Factorization machine model value, and then quickly obtain the prediction result of Factorization machine model.
In the present embodiment, by simplifying the cross term of Factorization machine model, and by the cross term meter of Factorization machine
Calculation is converted to the space length computational problem based on nearest neighbor search, further reduced the calculation force request of Factorization machine, is
The material for further increasing line server recommends efficiency to provide strong technical support.
In one embodiment, described according to the prediction result, it determines from the material to be recommended to the target
The material that user recommends, comprising:
Determine the value of the material characteristics cross term of the preset quantity material to be recommended；
According to the prediction result and described value, the predicted value of the preset quantity material to be recommended is calculated separately；
In the preset quantity material to be recommended, the maximum material to be recommended of the predicted value is determined as to described
The material that target user recommends.
The value meter in cross term is rethought when determining the material recommended to the target user for the present embodiment
The article cross term ignored when calculation, to retrieve the loss of significance of calculating.
Wherein, the value of the material characteristics cross term can be calculated by line server or off-line calculation server.
Line server can determine material characteristics vector and the user characteristics to span described based on nearest neighbor search
It is precalculated later from shortest preceding preset quantity material to be recommended or need to determine the preset quantity object to be recommended
The material characteristics cross term of the preset quantity material to be recommended is calculated when the value of the material characteristics cross term of material.
In addition, line server determines material characteristics vector and the user characteristics vector based on nearest neighbor search described
After shortest preceding preset quantity material to be recommended can by the material characteristics of the preset quantity material to be recommended to
Amount is sent to off-line calculation server, and offline service device calculates each material characteristics vector pair after receiving material characteristics vector
The material characteristics cross term answered, and the material characteristics cross term that off-line calculation obtains is uploaded to line server.By offline
Calculation server, which calculates material characteristics cross term, can further reduced the calculation force request of Factorization machine, to further increase
The material of line server recommends efficiency to provide strong technical support.
After determining the value of material characteristics cross term of preset quantity material to be recommended, obtains and be based on simplified processing
The obtained prediction result of Factorization machine model, i.e., user that the Factorization machine model of simplified processing is calculated and pre-
The correlation between quantity material to be recommended is set, each correlation is added with the value of corresponding material characteristics cross term, respectively
The predicted value of the preset quantity material to be recommended is calculated, and according to the predicted value, is determined from the material to be recommended
The material recommended to the target user.Specifically, according to the predicted value, again to the counted target user of meter and respectively
Correlation between a material to be recommended is ranked up by sequence from big to small, according to the ranking results of correlation from it is described to
Recommend to determine the material recommended to the target user in material.
Wherein, the material that Xiang Suoshu target user recommends can be one or more.
It, will be described pre- in the preset quantity material to be recommended when recommending a material to the target user
The maximum material to be recommended of measured value is determined as the material recommended to the target user；
When recommending at least two materials to the target user, in the preset quantity material to be recommended, by institute
It states the material to be recommended of predicted value maximum at least two and is determined as the material recommended to the target user.
In the present embodiment, by considering the value of material characteristics cross term to the Factorization machine model of simplified processing
Prediction result is modified, and to retrieve the precision of prediction loss of Factorization machine model, guarantees the accuracy that material is recommended.
Recommend material to determine method to more fully understand, is introduced below with reference to a specific embodiment.
As shown in figure 4, determining the flow diagram of method, the recommendation material determination side for the recommendation material of one embodiment
Method includes:
S1, user execute relevant operation to APP, specifically, carrying out some clicking, playing, thumbing up, commenting in APP
Equal behaviors.
S2, log system will collect the behavioral data of historical user, and garbage is rejected, arrange become user characteristics and
Material characteristics are input to machine learning system.
S3, pass through gradient descent method, the low-dimensional continuous representation of usage factor disassembler model learning feature.
S4, the model by parsing Factorization machine, by the model parameter of the model parameter of user characteristics and material characteristics
Separation, the model parameter are exactly that the lower dimensional space of feature indicates, i.e. user characteristics vector sum material characteristics vector, and store and arrive
Line server.
S5, material model is pre-entered into nearest neighbor search system, is indexed building using nearest neighbor search system, is
Quick search later creates conditions.
S6, when user requests material again, the model parameter of the user is inputted into nearest neighbor search system, solves and uses
Model parameter most similar material in family is as recommendation results.
In addition, the embodiment of the invention provides a kind of recommendation material determining devices, as shown in figure 5, described device includes: spy
It levies vector and obtains module 51, nearest neighbor search module 52, prediction module 53 and material determining module 54；Wherein,
Described eigenvector obtains module 51, for obtaining the user characteristics vector of target user；
The nearest neighbor search module 52, for the user characteristics vector to be inputted the index structure constructed in advance, base
Material characteristics vector and the shortest preceding preset quantity of the user characteristics vector distance object to be recommended are determined in nearest neighbor search
Material；
The prediction module 53, it is pre- for inputting the material characteristics vector of the user characteristics vector, material to be recommended
First trained Factorization machine model, obtains the prediction result of Factorization machine model；
The material determining module 54, for being determined to described from the material to be recommended according to the prediction result
The material that target user recommends.
In one embodiment, by following steps, training obtains the Factorization machine model in advance:
The behavioral data that acquisition historical user executes material；
The behavioral data is analyzed, user characteristics and material characteristics are extracted；
The user characteristics and the material characteristics are input to Factorization machine model to be trained to be trained, are obtained
Factorization machine model after training.
In one embodiment, described that the user characteristics and the material characteristics are input to Factorization to be trained
Machine model is trained, the Factorization machine model after being trained, comprising:
The user characteristics and the material characteristics are input to Factorization machine model to be trained；
Based on stochastic gradient descent method, when solving the loss function of the Factorization machine model and reaching predetermined convergence condition
Corresponding model parameter；
Factorization machine model according to the model parameter, after being trained.
In one embodiment, described that the user characteristics and the material characteristics are input to Factorization to be trained
Machine model is trained, after the Factorization machine model after being trained, further includes:
The Factorization machine model is parsed, it will be according to the user characteristics and the corresponding user generated of the material characteristics
Feature vector and the separation of material characteristics vector, respectively obtain user characteristics vector sum material characteristics vector；
Save material characteristics vector described in the user characteristics vector sum.
In one embodiment, after material characteristics vector described in the preservation user characteristics vector sum, further includes:
Obtain the material characteristics vector；
Index structure is constructed in advance according to the material characteristics vector.
In one embodiment, the cross term of the Factorization machine model is the cross term of simplified processing, the letter
The cross term for changing processing omits user characteristics cross term and material characteristics cross term；
The formula of the cross term of the simplified processing are as follows:
Wherein, V ∈ Rn×k, < vi,vj> indicate that two sizes are the vector V of kiAnd VjInner product of vectors, xixjIt is characterized xiWith
Feature xjCombination, Σ V (user) × Σ V (material) be user characteristics vector sum material characteristics vector inner product of vectors.
In one embodiment, the material determining module 54, is specifically used for:
Determine the value of the material characteristics cross term of the preset quantity material to be recommended；
According to the prediction result and described value, the predicted value of the preset quantity material to be recommended is calculated separately；
In the preset quantity material to be recommended, the maximum material to be recommended of the predicted value is determined as to described
The material that target user recommends.
In one embodiment, the material to be recommended includes: video, information, music, advertisement.
In one embodiment, as shown in fig. 6, for another recommendation material determining device provided in an embodiment of the present invention.
The recommendation material determining device, further includes: material recommending module 55；Wherein,
The material recommending module 55, for recommending the material to the target user.
Recommendation material determining device provided by the invention can be realized: by constructing index structure in advance, search in conjunction with arest neighbors
Rope and Factorization machine determine material recommended to the user, greatly reduce the calculation force request of Factorization machine, effectively mitigate meter
It calculates delay and storage pressure is no longer limited to operate in relatively small content so that the sphere of action of Factorization machine increases
On library, and the feature of Factorization machine combination user and material is predicted, is considered by multidimensional and is guaranteed that material is recommended accurate
Degree, realizes from lesser content library and precisely recommends to the leap that huge volumes of content library is precisely recommended.It can also be achieved: by by the factor
The user characteristics vector sum material characteristics vector that disassembler model generates is separated and is stored in advance to line server, can be
Subsequent creation index structure and material is recommended to provide convenience for target user, to recommend efficiency to be provided with to further increase material
The technical support of power；By constructing index structure in advance, can recommend to provide the condition of quick search when material for user to be subsequent,
When by can also significantly mitigate line server in off-line calculation server construction index structure and when being uploaded to line server
Computational burden, for further increase line server material recommend efficiency strong technical support is provided；By simplify because
The cross term of sub- disassembler model, and the calculating of the cross term of Factorization machine is converted into the space length based on nearest neighbor search
Computational problem further reduced the calculation force request of Factorization machine, recommend effect to further increase the material of line server
Rate provides strong technical support.
The embodiment of the method provided in an embodiment of the present invention for recommending material determining device that above-mentioned offer may be implemented, specific function
It is able to achieve the explanation referred in embodiment of the method, details are not described herein.
In addition, being deposited on computer readable storage medium the embodiment of the invention provides a kind of computer readable storage medium
Computer program is contained, recommendation material determination side described in above embodiments is realized when the computer program is executed by processor
Method.Wherein, the computer readable storage medium includes but is not limited to any kind of disk (including floppy disk, hard disk, CD, CD-
ROM and magneto-optic disk), ROM (Read-Only Memory, read-only memory), RAM (Random AcceSS Memory, immediately
Memory), EPROM (EraSable Programmable Read-Only Memory, Erarable Programmable Read only Memory),
(Electrically EraSable Programmable Read-Only Memory, electric erazable programmable is read-only to be deposited EEPROM
Reservoir), flash memory, magnetic card or light card.It is, storage equipment includes by equipment (for example, computer, mobile phone) with energy
Any medium for the form storage or transmission information enough read can be read-only memory, disk or CD etc..
Computer readable storage medium provided by the invention, it can be achieved that: by constructing index structure in advance, in conjunction with arest neighbors
Search and Factorization machine determine material recommended to the user, greatly reduce the calculation force request of Factorization machine, effectively mitigate
Computing relay and storage pressure are no longer limited to operate in relatively small so that the sphere of action of Factorization machine increases
On Rong Ku, and the feature of Factorization machine combination user and material is predicted, considers the essence for guaranteeing that material is recommended by multidimensional
Exactness realizes from lesser content library and precisely recommends to the leap that huge volumes of content library is precisely recommended.Can also be achieved: by by because
The user characteristics vector sum material characteristics vector that sub- disassembler model generates is separated and is stored in advance to line server, can
Material is recommended to provide convenience for subsequent creation index structure and for target user, to recommend efficiency to provide to further increase material
Strong technical support；By constructing index structure in advance, can recommend to provide the item of quick search when material for user to be subsequent
Part, when by can also significantly mitigate online clothes in off-line calculation server construction index structure and when being uploaded to line server
The computational burden of business device recommends efficiency to provide strong technical support to further increase the material of line server；Pass through letter
Change the cross term of Factorization machine model, and the calculating of the cross term of Factorization machine is converted into the space based on nearest neighbor search
Apart from computational problem, the calculation force request of Factorization machine further reduced, push away to further increase the material of line server
It recommends efficiency and strong technical support is provided.
The embodiment of the method for above-mentioned offer may be implemented in computer readable storage medium provided in an embodiment of the present invention, specifically
Function realizes the explanation referred in embodiment of the method, and details are not described herein.
In addition, the embodiment of the invention also provides a kind of computer equipments, as shown in Figure 7.Calculating described in the present embodiment
Machine equipment can be the equipment such as server, personal computer and the network equipment.The computer equipment include processor 702,
The devices such as memory 703, input unit 704 and display unit 705.It will be understood by those skilled in the art that setting shown in Fig. 7
Standby structure devices do not constitute the restriction to all devices, may include components more more or fewer than diagram, or combine certain
A little components.Memory 703 can be used for storing computer program 701 and each functional module, and the operation of processor 702 is stored in storage
The computer program 701 of device 703, thereby executing the various function application and data processing of equipment.Memory can be memory
Reservoir or external memory, or including both built-in storage and external memory.Built-in storage may include read-only memory
(ROM), programming ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash
Device or random access memory.External memory may include hard disk, floppy disk, ZIP disk, USB flash disk, tape etc..It is disclosed in this invention to deposit
Reservoir includes but is not limited to the memory of these types.Memory disclosed in this invention is only used as example rather than as restriction.
Input unit 704 is used to receive the input of signal, and receives the keyword of user's input.Input unit 704 can
Including touch panel and other input equipments.Touch panel collects the touch operation of user on it or nearby and (for example uses
Family uses the operations of any suitable object or attachment on touch panel or near touch panel such as finger, stylus), and root
According to the corresponding attachment device of preset driven by program；Other input equipments can include but is not limited to physical keyboard, function
One of key (such as broadcasting control button, switch key etc.), trace ball, mouse, operating stick etc. are a variety of.Display unit
705 can be used for showing the information of user's input or be supplied to the information of user and the various menus of computer equipment.Display is single
The forms such as liquid crystal display, Organic Light Emitting Diode can be used in member 705.Processor 702 is the control centre of computer equipment, benefit
With the various pieces of various interfaces and the entire computer of connection, by running or executing the software being stored in memory 702
Program and/or module, and the data being stored in memory are called, perform various functions and handle data.
As one embodiment, the computer equipment includes: one or more processors 702, memory 703, and one
Or multiple computer programs 701, wherein one or more of computer programs 701 are stored in memory 703 and are matched
It is set to and is executed by one or more of processors 702, one or more of computer programs 701 are configured to carry out above
Recommendation material described in any embodiment determines method.
Computer equipment provided by the invention, it can be achieved that: by constructing index structure in advance, in conjunction with nearest neighbor search and because
Sub- disassembler determines material recommended to the user, greatly reduces the calculation force request of Factorization machine, effectively mitigation computing relay
It is no longer limited to operate in relatively small content library so that the sphere of action of Factorization machine increases with storage pressure, and
The feature of Factorization machine combination user and material is predicted, is considered the accuracy for guaranteeing that material is recommended by multidimensional, is realized
From lesser content library precisely recommend to the leap that huge volumes of content library is precisely recommended.It can also be achieved: by by Factorization machine
The user characteristics vector sum material characteristics vector that model generates is separated and is stored in advance to line server, can be subsequent wound
It indexes structure and recommends material to provide convenience for target user, to recommend efficiency to provide strong skill to further increase material
Art is supported；By constructing index structure in advance, can recommend to provide the condition of quick search when material for user to be subsequent, when passing through
In off-line calculation server construction index structure and when being uploaded to line server, it can also significantly mitigate the operation of line server
Burden recommends efficiency to provide strong technical support to further increase the material of line server；By simplifying Factorization
The cross term of machine model, and the calculating of the cross term of Factorization machine is converted into the calculating of the space length based on nearest neighbor search and is asked
Topic, further reduced the calculation force request of Factorization machine, recommend efficiency to provide to further increase the material of line server
Strong technical support.
The embodiment of the method for above-mentioned offer may be implemented in computer equipment provided in an embodiment of the present invention, and concrete function is realized
The explanation in embodiment of the method is referred to, details are not described herein.
It, can also be in addition, each functional unit in each embodiment of the present invention can integrate in a processing module
It is that each unit physically exists alone, can also be integrated in two or more units in a module.Above-mentioned integrated mould
Block both can take the form of hardware realization, can also be realized in the form of software function module.The integrated module is such as
Fruit is realized and when sold or used as an independent product in the form of software function module, also can store in a computer
In read/write memory medium.
The above is only some embodiments of the invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (12)
1. a kind of recommendation material determines method, which comprises the steps of:
Obtain the user characteristics vector of target user；
The user characteristics vector is inputted into the index structure that constructs in advance, based on nearest neighbor search determine material characteristics vector with
The shortest preceding preset quantity of user characteristics vector distance material to be recommended；
The Factorization machine model that the material characteristics vector input of the user characteristics vector, material to be recommended is trained in advance,
Obtain the prediction result of Factorization machine model；
According to the prediction result, the material recommended to the target user is determined from the material to be recommended.
2. recommendation material according to claim 1 determines method, which is characterized in that the Factorization machine model by with
Training obtains lower step in advance:
The behavioral data that acquisition historical user executes material；
The behavioral data is analyzed, user characteristics and material characteristics are extracted；
The user characteristics and the material characteristics are input to Factorization machine model to be trained to be trained, are trained
Factorization machine model afterwards.
3. recommendation material according to claim 2 determines method, which is characterized in that described by the user characteristics and described
Material characteristics are input to Factorization machine model to be trained and are trained, the Factorization machine model after being trained, comprising:
The user characteristics and the material characteristics are input to Factorization machine model to be trained；
Based on stochastic gradient descent method, the loss function for solving the Factorization machine model reaches corresponding when predetermined convergence condition
Model parameter；
Factorization machine model according to the model parameter, after being trained.
4. recommendation material according to claim 2 determines method, which is characterized in that described by the user characteristics and described
Material characteristics are input to Factorization machine model to be trained and are trained, after the Factorization machine model after being trained,
Further include:
The Factorization machine model is parsed, it will be according to the user characteristics and the corresponding user characteristics generated of the material characteristics
The separation of vector sum material characteristics vector, respectively obtains user characteristics vector sum material characteristics vector；
Save material characteristics vector described in the user characteristics vector sum.
5. recommendation material according to claim 4 determines method, which is characterized in that described to save the user characteristics vector
After the material characteristics vector, further includes:
Obtain the material characteristics vector；
Index structure is constructed in advance according to the material characteristics vector.
6. recommendation material according to any one of claims 1 to 5 determines method, which is characterized in that the Factorization machine
The cross term of model is the cross term of simplified processing, and the cross term of the simplified processing omits user characteristics cross term and material
Characteristic crossover item；
The formula of the cross term of the simplified processing are as follows:
Wherein, V ∈ Rn×k, < vi,vj> indicate that two sizes are the vector V of kiAnd VjInner product of vectors, xixjIt is characterized xiAnd feature
xjCombination, Σ V (user) × Σ V (material) be user characteristics vector sum material characteristics vector inner product of vectors.
7. recommendation material according to claim 6 determines method, which is characterized in that it is described according to the prediction result, from
The material recommended to the target user is determined in the material to be recommended, comprising:
Determine the value of the material characteristics cross term of the preset quantity material to be recommended；
According to the prediction result and described value, the predicted value of the preset quantity material to be recommended is calculated separately；
In the preset quantity material to be recommended, the maximum material to be recommended of the predicted value is determined as to the target
The material that user recommends.
8. recommendation material according to claim 1 determines method, which is characterized in that the material to be recommended include: video,
Information, music, advertisement.
9. recommendation material according to claim 1 determines method, which is characterized in that according to the prediction result, from described
In material to be recommended after the determining material recommended to the target user, further includes:
Recommend the material to the target user.
10. a kind of recommendation material determining device characterized by comprising
Feature vector obtains module, for obtaining the user characteristics vector of target user；
Nearest neighbor search module is searched for the user characteristics vector to be inputted the index structure constructed in advance based on arest neighbors
Rope determines material characteristics vector and the shortest preceding preset quantity of the user characteristics vector distance material to be recommended；
Prediction module, for by the input of the material characteristics vector of the user characteristics vector, material to be recommended it is trained in advance because
Sub- disassembler model, obtains the prediction result of Factorization machine model；
Material determining module, for according to the prediction result, determination to be pushed away to the target user from the material to be recommended
The material recommended.
11. a kind of computer readable storage medium, which is characterized in that be stored with computer on the computer readable storage medium
Program realizes claim 1 to 9 described in any item recommendation material determination sides when the computer program is executed by processor
Method.
12. a kind of computer equipment, characterized in that it comprises:
One or more processors；
Memory；
One or more computer programs, wherein one or more of computer programs are stored in the memory and quilt
It is configured to be executed by one or more of processors, one or more of computer programs are configured to: execute according to power
Benefit requires 1 to 9 described in any item recommendation materials to determine method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811482337.5A CN109408729B (en) | 2018-12-05 | 2018-12-05 | Recommended material determination method and device, storage medium and computer equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811482337.5A CN109408729B (en) | 2018-12-05 | 2018-12-05 | Recommended material determination method and device, storage medium and computer equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109408729A true CN109408729A (en) | 2019-03-01 |
| CN109408729B CN109408729B (en) | 2022-02-08 |
Family
ID=65457422
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811482337.5A Active CN109408729B (en) | 2018-12-05 | 2018-12-05 | Recommended material determination method and device, storage medium and computer equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109408729B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110162693A (en) * | 2019-03-04 | 2019-08-23 | 腾讯科技（深圳）有限公司 | A kind of method and server of information recommendation |
| CN110175857A (en) * | 2019-03-25 | 2019-08-27 | 阿里巴巴集团控股有限公司 | It is preferred that business determines method and device |
| CN110795631A (en) * | 2019-10-29 | 2020-02-14 | 支付宝(杭州)信息技术有限公司 | Push model optimization and prediction method and device based on factorization machine |
| CN111143670A (en) * | 2019-12-09 | 2020-05-12 | 中国平安财产保险股份有限公司 | Information determination method and related product |
| CN111274473A (en) * | 2020-01-13 | 2020-06-12 | 腾讯科技（深圳）有限公司 | Training method and device for recommendation model based on artificial intelligence and storage medium |
| CN111291264A (en) * | 2020-01-23 | 2020-06-16 | 腾讯科技（深圳）有限公司 | Access object prediction method and device based on machine learning and computer equipment |
| CN111597380A (en) * | 2020-05-14 | 2020-08-28 | 北京奇艺世纪科技有限公司 | Recommended video determining method and device, electronic equipment and storage medium |
| CN112417216A (en) * | 2019-08-23 | 2021-02-26 | 腾讯科技（深圳）有限公司 | Object recommendation method and device, server and storage medium |
| CN112506980A (en) * | 2020-12-22 | 2021-03-16 | 北京明略软件系统有限公司 | Streaming data processing frequency control method and system based on recommendation scene |
| CN112529636A (en) * | 2020-12-18 | 2021-03-19 | 平安科技（深圳）有限公司 | Commodity recommendation method and device, computer equipment and medium |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102982107A (en) * | 2012-11-08 | 2013-03-20 | 北京航空航天大学 | Recommendation system optimization method with information of user and item and context attribute integrated |
| US20170164049A1 (en) * | 2015-12-02 | 2017-06-08 | Le Holdings (Beijing) Co., Ltd. | Recommending method and device thereof |
| US20170169349A1 (en) * | 2015-12-09 | 2017-06-15 | Le Holdings (Beijing) Co., Ltd. | Recommending method and electronic device |
| CN107515909A (en) * | 2017-08-11 | 2017-12-26 | 深圳市耐飞科技有限公司 | A kind of video recommendation method and system |
| CN107635151A (en) * | 2017-09-25 | 2018-01-26 | 四川长虹电器股份有限公司 | A kind of machine learning TV programme suggesting method based on domain disassembler |
| CN107729488A (en) * | 2017-10-17 | 2018-02-23 | 北京搜狐新媒体信息技术有限公司 | A kind of information recommendation method and device |
| CN108345419A (en) * | 2017-01-25 | 2018-07-31 | 华为技术有限公司 | A kind of generation method and device of information recommendation list |
| US20180232794A1 (en) * | 2017-02-14 | 2018-08-16 | Idea Labs Inc. | Method for collaboratively filtering information to predict preference given to item by user of the item and computing device using the same |
| CN108647226A (en) * | 2018-03-26 | 2018-10-12 | 浙江大学 | A kind of mixing recommendation method based on variation autocoder |
| CN108769125A (en) * | 2018-04-28 | 2018-11-06 | 广州优视网络科技有限公司 | Using recommendation method, apparatus, storage medium and computer equipment |
| CN108804646A (en) * | 2018-06-06 | 2018-11-13 | 重庆邮电大学 | The point of interest of a kind of fusion deep learning and Factorization machine is registered prediction technique |
| CN108830680A (en) * | 2018-05-30 | 2018-11-16 | 山东大学 | Personalized recommendation method, system and storage medium based on discrete disassembler |
-
2018
- 2018-12-05 CN CN201811482337.5A patent/CN109408729B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102982107A (en) * | 2012-11-08 | 2013-03-20 | 北京航空航天大学 | Recommendation system optimization method with information of user and item and context attribute integrated |
| US20170164049A1 (en) * | 2015-12-02 | 2017-06-08 | Le Holdings (Beijing) Co., Ltd. | Recommending method and device thereof |
| US20170169349A1 (en) * | 2015-12-09 | 2017-06-15 | Le Holdings (Beijing) Co., Ltd. | Recommending method and electronic device |
| CN108345419A (en) * | 2017-01-25 | 2018-07-31 | 华为技术有限公司 | A kind of generation method and device of information recommendation list |
| US20180232794A1 (en) * | 2017-02-14 | 2018-08-16 | Idea Labs Inc. | Method for collaboratively filtering information to predict preference given to item by user of the item and computing device using the same |
| CN107515909A (en) * | 2017-08-11 | 2017-12-26 | 深圳市耐飞科技有限公司 | A kind of video recommendation method and system |
| CN107635151A (en) * | 2017-09-25 | 2018-01-26 | 四川长虹电器股份有限公司 | A kind of machine learning TV programme suggesting method based on domain disassembler |
| CN107729488A (en) * | 2017-10-17 | 2018-02-23 | 北京搜狐新媒体信息技术有限公司 | A kind of information recommendation method and device |
| CN108647226A (en) * | 2018-03-26 | 2018-10-12 | 浙江大学 | A kind of mixing recommendation method based on variation autocoder |
| CN108769125A (en) * | 2018-04-28 | 2018-11-06 | 广州优视网络科技有限公司 | Using recommendation method, apparatus, storage medium and computer equipment |
| CN108830680A (en) * | 2018-05-30 | 2018-11-16 | 山东大学 | Personalized recommendation method, system and storage medium based on discrete disassembler |
| CN108804646A (en) * | 2018-06-06 | 2018-11-13 | 重庆邮电大学 | The point of interest of a kind of fusion deep learning and Factorization machine is registered prediction technique |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110162693B (en) * | 2019-03-04 | 2024-05-10 | 深圳市雅阅科技有限公司 | Information recommendation method and server |
| CN110162693A (en) * | 2019-03-04 | 2019-08-23 | 腾讯科技（深圳）有限公司 | A kind of method and server of information recommendation |
| CN110175857A (en) * | 2019-03-25 | 2019-08-27 | 阿里巴巴集团控股有限公司 | It is preferred that business determines method and device |
| CN112417216A (en) * | 2019-08-23 | 2021-02-26 | 腾讯科技（深圳）有限公司 | Object recommendation method and device, server and storage medium |
| CN112417216B (en) * | 2019-08-23 | 2023-09-22 | 腾讯科技（深圳）有限公司 | Object recommendation method, device, server and storage medium |
| CN110795631A (en) * | 2019-10-29 | 2020-02-14 | 支付宝(杭州)信息技术有限公司 | Push model optimization and prediction method and device based on factorization machine |
| CN111143670A (en) * | 2019-12-09 | 2020-05-12 | 中国平安财产保险股份有限公司 | Information determination method and related product |
| CN111274473B (en) * | 2020-01-13 | 2021-03-12 | 腾讯科技（深圳）有限公司 | Training method and device for recommendation model based on artificial intelligence and storage medium |
| CN111274473A (en) * | 2020-01-13 | 2020-06-12 | 腾讯科技（深圳）有限公司 | Training method and device for recommendation model based on artificial intelligence and storage medium |
| CN111291264B (en) * | 2020-01-23 | 2023-06-23 | 腾讯科技（深圳）有限公司 | Access object prediction method and device based on machine learning and computer equipment |
| CN111291264A (en) * | 2020-01-23 | 2020-06-16 | 腾讯科技（深圳）有限公司 | Access object prediction method and device based on machine learning and computer equipment |
| CN111597380A (en) * | 2020-05-14 | 2020-08-28 | 北京奇艺世纪科技有限公司 | Recommended video determining method and device, electronic equipment and storage medium |
| CN111597380B (en) * | 2020-05-14 | 2023-06-02 | 北京奇艺世纪科技有限公司 | Recommended video determining method and device, electronic equipment and storage medium |
| CN112529636A (en) * | 2020-12-18 | 2021-03-19 | 平安科技（深圳）有限公司 | Commodity recommendation method and device, computer equipment and medium |
| CN112506980A (en) * | 2020-12-22 | 2021-03-16 | 北京明略软件系统有限公司 | Streaming data processing frequency control method and system based on recommendation scene |
| CN112506980B (en) * | 2020-12-22 | 2024-02-23 | 北京明略软件系统有限公司 | Streaming data processing frequency control method and system based on recommended scene |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109408729B (en) | 2022-02-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109408729A (en) | Material is recommended to determine method, apparatus, storage medium and computer equipment | |
| CN107613022B (en) | Content pushing method and device and computer equipment | |
| CN110149540B (en) | Recommendation processing method and device for multimedia resources, terminal and readable medium | |
| CN110543598B (en) | Information recommendation method and device and terminal | |
| US9798806B2 (en) | Information retrieval using dynamic guided navigation | |
| CN110162690A (en) | Determine user to the method and apparatus of the interest-degree of article, equipment and storage medium | |
| CN109284417A (en) | Video pushing method, device, computer equipment and storage medium | |
| CN105912669B (en) | Method and device for complementing search terms and establishing individual interest model | |
| US10909427B2 (en) | Method and device for classifying webpages | |
| CN104899246B (en) | Collaborative filtering recommending method based on blurring mechanism user scoring neighborhood information | |
| US20120084282A1 (en) | Content quality filtering without use of content | |
| CN109190043A (en) | Recommended method and device, storage medium, electronic equipment and recommender system | |
| CN110457581A (en) | A kind of information recommended method, device, electronic equipment and storage medium | |
| CN108829771A (en) | Main broadcaster's recommended method, device, computer storage medium and server | |
| CN112632359B (en) | Information recommendation method, device, electronic equipment and storage medium | |
| WO2014160648A1 (en) | Ranking product search results | |
| CN109167816A (en) | Information-pushing method, device, equipment and storage medium | |
| CN108132963A (en) | Resource recommendation method and device, computing device and storage medium | |
| CN111242310A (en) | Feature validity evaluation method and device, electronic equipment and storage medium | |
| CN108665148B (en) | Electronic resource quality evaluation method and device and storage medium | |
| CN105302880A (en) | Content correlation recommendation method and apparatus | |
| Yin et al. | Exploring social activeness and dynamic interest in community-based recommender system | |
| CN112579854A (en) | Information processing method, device, equipment and storage medium | |
| CN113961823B (en) | News recommendation method, system, storage medium and equipment | |
| CN111444438A (en) | Method, device, equipment and storage medium for determining recall permission rate of recall strategy |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20221123 Address after: 31a, 15th floor, building 30, maple commercial city, bangrang Road, Brazil Patentee after: Baiguoyuan Technology (Singapore) Co.,Ltd. Address before: Building B-1, North District, Wanda Commercial Plaza, Wanbo business district, No. 79, Wanbo 2nd Road, Nancun Town, Panyu District, Guangzhou City, Guangdong Province Patentee before: GUANGZHOU BAIGUOYUAN INFORMATION TECHNOLOGY Co.,Ltd. |