CN109903117A - A kind of knowledge mapping processing method and processing device for commercial product recommending - Google Patents
A kind of knowledge mapping processing method and processing device for commercial product recommending Download PDFInfo
- Publication number
- CN109903117A CN109903117A CN201910007982.XA CN201910007982A CN109903117A CN 109903117 A CN109903117 A CN 109903117A CN 201910007982 A CN201910007982 A CN 201910007982A CN 109903117 A CN109903117 A CN 109903117A
- Authority
- CN
- China
- Prior art keywords
- vector
- commodity
- user
- network model
- initial
- 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
- 238000013507 mapping Methods 0.000 title claims abstract description 28
- 238000012545 processing Methods 0.000 title claims abstract description 17
- 239000013065 commercial product Substances 0.000 title claims abstract description 13
- 238000003672 processing method Methods 0.000 title claims abstract description 7
- 239000013598 vector Substances 0.000 claims abstract description 207
- 210000005036 nerve Anatomy 0.000 claims abstract description 21
- 238000000034 method Methods 0.000 claims description 26
- 239000000047 product Substances 0.000 claims description 21
- 238000013528 artificial neural network Methods 0.000 claims description 16
- 238000012549 training Methods 0.000 claims description 16
- 238000000605 extraction Methods 0.000 claims description 15
- 238000003062 neural network model Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 6
- 230000006399 behavior Effects 0.000 claims description 6
- 230000000306 recurrent effect Effects 0.000 claims description 6
- 238000004519 manufacturing process Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 238000013139 quantization Methods 0.000 description 12
- 238000010801 machine learning Methods 0.000 description 9
- 239000000284 extract Substances 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 210000004218 nerve net Anatomy 0.000 description 2
- 230000001737 promoting effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009897 systematic effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
Abstract
The embodiment of the invention discloses a kind of knowledge mapping processing method and processing devices for commercial product recommending, are related to Internet technical field, can construct the knowledge mapping between commodity and user.Wherein, the initial commodity vector according to the initial user vector sum, the vector result for being trained study to first nerves network model, and obtaining the hidden layer in the first nerves network model calculated result is indicated as the vector of user and the vector of commodity indicates；The vector for calculating the user indicates the distance between the vector of commodity expression, using calculated distance as similarity；It is indicated using the initial user vector and the initial commodity vector indicates building nervus opticus network model, by the vector result of the hidden layer in the calculated result of the nervus opticus network model, the neighborhood vector as commodity is indicated；It is indicated according to obtained neighborhood vector, calculates the degree of correlation between each commodity；Construct the knowledge mapping between the user and commodity.
Description
Technical field
The present invention relates to Internet technical field more particularly to a kind of knowledge mapping processing method for commercial product recommending and
Device.
Background technique
Knowledge mapping is as relationship representation method between a kind of entity or concept and entity/concept, in search, natural language
There is more application in the fields such as speech processing, question and answer, retail or e-commerce field application still in its infancy.Current electricity
Sub- commercial field, knowledge mapping the relevant technologies focus primarily upon how to extract commodity/user tag, special by the label newly extracted
Sign directly applies to search or recommends field.
In electric business field, due to needing to handle the data of magnanimity and real-time change, mainly study in the industry at present
It is the label by how to train knowledge mapping, or how from the data sources such as picture, webpage extracts label.And label mentions
Processing scheme after taking then rarely has research, and the incidence relation between each label is usually manual maintenance.And it is limited to manpower
Cost and data volume, there are no systematic about commodity/custom system quantization means scheme, it is also difficult to which realization serves zero
Sell the Relation extraction of relationship.
So that knowledge mapping technology developing slowly in electric business retail domain, use scope is little.
Summary of the invention
The embodiment of the present invention provides a kind of knowledge mapping processing method and processing device for commercial product recommending, can construct quotient
Knowledge mapping between product and user.
In order to achieve the above objectives, the embodiment of the present invention adopts the following technical scheme that
One-hot is carried out using user's portrait information to encode to obtain initial user vector, is carried out using information attribute value
One-hot encodes to obtain initial commodity vector；
Initial commodity vector, is trained to first nerves network model according to the initial user vector sum
It practises, and obtains the vector result of the hidden layer in the first nerves network model calculated result, the vector as user indicates
It is indicated with the vector of commodity；
The vector for calculating the user indicates the distance between the vector of commodity expression, by calculated distance
As similarity；
It is indicated using the initial user vector and the initial commodity vector indicates, construct nervus opticus network model,
By the vector result of the hidden layer in the calculated result of the nervus opticus network model, the neighborhood vector as commodity is indicated；
It is indicated according to obtained neighborhood vector, calculates the degree of correlation between each commodity；
Using the vector of the user, the vector of the commodity, obtained similarity and the degree of correlation, the user is constructed
Knowledge mapping between commodity.
In the present embodiment, for current commodity/user knowledge quantization means problem, proposed in the present embodiment a kind of logical
Cross building neural network, the method that training indicates vector.With specific reference to the retail knowledge base that commodity/user tag is constituted, it is based on
Neural network method obtains commodity/user knowledge, is calculated by machine learning method or simple distance and extracts commodity/user
Knowledge is similar to correlativity, and the commodity and user knowledge map of a kind of retail domain are constructed using the relationship and label of extraction.From
And realize and quantization means are carried out to retail knowledge, similar correlativity is extracted based on quantization means, and then construct a kind of quotient
Product and user knowledge map.
By that using trained vector row, can be calculated with further progress or extract similar related pass using machine learning
System, so construct a kind of retail domain commodity user knowledge map.By this knowledge mapping, retail domain can be promoted
The efficiency and precision of many machine learning applications, such as solve the problems, such as the cold start-up of commercial product recommending technical field, promote advertisement and launch
Precision etc..
Detailed description of the invention
It to describe the technical solutions in the embodiments of the present invention more clearly, below will be to needed in the embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for ability
For the those of ordinary skill of domain, without creative efforts, it can also be obtained according to these attached drawings other attached
Figure.
Fig. 1 is system architecture schematic diagram provided in an embodiment of the present invention；
Fig. 2 a is method flow schematic diagram provided in an embodiment of the present invention；
Fig. 2 b is the example schematic of knowledge mapping construction logic provided in an embodiment of the present invention；
Fig. 3 is a kind of model schematic of trained commodity/user tag knowledge vector provided in an embodiment of the present invention；
Fig. 4 is a kind of trained commodity provided in an embodiment of the present invention/user's neighborhood vector field homoemorphism type schematic diagram；
Fig. 5 is apparatus structure schematic diagram provided in an embodiment of the present invention.
Specific embodiment
Technical solution in order to enable those skilled in the art to better understand the present invention, with reference to the accompanying drawing and specific embodiment party
Present invention is further described in detail for formula.Embodiments of the present invention are described in more detail below, the embodiment is shown
Example is shown in the accompanying drawings, and in which the same or similar labels are throughly indicated same or similar element or has identical or class
Like the element of function.It is exemplary below with reference to the embodiment of attached drawing description, for explaining only the invention, and cannot
It is construed to limitation of the present invention.Those skilled in the art of the present technique are appreciated that unless expressly stated, odd number shape used herein
Formula " one ", "one", " described " and "the" may also comprise plural form.It is to be further understood that specification of the invention
Used in wording " 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 the one or more of the other feature of addition, integer, step, operation, element, component and/or their group.It should be understood that
When we say that an element is " connected " or " coupled " to another element, it can be directly connected or coupled to other elements, or
There may also be intermediary elements.In addition, " connection " used herein or " coupling " may include being wirelessly connected or coupling.Here make
Wording "and/or" includes one or more associated any cells for listing item and all combinations.The art
Technical staff is appreciated that unless otherwise defined all terms (including technical terms and scientific terms) used herein have
Meaning identical with the general understanding of the those of ordinary skill in fields of the present invention.It should also be understood that such as general
Those terms, which should be understood that, defined in dictionary has a meaning that is consistent with the meaning in the context of the prior art, and
Unless defined as here, it will not be explained in an idealized or overly formal meaning.
Method flow in the present embodiment can specifically execute in a kind of system as shown in Figure 1, which includes:
Background server and database.
Background server obtains commodity portrait label and user's portrait information by access database.It is taken off in the present embodiment
The background server shown specifically can be the equipment such as server, work station, supercomputer, or by multiple server groups
At a kind of server cluster system for data processing.
The operation systems such as connection commodity library, commodity transaction platform and Retailer system, such as can be from commodity library
Middle acquisition commodity portrait label obtains user's portrait information inside commodity transaction platform or Retailer system.The present embodiment
Disclosed in database, specifically can be a kind of chart database, such as Neo4j, OrientDB, be also possible to relational data
Library, such as mysql, oracle, db2, specifically can be including store equipment data server and with data server phase
Storage equipment even, or a kind of server set for database being made of multiple data servers and storage server
Group's system.
The embodiment of the present invention provides a kind of knowledge mapping processing method for commercial product recommending, as shown in Figure 2 a, comprising:
S101, information of being drawn a portrait using user are carried out one-hot and encode to obtain initial user vector, and information attribute value is utilized
One-hot is carried out to encode to obtain initial commodity vector.
Wherein, user's portrait information includes at least: customer attribute information, user behavior information and commodity preference etc. are in the industry
The normally understood information drawn a portrait for describing user.The information attribute value includes at least: the functional parameter of commodity produces
Ground and suitable crowd etc. are normally understood for describing the information of item property in the industry.
One-hot coding is carried out to user's portrait information and information attribute value respectively, formation is expressed as (x1, x2... ...,
xn) etc structure high dimension vector, draw a portrait the obtained high dimension vector of information as initial user vector using user, utilize quotient
The high dimension vector that product attribute information obtains is as initial commodity vector.
S102, the initial commodity vector according to the initial user vector sum, instruct first nerves network model
Practice study, and obtains the vector result of the hidden layer in the first nerves network model calculated result, the vector as user
It indicates and the vector of commodity indicates.
In the present embodiment, it needs to utilize first nerves network model and nervus opticus network model, wherein first nerves
The effect of network model is, learns to the initial commodity vector of initial user vector sum, obtains label vector (specific manifestation
For the form of vectorization coding), label vector is primarily due to calculate the label vector distance between commodity/user, as similar
Relationship.
Specifically, the form of first nerves network model can multiplicity, a kind of model form as shown in Figure 3, model
The number of plies can increase, and form also can change.Here vector (the h after model training1, h2... ..., hn) commodity/user the most
Vector indicates that under normal circumstances, m is much smaller than n.In the present embodiment, first nerves network model is mainly used for vector dimensionality reduction, greatly
Partial neural network model can be applicable in, and existing neural network model work at present can be selected according to specific application scenarios
For first nerves network model.
Nervus opticus network model is constructed using initial commodity vector described in the initial user vector sum
, nervus opticus network model obtains field vector and (and shows as vectorization coding for learning to label vector
Form).About field vector field homoemorphism type training method, the form of neural network can be circulation nerve net, such as simple rnn,
Bi-rnn, lstm, gru etc..In the present embodiment, nervus opticus network model is used for similarity calculation, mainly uses RNN
(Recurrent Neural Networks, Recognition with Recurrent Neural Network) is used as nervus opticus network model.
Whether first nerves network model or nervus opticus network model, the initial use in every kind of neural network model
Input model is calculated the initial commodity vector needs of family vector sum together.
The vector of user indicates, is to obtain calculated result by first nerves network model with initial user vector, calculates
As a result the vector result of the hidden layer in is used as the vector of user to indicate.The vector of commodity indicates, is with initial commodity vector
Calculated result is obtained by first nerves network model, the vector result of the hidden layer in calculated result is the vector for being used as commodity
It indicates.
S103, the vector for calculating the user indicate the distance between the vector of commodity expression, will be calculated
Distance as similarity.
In the present embodiment, the vector similitude between user and commodity is calculated, and can use different distance definitions, such as Europe
Formula distance, included angle cosine, Chebyshev's distance etc., form can be different, as long as can quantify the similarity degree between object, formed
Similarity relation.
S104, initial commodity vector described in the initial user vector sum, building nervus opticus network model, by institute are utilized
The vector result for stating the hidden layer in the calculated result of nervus opticus network model, the neighborhood vector as commodity indicate.
S105, it is indicated according to obtained neighborhood vector, calculates the degree of correlation between each commodity.
S106, indicated using the vector of the user, the vectors of the commodity indicates, obtained similarity and correlation
Degree, constructs the knowledge mapping between the user and commodity.
In the present embodiment, as shown in Figure 2 b, using commodity/user as entity, commodity/user is embodied in label
Vector form, such as the initial user vector/initial commodity vector, alternatively, the vector of the user indicates the/commodity
Vector indicate.And using in acquired S103 acquired similarity and the degree of correlation acquired in S105 as similar related
Relationship constructs a kind of commodity/user knowledge map of retail domain.
Knowledge mapping is a branch of artificial intelligence, and the quantization means of research object are the bases of artificial intelligence.Zero
Field is sold, current technology is a tag extraction method in knowledge mapping field, and there are no systematic about commodity/use
The quantization means method of family system, and serve the Relation extraction method of retail relationship.And the scheme purpose of the present embodiment exists
In how expressing label, specifically how being associated with, forming topology, obtaining similitude and correlation, i.e., how training between label
Relationship.
In the present embodiment, for current commodity/user knowledge quantization means problem, proposed in the present embodiment a kind of logical
Cross building neural network, the method that training indicates vector.With specific reference to the retail knowledge base that commodity/user tag is constituted, it is based on
Neural network method obtains commodity/user knowledge, is calculated by machine learning method or simple distance and extracts commodity/user
Knowledge is similar to correlativity, and the commodity and user knowledge map of a kind of retail domain are constructed using the relationship and label of extraction.From
And realize and quantization means are carried out to retail knowledge, similar correlativity is extracted based on quantization means, and then construct a kind of quotient
Product and user knowledge map.
In the present embodiment, by the vector row using training, it can be calculated with further progress or be mentioned using machine learning
Take similar correlativity, so construct a kind of retail domain commodity user knowledge map.It, can be with by this knowledge mapping
The efficiency and precision for promoting many machine learning applications of retail domain, such as solve the problems, such as the cold start-up of commercial product recommending technical field,
Promote the precision etc. that advertisement is launched.
It encodes to obtain initial user vector specifically, carrying out one-hot using user's portrait information described in step S101,
Include:
User's portrait information, middle extraction user characteristics label are obtained, the user characteristics label corresponds to the user
Content in information of drawing a portrait.
User tag dictionary is constructed using the user characteristics label, and carries out one-hot coding, the height that coding is obtained
Initial vector is tieed up as the initial user vector.
It obtains or building commodity is drawn a portrait with user's relevant knowledge label or user, make characteristics dictionary, carry out one-hot volume
Code.Specifically the users such as user property, behavior, commodity preference can be obtained inside commodity transaction platform or Retailer system to draw
As information, commodity portrait label, such as functional parameter, suitable crowd, shelf-life are obtained from commodity library.And further directed to
Family, obtains or building user property, user behavior, building shot and long term Demand perference, user's portrait etc. classify to feature whole
It closes, is then encoded, form initial user vector.Then to labeling, duplicate removal, disambiguation, label dictionary is constructed, it is most laggard
Row one-hot coding, forms (x1, x2... ..., xn) higher-dimension initial vector.
It encodes to obtain initial commodity vector specifically, carrying out one-hot using information attribute value described in step S101,
Include:
Obtain the information attribute value, middle extraction product features label, the use of the product features label corresponding goods
Family feature and user group's positioning.
Commercial goods labels dictionary is constructed using the product features label, and carries out one-hot coding, the height that coding is obtained
Initial vector is tieed up as the initial commodity vector.
It obtains or building commodity is drawn a portrait with user's relevant knowledge label or user, make characteristics dictionary, carry out one-hot volume
Code.Specifically the users such as user property, behavior, commodity preference can be obtained inside commodity transaction platform or Retailer system to draw
As information, commodity portrait label, such as functional parameter, suitable crowd, shelf-life are obtained from commodity library.And further directed to quotient
Product obtain the attribute tags such as commodity function parameter, the place of production, shelf-life, extract product by the methods of cluster or keyword extraction
User characteristics, user group positioning etc. features, classification integration is carried out to feature, is then encoded, initial commodity vector is formed.So
Afterwards to labeling, duplicate removal, disambiguation, label dictionary is constructed, finally carries out one-hot coding, forms (x1, x2... ..., xn) high
Tie up initial vector.
Specifically, constructing nervus opticus network model in step S104, comprising:
By the initial commodity vector, or by the vectors of the commodity indicate to input the nervus opticus network model into
Row training, wherein the nervus opticus network model is Recognition with Recurrent Neural Network model.
The process of the forward direction iteration of the nervus opticus network model includes: St=f (U*Xt+W*St-1) wherein, XtIndicate t
Moment neighborhood commodity vector, otIndicate the output commodity vector of t moment, StIndicate the memory of t moment, f indicates neural network activation
Function.
In the present embodiment, about field vector field homoemorphism type training method, the form of neural network can be circulation nerve net,
Such as simple rnn, bi-rnn, lstm, gru.Model citing as shown in Figure 4, mode input vector X, can be just here
Beginning user vector and initial commodity vector；During model iteration, input vector X be also possible to user vector indicate and
The vector of commodity indicates.Centre is the network unit that may be reused.If using XtIndicate some neighborhood commodity of t moment
Vector, otIndicate the output commodity vector of t moment, StIndicate the memory of t moment, f indicates neural network activation primitive, then model
Forward direction iterative formula can indicate are as follows: St=f (U*Xt+W*St-1)
Activation primitive f therein can choose relu, softmax etc., output are as follows: ot=softmax (V*St)
By taking softmax as an example, is defined as:
Specifically, in step S104, the hidden layer by the calculated result of the nervus opticus network model to
Amount is as a result, the neighborhood vector as commodity indicates, comprising:
It is indicated using commodity/user vector, constructs neural network, using a certain commodity code finally bought as model
Tag variable carries out model training, obtains most to buy other commodity codes of the commodity simultaneously as the input variable of model
Hidden layer vector afterwards, the neighborhood vector as the tagged items indicate.Such as:
Using the coding of xth commodity as the tag variable of the nervus opticus network model.The training nervus opticus net
Network model obtains the vector result of the hidden layer in the nervus opticus network model calculated result, as the xth commodity
Neighborhood vector indicate, wherein other commodity be have purchased the xth commodity user simultaneously buy commodity, 1≤x≤
N, n are the positive integer greater than 1.
Specifically, in step S105, it is described to be indicated according to obtained neighborhood vector, calculate the correlation between each commodity
The concrete mode of degree may include: to be indicated according to the neighborhood vector of the xth commodity, calculate the commodity other than the xth commodity
With the degree of correlation between the xth commodity.
In the present embodiment, initial vector, which is converted to corresponding vector, to be indicated, is obtained by first nerves network model
Vector indicates, is actually the dimensionality reduction mark carried out for the label data of magnanimity, and be further processed using vector expression
Correlation and relevance are analyzed, realizes and shiploads of merchandise/user attribute and label is concentrated with lesser dimension, that is, realize
Dimension Reduction Analysis correlation, and can embody closeness relation between object.By this amount can directly computing object away from
From the similarity relation between object being obtained, further, it is possible to obtain the correlativity of object by model training.
The important relationship of correlation and similitude both relationships as retail domain, is the base of the applications such as recommendation, advertisement
Plinth.For example, the cold start-up problem of recommendation function may be implemented in terms of commercial product recommending, and conventional method is compared in accuracy
Also it is greatly improved.Knowledge mapping is constructed using these relationships, is served by the form of chart database, system may be implemented
Millisecond response, to greatly improve the operational efficiency of operation system.
In the present embodiment, a kind of knowledge mapping processing unit for commercial product recommending is also provided, as shown in Figure 5, comprising:
Coding module encodes to obtain initial user vector, utilizes commodity for carrying out one-hot using user's portrait information
Attribute information carries out one-hot and encodes to obtain initial commodity vector.
First processing module is used for the initial commodity vector according to the initial user vector sum, to first nerves net
Network model is trained study, and obtains the vector result of the hidden layer in the first nerves network model calculated result, makees
Vector for user indicates and the vector of commodity indicates.
Similarity calculation module, for calculate the user vector indicate and the commodity vector expression between away from
From using calculated distance as similarity.
Second processing module, for indicating to indicate with the initial commodity vector using the initial user vector, building
Nervus opticus network model, by the vector result of the hidden layer in the calculated result of the nervus opticus network model, as quotient
The neighborhood vector of product indicates.
Relatedness computation module calculates the degree of correlation between each commodity for indicating according to obtained neighborhood vector.
Map construction module, for utilizing the vector of the user, the vector of the commodity, obtained similarity and phase
Guan Du constructs the knowledge mapping between the user and commodity.
Wherein, the coding module is specifically used for obtaining the user and draws a portrait information, middle extraction user characteristics label,
In, user's portrait information includes at least: customer attribute information, user behavior information and commodity preference, the user are special
Sign label corresponds to the content in user's portrait information.User tag dictionary is constructed using the user characteristics label, is gone forward side by side
Row one-hot coding, the higher-dimension initial vector that coding is obtained is as the initial user vector.
The coding module, is specifically also used to: obtaining the information attribute value, middle extraction product features label, wherein
The information attribute value includes at least: functional parameter, the place of production and the suitable crowd of commodity, and the product features label is corresponding
The user characteristics of commodity and user group's positioning.Commercial goods labels dictionary is constructed using the product features label, and carries out one-
Hot coding, the higher-dimension initial vector that coding is obtained is as the initial commodity vector.
The Second processing module is specifically used for the initial commodity vector, or the vector of the commodity is indicated
It inputs the nervus opticus network model to be trained, wherein the nervus opticus network model is Recognition with Recurrent Neural Network model.
The process of the forward direction iteration of the nervus opticus network model includes: St=f (U*Xt+W*St-1), wherein XtIndicate that t moment is adjacent
Domain commodity vector, otIndicate the output commodity vector of t moment, StIndicate the memory of t moment, f indicates neural network activation primitive.
The Second processing module, specifically for the mark using the coding of xth commodity as the nervus opticus network model
Sign variable.The training nervus opticus network model, obtains the hidden layer in the nervus opticus network model calculated result
Vector result, the neighborhood vector as the xth commodity indicate, wherein other commodity are to have purchased the xth commodity
The commodity that user buys simultaneously, 1≤x≤n, n are the positive integer greater than 1.
The relatedness computation module calculates the xth specifically for indicating according to the neighborhood vector of the xth commodity
The degree of correlation between commodity and the xth commodity other than commodity.
In the present embodiment, for current commodity/user knowledge quantization means problem, proposed in the present embodiment a kind of logical
Cross building neural network, the method that training indicates vector.With specific reference to the retail knowledge base that commodity/user tag is constituted, it is based on
Neural network method obtains commodity/user knowledge, is calculated by machine learning method or simple distance and extracts commodity/user
Knowledge is similar to correlativity, and the commodity and user knowledge map of a kind of retail domain are constructed using the relationship and label of extraction.From
And realize and quantization means are carried out to retail knowledge, similar correlativity is extracted based on quantization means, and then construct a kind of quotient
Product and user knowledge map.
In the present embodiment, by the vector row using training, it can be calculated with further progress or be mentioned using machine learning
Take similar correlativity, so construct a kind of retail domain commodity user knowledge map.It, can be with by this knowledge mapping
The efficiency and precision for promoting many machine learning applications of retail domain, such as solve the problems, such as the cold start-up of commercial product recommending technical field,
Promote the precision etc. that advertisement is launched.
All the embodiments in this specification are described in a progressive manner, same and similar portion between each embodiment
Dividing may refer to each other, and the highlights of each of the examples are differences from other embodiments.Especially for equipment reality
For applying example, since it is substantially similar to the method embodiment, so describing fairly simple, related place is referring to embodiment of the method
Part explanation.The above description is merely a specific embodiment, but protection scope of the present invention is not limited to
This, anyone skilled in the art in the technical scope disclosed by the present invention, the variation that can readily occur in or replaces
It changes, should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with the protection model of claim
Subject to enclosing.
Claims (10)
1. a kind of knowledge mapping processing method for commercial product recommending characterized by comprising
One-hot is carried out using user's portrait information to encode to obtain initial user vector, and information attribute value is utilized to carry out one-
Hot encodes to obtain initial commodity vector；
Initial commodity vector, is trained study to first nerves network model according to the initial user vector sum, and
The vector result of the hidden layer in the first nerves network model calculated result is obtained, the vector as user indicates and commodity
Vector indicate；
The vector for calculating the user the distance between indicates to indicate with the vectors of the commodity, using calculated distance as
Similarity；
It is indicated using the initial user vector and the initial commodity vector indicates, nervus opticus network model is constructed, by institute
The vector result for stating the hidden layer in the calculated result of nervus opticus network model, the neighborhood vector as commodity indicate；
It is indicated according to obtained neighborhood vector, calculates the degree of correlation between each commodity；
Using the vector of the user, the vector of the commodity, obtained similarity and the degree of correlation, the user and quotient are constructed
Knowledge mapping between product.
2. the method according to claim 1, wherein described carry out one-hot coding using user's portrait information
Obtain initial user vector, comprising:
Obtain user's portrait information, middle extraction user characteristics label, wherein user's portrait information includes at least:
Customer attribute information, user behavior information and commodity preference, the user characteristics label correspond in user's portrait information
Content；
User tag dictionary is constructed using the user characteristics label, and carries out one-hot coding, at the beginning of the higher-dimension that coding is obtained
Beginning vector is as the initial user vector.
3. the method according to claim 1, wherein described carry out one-hot coding using information attribute value
Obtain initial commodity vector, comprising:
The information attribute value is obtained, middle extraction product features label, wherein the information attribute value includes at least:
Functional parameter, the place of production and the suitable crowd of commodity, the user characteristics of the product features label corresponding goods and user group are fixed
Position；
Commercial goods labels dictionary is constructed using the product features label, and carries out one-hot coding, at the beginning of the higher-dimension that coding is obtained
Beginning vector is as the initial commodity vector.
4. the method according to claim 1, wherein the building nervus opticus network model, comprising:
Indicate that inputting the nervus opticus network model instructs by the initial commodity vector, or by the vector of the commodity
Practice, wherein the nervus opticus network model is Recognition with Recurrent Neural Network model；
The process of the forward direction iteration of the nervus opticus network model includes:
St=f (U*Xt+W*St-1), wherein XtIndicate t moment neighborhood commodity vector, otIndicate the output commodity vector of t moment, St
Indicate the memory of t moment, f indicates neural network activation primitive.
5. according to the method described in claim 4, it is characterized in that, the calculated result by the nervus opticus network model
In hidden layer vector result, as commodity neighborhood vector indicate, comprising:
Using the coding of xth commodity as the tag variable of the nervus opticus network model；
The training nervus opticus network model, obtains the vector of the hidden layer in the nervus opticus network model calculated result
As a result, the neighborhood vector as the xth commodity indicates, wherein other commodity are the user for having purchased the xth commodity
The commodity of purchase simultaneously, 1≤x≤n, n are the positive integer greater than 1.
6. according to the method described in claim 5, calculating is each it is characterized in that, described indicate according to obtained neighborhood vector
The degree of correlation between a commodity are as follows: indicated according to the neighborhood vector of the xth commodity, calculate the commodity other than the xth commodity
With the degree of correlation between the xth commodity.
7. a kind of knowledge mapping processing unit for commercial product recommending characterized by comprising
Coding module encodes to obtain initial user vector, utilizes item property for carrying out one-hot using user's portrait information
Information carries out one-hot and encodes to obtain initial commodity vector；
First processing module is used for the initial commodity vector according to the initial user vector sum, to first nerves network mould
Type is trained study, and obtains the vector result of the hidden layer in the first nerves network model calculated result, as with
The vector at family indicates and the vector of commodity indicates；
Similarity calculation module, the vector for calculating the user indicate the distance between the vector of commodity expression,
Using calculated distance as similarity；
Second processing module, for indicating to indicate with the initial commodity vector using the initial user vector, building second
Neural network model, by the vector result of the hidden layer in the calculated result of the nervus opticus network model, as commodity
Neighborhood vector indicates；
Relatedness computation module calculates the degree of correlation between each commodity for indicating according to obtained neighborhood vector；
Map construction module, for utilizing the vector of the user, the vector of the commodity, obtained similarity and correlation
Degree, constructs the knowledge mapping between the user and commodity.
8. device according to claim 7, which is characterized in that the coding module is specifically used for obtaining user's picture
As information, middle extraction user characteristics label, wherein user's portrait information includes at least: customer attribute information, Yong Huhang
For information and commodity preference, the user characteristics label corresponds to the content in user's portrait information；It is special using the user
It levies label and constructs user tag dictionary, and carry out one-hot coding, the higher-dimension initial vector that coding is obtained is as described initial
User vector；
The coding module, is specifically also used to: obtaining the information attribute value, middle extraction product features label, wherein described
Information attribute value includes at least: functional parameter, the place of production and the suitable crowd of commodity, the product features label corresponding goods
User characteristics and user group positioning；Commercial goods labels dictionary is constructed using the product features label, and carries out one-hot volume
Code, the higher-dimension initial vector that coding is obtained is as the initial commodity vector.
9. device according to claim 7, which is characterized in that the Second processing module, being specifically used for will be described initial
Commodity vector, or the vectors of the commodity is indicated that inputting the nervus opticus network model is trained, wherein described the
Two neural network models are Recognition with Recurrent Neural Network model；The process of the forward direction iteration of the nervus opticus network model includes: St
=f (U*Xt+W*St-1), wherein XtIndicate t moment neighborhood commodity vector, otIndicate the output commodity vector of t moment, StIndicate t
The memory at moment, f indicate neural network activation primitive.
10. device according to claim 9, which is characterized in that the Second processing module is specifically used for xth commodity
Tag variable of the coding as the nervus opticus network model；The training nervus opticus network model obtains described the
The vector result of hidden layer in two neural network model calculated results, as the xth commodity neighborhood vector indicate,
In, other commodity are the user for having purchased the xth commodity while the commodity of purchase, and 1≤x≤n, n are just whole greater than 1
Number；
The relatedness computation module calculates the xth commodity specifically for indicating according to the neighborhood vector of the xth commodity
The degree of correlation between commodity and the xth commodity in addition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910007982.XA CN109903117B (en) | 2019-01-04 | 2019-01-04 | Knowledge graph processing method and device for commodity recommendation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910007982.XA CN109903117B (en) | 2019-01-04 | 2019-01-04 | Knowledge graph processing method and device for commodity recommendation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109903117A true CN109903117A (en) | 2019-06-18 |
| CN109903117B CN109903117B (en) | 2022-10-25 |
Family
ID=66943538
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910007982.XA Active CN109903117B (en) | 2019-01-04 | 2019-01-04 | Knowledge graph processing method and device for commodity recommendation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109903117B (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110275964A (en) * | 2019-06-26 | 2019-09-24 | 程淑玉 | The recommended models of knowledge based map and Recognition with Recurrent Neural Network |
| CN110598766A (en) * | 2019-08-28 | 2019-12-20 | 第四范式（北京）技术有限公司 | Training method and device for commodity recommendation model and electronic equipment |
| CN110880089A (en) * | 2019-11-21 | 2020-03-13 | 秒针信息技术有限公司 | Knowledge graph-based warehouse space distribution method and system |
| CN111199459A (en) * | 2019-12-30 | 2020-05-26 | 深圳市盟天科技有限公司 | Commodity recommendation method and device, electronic equipment and storage medium |
| CN111310074A (en) * | 2020-02-13 | 2020-06-19 | 北京百度网讯科技有限公司 | Interest point label optimization method and device, electronic equipment and computer readable medium |
| CN111310051A (en) * | 2019-12-13 | 2020-06-19 | 北京邮电大学 | Cold start solution based on knowledge graph |
| CN111507796A (en) * | 2020-04-07 | 2020-08-07 | 桂林电子科技大学 | Online shopping mall commodity recommendation method based on knowledge graph |
| CN112102029A (en) * | 2020-08-20 | 2020-12-18 | 浙江大学 | Knowledge graph-based long-tail recommendation calculation method |
| CN112200636A (en) * | 2020-10-24 | 2021-01-08 | 朱丽勤 | Intelligent shopping recommendation method based on big data |
| CN112395429A (en) * | 2020-12-02 | 2021-02-23 | 上海三稻智能科技有限公司 | Method, system and storage medium for determining, pushing and applying HS (high speed coding) codes based on graph neural network |
| CN112446741A (en) * | 2020-12-10 | 2021-03-05 | 华院数据技术(上海)有限公司 | User portrayal method and system based on probability knowledge graph |
| CN112749323A (en) * | 2019-10-31 | 2021-05-04 | 北京沃东天骏信息技术有限公司 | Method and device for constructing user portrait |
| CN112837118A (en) * | 2021-01-21 | 2021-05-25 | 北京电解智科技有限公司 | Commodity recommendation method and device for enterprise users |
| WO2021171126A1 (en) * | 2020-02-28 | 2021-09-02 | International Business Machines Corporation | Personalized automated machine learning |
| WO2021223165A1 (en) * | 2020-05-07 | 2021-11-11 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for object evaluation |
| WO2022135118A1 (en) * | 2020-12-23 | 2022-06-30 | 浙江大学 | Combined product mining method based on knowledge graph rule embedding |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160189028A1 (en) * | 2014-12-31 | 2016-06-30 | Verizon Patent And Licensing Inc. | Systems and Methods of Using a Knowledge Graph to Provide a Media Content Recommendation |
| CN107767262A (en) * | 2017-10-24 | 2018-03-06 | 上海点融信息科技有限责任公司 | Information processing method, device and computer-readable recording medium |
| CN108804419A (en) * | 2018-05-22 | 2018-11-13 | 湖南大学 | Medicine is sold accurate recommended technology under a kind of line of knowledge based collection of illustrative plates |
| CN108960975A (en) * | 2018-06-15 | 2018-12-07 | 广州麦优网络科技有限公司 | Personalized Precision Marketing Method, server and storage medium based on user's portrait |
-
2019
- 2019-01-04 CN CN201910007982.XA patent/CN109903117B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160189028A1 (en) * | 2014-12-31 | 2016-06-30 | Verizon Patent And Licensing Inc. | Systems and Methods of Using a Knowledge Graph to Provide a Media Content Recommendation |
| CN107767262A (en) * | 2017-10-24 | 2018-03-06 | 上海点融信息科技有限责任公司 | Information processing method, device and computer-readable recording medium |
| CN108804419A (en) * | 2018-05-22 | 2018-11-13 | 湖南大学 | Medicine is sold accurate recommended technology under a kind of line of knowledge based collection of illustrative plates |
| CN108960975A (en) * | 2018-06-15 | 2018-12-07 | 广州麦优网络科技有限公司 | Personalized Precision Marketing Method, server and storage medium based on user's portrait |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110275964B (en) * | 2019-06-26 | 2023-07-14 | 程淑玉 | Recommendation model based on knowledge graph and cyclic neural network |
| CN110275964A (en) * | 2019-06-26 | 2019-09-24 | 程淑玉 | The recommended models of knowledge based map and Recognition with Recurrent Neural Network |
| CN110598766B (en) * | 2019-08-28 | 2022-05-10 | 第四范式（北京）技术有限公司 | Training method and device for commodity recommendation model and electronic equipment |
| CN110598766A (en) * | 2019-08-28 | 2019-12-20 | 第四范式（北京）技术有限公司 | Training method and device for commodity recommendation model and electronic equipment |
| CN112749323A (en) * | 2019-10-31 | 2021-05-04 | 北京沃东天骏信息技术有限公司 | Method and device for constructing user portrait |
| CN110880089A (en) * | 2019-11-21 | 2020-03-13 | 秒针信息技术有限公司 | Knowledge graph-based warehouse space distribution method and system |
| CN111310051A (en) * | 2019-12-13 | 2020-06-19 | 北京邮电大学 | Cold start solution based on knowledge graph |
| CN111199459A (en) * | 2019-12-30 | 2020-05-26 | 深圳市盟天科技有限公司 | Commodity recommendation method and device, electronic equipment and storage medium |
| CN111199459B (en) * | 2019-12-30 | 2024-02-09 | 深圳市盟天科技有限公司 | Commodity recommendation method, commodity recommendation device, electronic equipment and storage medium |
| CN111310074A (en) * | 2020-02-13 | 2020-06-19 | 北京百度网讯科技有限公司 | Interest point label optimization method and device, electronic equipment and computer readable medium |
| CN111310074B (en) * | 2020-02-13 | 2023-08-18 | 北京百度网讯科技有限公司 | Method and device for optimizing labels of interest points, electronic equipment and computer readable medium |
| GB2608540A (en) * | 2020-02-28 | 2023-01-04 | Ibm | Personalized automated machine learning |
| US11379710B2 (en) * | 2020-02-28 | 2022-07-05 | International Business Machines Corporation | Personalized automated machine learning |
| WO2021171126A1 (en) * | 2020-02-28 | 2021-09-02 | International Business Machines Corporation | Personalized automated machine learning |
| CN111507796A (en) * | 2020-04-07 | 2020-08-07 | 桂林电子科技大学 | Online shopping mall commodity recommendation method based on knowledge graph |
| WO2021223165A1 (en) * | 2020-05-07 | 2021-11-11 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for object evaluation |
| CN112102029A (en) * | 2020-08-20 | 2020-12-18 | 浙江大学 | Knowledge graph-based long-tail recommendation calculation method |
| CN112200636A (en) * | 2020-10-24 | 2021-01-08 | 朱丽勤 | Intelligent shopping recommendation method based on big data |
| CN112395429A (en) * | 2020-12-02 | 2021-02-23 | 上海三稻智能科技有限公司 | Method, system and storage medium for determining, pushing and applying HS (high speed coding) codes based on graph neural network |
| CN112446741B (en) * | 2020-12-10 | 2022-02-15 | 华院计算技术(上海)股份有限公司 | User portrayal method and system based on probability knowledge graph |
| CN112446741A (en) * | 2020-12-10 | 2021-03-05 | 华院数据技术(上海)有限公司 | User portrayal method and system based on probability knowledge graph |
| WO2022135118A1 (en) * | 2020-12-23 | 2022-06-30 | 浙江大学 | Combined product mining method based on knowledge graph rule embedding |
| CN112837118A (en) * | 2021-01-21 | 2021-05-25 | 北京电解智科技有限公司 | Commodity recommendation method and device for enterprise users |
| WO2022156529A1 (en) * | 2021-01-21 | 2022-07-28 | 北京电解智科技有限公司 | Commodity recommendation method and apparatus for enterprise user |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109903117B (en) | 2022-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109903117A (en) | A kind of knowledge mapping processing method and processing device for commercial product recommending | |
| CN103593792B (en) | A kind of personalized recommendation method based on Chinese knowledge mapping and system | |
| CN111784455A (en) | Article recommendation method and recommendation equipment | |
| Rupapara et al. | Auto-encoders for content-based image retrieval with its implementation using handwritten dataset | |
| Gao et al. | Small sample classification of hyperspectral image using model-agnostic meta-learning algorithm and convolutional neural network | |
| CN112364976A (en) | User preference prediction method based on session recommendation system | |
| WO2023011382A1 (en) | Recommendation method, recommendation model training method, and related product | |
| Zhang et al. | Depth-based subgraph convolutional auto-encoder for network representation learning | |
| CN112380453B (en) | Article recommendation method and device, storage medium and equipment | |
| Guo et al. | Attention based consistent semantic learning for micro-video scene recognition | |
| CN115982463A (en) | Resource recommendation method, device, equipment and storage medium | |
| CN115687760A (en) | User learning interest label prediction method based on graph neural network | |
| CN116521906A (en) | Meta description generation method, device, equipment and medium thereof | |
| CN113298546A (en) | Sales prediction method and device, and commodity processing method and device | |
| Zhang | E-commerce products recognition based on a deep learning architecture: Theory and implementation | |
| CN112199884A (en) | Article molecule generation method, device, equipment and storage medium | |
| CN116821516A (en) | Resource recommendation method, device, equipment and storage medium | |
| CN116975686A (en) | Method for training student model, behavior prediction method and device | |
| CN114417161B (en) | Virtual article time sequence recommendation method, device, medium and equipment based on special-purpose map | |
| CN114168804B (en) | Similar information retrieval method and system based on heterogeneous subgraph neural network | |
| Wang et al. | One-shot retail product identification based on improved Siamese neural networks | |
| CN112307343B (en) | Cross-E-book city user alignment method based on double-layer iterative compensation and full-face representation | |
| Qiang et al. | Large-scale multi-label image retrieval using residual network with hash layer | |
| Song et al. | Human action recognition with 3D convolution skip-connections and RNNs | |
| Hu et al. | Unsupervised segmentation for terracotta warrior with seed-region-growing CNN (SRG-Net) |
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 |