CN108897823A - Personalized commercial search method and device based on deep learning attention mechanism - Google Patents
Personalized commercial search method and device based on deep learning attention mechanism Download PDFInfo
- Publication number
- CN108897823A CN108897823A CN201810645528.2A CN201810645528A CN108897823A CN 108897823 A CN108897823 A CN 108897823A CN 201810645528 A CN201810645528 A CN 201810645528A CN 108897823 A CN108897823 A CN 108897823A
- Authority
- CN
- China
- Prior art keywords
- attention mechanism
- term preference
- long
- commodity
- inquiry
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of personalized commercial search method and device based on deep learning attention mechanism.Wherein, this method includes：Construct the short-term preference pattern based on attention mechanism；Construct the long-term preference pattern based on attention mechanism；Inquiry indicates again；Inquiring the process that indicates again is：Merge the short-term preference pattern based on attention mechanism, long-term preference pattern and current queries based on attention mechanism, learn the interactive relation between three by multilayer fully-connected network, the inquiry recombinated indicates, and the degree of correlation of all commodity and current queries is defined using a distance function；After the training multilayer fully-connected network, to the new inquiry that each user submits, the distance value of all commodity and current queries is obtained, all distance values are ranked up from high to low again, and the corresponding commodity of preceding n distance value are returned into user, wherein n is positive integer.
Description
Technical field
The invention belongs to field of data retrieval more particularly to a kind of personalized commercials based on deep learning attention mechanism
Search method and device.
Background technique
Prevailing with internet, e-commerce also becomes more and more popular.When e-commerce website is (such as day
Cat) in user want purchase one commodity when, it usually needs find him from millions of commodity in a manner of retrieval
Admire that.For this online commodity retrieval, a relatively common situation is that user submits inquiry first,
Then search engine returns to the items list to have sorted relevant to current queries.However, the inquiry that user submits is in general
Only by several crucial phrases at (for example, man's long sleeve blouse), this, which also has led to it, can not accurately convey the demand of user, from
And it causes user and search result is discontented with.
In addition, the shopping preferences of user can be, very wide in range (due to different backgrounds, such as age, gender is received
Enter), or (such as season, positioning) is influenced by current environment.Therefore, the same inquiry from different user is come
It says, different degrees of economic loss can be caused to e-commerce website by returning to identical search result.In view of this, considering user
Shopping under different situations is intended to return to the inquiry that user submits relevant commodity, i.e., personalized commodity retrieval is right
Meeting the current shopping need of user is just particularly important.
Traditional commodity search method be limited only to inquiry commodity between simple match without by user itself
Attribute considers wherein.These methods are drawn due to ignoring the heterogeneity of numerous individual subscriber demands so often will lead to search
It is limited to hold up retrieval performance, satisfied search result can not be provided for user.
Ai et al. is recently proposed a kind of commodity search method of personalization, they can learn to use jointly by one
The latent space that family, commodity and inquiry indicate models the long-term shopping preferences of user.But this method also there are two types of
Defect：(1) assume that the long-term shopping preferences of user are stable, and be actually to change at any time and slowly；(2) do not have
The short-term shopping preferences of user are taken into account, and the short-term buying behavior of user is likely to reflection user's nearest a period of time
Buying habit.
Wherein, refer to for a long time in user and metastable shopping preferences, such as the color liked, suitable size
With consuming capacity etc..It will receive the respective background influence of user, such as age, marriage, education, income etc. simultaneously.In contrast, short
The shopping preferences of phase reflect shopping of the user within a relatively short period and are intended to, and will receive the influence of emergency event,
Such as new product release, season changes and special personal red-letter day (such as birthday).These quotient that can be bought recently from user
Be inferred in product attribute, compared with long-term shopping preferences, short-term shopping preferences update ground more frequently with it is difficult to predict.
Currently, being had the following problems in personalized commercial search method：
First is that being accurately extremely complex for the long-term and short-term shopping preferences modeling of user.The long-term shopping of user is inclined
Well include many aspects, such as consuming capacity or favorite color and brand, and can change with the background (as taken in) of user
Become and changes.It corresponds, the short-term shopping preferences of user are generally also what dynamic changed, and are highly prone to emergency event
Influence；
Second is that user by one only by several crucial phrases at text query oneself shopping need, this meeting described
It is remarkable for causing to be accurately positioned in the long-term shopping preferences of user with current queries related aspect.For example, a cotta
The design of jacket, rather than price, can be more to the customer impact for having economic capability.For the short-term preference of user, most
The multiple commodity closely bought also can generate different influences to the next buying behavior of user；
It is also difficult third is that long-term, the short-term shopping preferences of user are combined with current inquiry.
Summary of the invention
In order to solve the deficiencies in the prior art, the first object of the present invention is to provide a kind of based on deep learning attention machine
The personalized commercial search method of system, which raises the accuracy of retrieval, to improve the retrieval experience of user.
A kind of personalized commercial search method based on deep learning attention mechanism of the invention, including：
Step 1：Construct the short-term preference pattern based on attention mechanism；The step 1 specifically includes：
Current inquiry is given, the degree of correlation before being estimated using neural network between m inquiry and current queries, and
It is indicated using different attention weights；Wherein, m is positive integer；
The attention weight of acquisition is indicated multiplied by corresponding commodity, is input in RNN model, output is indicated based on attention
M vector of the short-term preference of power mechanism；
Step 2：Construct the long-term preference pattern based on attention mechanism；The step 2 specifically includes：
It is indicated using m commodity of initial purchase to initialize its long-term preference, and is based on attention machine using with building
The same principle of the short-term preference pattern of system, obtains the long-term preference pattern based on attention mechanism；
It is every later to pass through m commodity, the long-term preference of an active user is just updated, is obtained updated based on attention
The long-term preference pattern of mechanism；
Step 3：Inquiry indicates again；Its process is：It merges the short-term preference pattern based on attention mechanism, be based on attention
The long-term preference pattern and current queries of mechanism, are learnt the interactive relation between three by multilayer fully-connected network, obtained
Inquiry to recombination indicates, and the degree of correlation of all commodity and current queries is defined using a distance function；
Step 4：After the training multilayer fully-connected network, to the new inquiry that each user submits, all quotient are obtained
The distance value of product and current queries, then all distance values are ranked up from high to low, and by the corresponding commodity of preceding n distance value
Return to user, wherein n is positive integer.
Further, this method further includes：
The multilayer fully-connected network is trained using the learning method of " based in pairs ".
Further, it in the step 1, is inquired using two layers neural network come m before estimating and is currently looked into
Degree of correlation between inquiry.
Further, in the step 3, the distance function is COS distance or dot product or Euclidean distance or graceful
Hatton's distance.
The second object of the present invention is to provide a kind of personalized commercial retrieval device based on deep learning attention mechanism.
A kind of personalized commercial based on deep learning attention mechanism of the invention retrieves device, including personalized commercial
Retrieval process device, the personalized commercial retrieval process device include：
Short-term preference pattern constructs module, is configured as：Construct the short-term preference pattern based on attention mechanism；It is described
Short-term preference pattern building module specifically includes：
Attention Weight Acquisition submodule, is configured as：Current inquiry is given, preceding m is estimated using neural network
Degree of correlation between a inquiry and current queries, and indicated using different attention weights；Wherein, m is positive integer；And
Short-term preference indicates submodule, is configured as：The attention weight of acquisition is indicated multiplied by corresponding commodity, it is defeated
Enter into RNN model, output indicates m vector of the short-term preference based on attention mechanism；
Long-term preference pattern constructs module, is configured as：Construct the long-term preference pattern based on attention mechanism；It is described
Long-term preference pattern building module specifically includes：
Long-term preference initialization and building submodule, are configured as：It is indicated using m commodity of initial purchase come initial
Change its long-term preference, and using the same principle with short-term preference pattern of the building based on attention mechanism, obtains based on attention
The long-term preference pattern of power mechanism；
Long-term preference updates submodule, is configured as：It is every later to pass through m commodity, just update an active user's
Long-term preference obtains the updated long-term preference pattern based on attention mechanism；
Representation module again is inquired, is configured to：It merges the short-term preference pattern based on attention mechanism, be based on attention machine
The long-term preference pattern and current queries of system, are learnt the interactive relation between three by multilayer fully-connected network, obtained
The inquiry of recombination indicates, and the degree of correlation of all commodity and current queries is defined using a distance function；
Commodity return module, is configured to：After the training multilayer fully-connected network, each user is submitted new
Inquiry, obtains the distance value of all commodity and current queries, then all distance values are ranked up from high to low, and by preceding n
The corresponding commodity of distance value return to user, wherein n is positive integer.
Further, the personalized commercial retrieval process device further includes：
Model training module is configured as：The multilayer is trained to connect entirely using the learning method of " based in pairs "
Network.
Further, in the attention Weight Acquisition submodule, preceding m is estimated using two layers neural network
Degree of correlation between a inquiry and current queries.
Further, in the inquiry again representation module, the distance function is COS distance or dot product or European
Distance or manhatton distance.
Compared with prior art, the beneficial effects of the invention are as follows：
(1) in order to alleviate the problems faced in existing method, the present invention can be in conjunction with the metastable of user
Long-term shopping preferences and personalized retrieval is carried out to commodity to the short-term shopping preferences of time-sensitive, finally improves the standard of retrieval
Exactness, to improve the retrieval experience of user.
(2) inquiry that the present invention can effectively be submitted the long-term and short-term shopping preferences of user and active user
In conjunction with, and the shopping need of user is indicated again.
(3) two respective attention mechanism in the long-term and short-term shopping preferences modeling that the present invention passes through user, energy
Factor relevant to current queries in enough prominent two preferences.
(4) present invention improves the accuracy of personalized commercial retrieval, to a certain extent be e-commerce website
Retain more users and improves income.
Detailed description of the invention
The accompanying drawings constituting a part of this application is used to provide further understanding of the present application, and the application's shows
Meaning property embodiment and its explanation are not constituted an undue limitation on the present application for explaining the application.
Fig. 1 is a kind of personalized commercial search method flow chart based on deep learning attention mechanism of the invention.
Fig. 2 is personalized commercial retrieval process device structural schematic diagram of the invention.
Specific embodiment
It is noted that following detailed description is all illustrative, it is intended to provide further instruction to the application.Unless another
It indicates, all technical and scientific terms used herein has usual with the application person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, additionally, it should be understood that, when in the present specification using term "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, device, component and/or their combination.
Below with reference to examples of implementation and attached drawing to a kind of personalized quotient based on deep learning attention mechanism of the invention
Product search method is described in detail.
As shown in Figure 1, a kind of personalized commercial search method based on deep learning attention mechanism of the invention, including
Following steps：
Step 1：Construct the short-term preference pattern based on attention mechanism.
Specifically, the process of short-term preference pattern of the building based on attention mechanism includes：
Step 1.1：Current inquiry is given, m inquiry and the phase between current queries before estimating using neural network
Pass degree, and indicated using different attention weights；Wherein, m is positive integer；
Due to the inquiry commodity of purchase (or) before and be not all it is closely related with current queries (or end article),
For example, compare with the inquiry " mobile phone " submitted before, " mouse " and " keyboard " obviously with current inquiry " display screen " more
It is close.Due to the reason of discrepancies of inquiry and current queries degree of correlation before this, present invention employs two layers nerves
Network estimates the degree of correlation between them, is herein indicated with different attention weights.
Step 1.2：The attention weight of acquisition is indicated multiplied by corresponding commodity, is input in RNN model, output indicates
M vector of the short-term preference based on attention mechanism.
Corresponding with inquiry, the corresponding purchase commodity of these inquiries before are also intended to the current purchase commodity of user
Generate different influences.So corresponding the bought quotient of inquiry before being determined with inquiry before with the degree of correlation of current queries
The correlation of product and current buying intention, it is specific to indicate to be that the attention weight that obtained before is indicated multiplied by corresponding commodity
To obtain the short-term preference based on attention mechanism.
Before building is based on the short-term preference pattern of attention mechanism, further include：
The inquiry of all comment informations of current commodity and a text is all learned by a unified PV-DM model
The expression of commodity and inquiry is obtained, then this expression is projected in same sub-spaces.
Step 2：Construct the long-term preference pattern based on attention mechanism.
The long-term shopping preferences of user are more stable for short-term, but also will be slow variation.With active user institute
Some purchase commodity slowly update it to indicate long-term preference.
Specifically, the process of long-term preference pattern of the building based on attention mechanism includes：
Step 2.1：It is indicated using m commodity of initial purchase to initialize its long-term preference, and utilizes and be based on building
The same principle of the short-term preference pattern of attention mechanism obtains the long-term preference pattern based on attention mechanism；
Step 2.1：It is every later to pass through m commodity, the long-term preference of an active user is just updated, updated base is obtained
In the long-term preference pattern of attention mechanism.
Step 3：Inquiry indicates again.
Specifically, the process that indicates again of inquiry is：
Merge the short-term preference pattern based on attention mechanism, the long-term preference pattern based on attention mechanism and current
Inquiry, learns the interactive relation between three by multilayer fully-connected network, and the inquiry recombinated indicates, and uses one
Distance function defines the degrees of correlation of all commodity and current queries.
Wherein, the distance function is COS distance or dot product or Euclidean distance or manhatton distance.
Step 4：After the training multilayer fully-connected network, to the new inquiry that each user submits, all quotient are obtained
The distance value of product and current queries, then all distance values are ranked up from high to low, and by the corresponding commodity of preceding n distance value
Return to user, wherein n is positive integer.
Such as：N is 10 or 20.
In another embodiment, this method further includes：
The multilayer fully-connected network is trained using the learning method of " based in pairs ".
It can be improved the robustness of the multilayer fully-connected network to learn in this way, wherein BPR (Bayes's personalized ordering) damage
The loss that function is used for the end is lost, the SGD (stochastic gradient descent) with momentum is used optimization method.
In order to alleviate the problems faced in existing method, the present invention can be in conjunction with the metastable long-term of user
Shopping preferences and personalized retrieval is carried out to commodity to the short-term shopping preferences of time-sensitive, finally improves the accurate of retrieval
Degree, to improve the retrieval experience of user.
The inquiry knot that the present invention can effectively be submitted the long-term and short-term shopping preferences of user and active user
It closes, and indicates the shopping need of user again.
Two respective attention mechanism in the modeling of long-term and short-term shopping preferences that the present invention passes through user, can dash forward
Factor relevant to current queries in two preferences out.
The present invention improves the accuracy of personalized commercial retrieval, to retain to a certain extent for e-commerce website
More users and raising income.
The present invention also provides a kind of, and the personalized commercial based on deep learning attention mechanism retrieves device.
A kind of personalized commercial based on deep learning attention mechanism of the invention retrieves device, including personalized commercial
Retrieval process device, as shown in Fig. 2, the personalized commercial retrieval process device includes：
(1) short-term preference pattern constructs module, is configured as：Construct the short-term preference pattern based on attention mechanism.
The short-term preference pattern building module specifically includes：
(1.1) attention Weight Acquisition submodule, is configured as：Current inquiry is given, is estimated using neural network
Degree of correlation before meter between m inquiry and current queries, and indicated using different attention weights；Wherein, m is positive whole
Number.
Due to the inquiry commodity of purchase (or) before and be not all it is closely related with current queries (or end article),
For example, compare with the inquiry " mobile phone " submitted before, " mouse " and " keyboard " obviously with current inquiry " display screen " more
It is close.Due to the reason of discrepancies of inquiry and current queries degree of correlation before this, present invention employs two layers nerves
Network estimates the degree of correlation between them, is herein indicated with different attention weights.
(1.2) short-term preference indicates submodule, is configured as：By the attention weight of acquisition multiplied by corresponding commodity list
Show, be input in RNN model, output indicates m vector of the short-term preference based on attention mechanism.
Corresponding with inquiry, the corresponding purchase commodity of these inquiries before are also intended to the current purchase commodity of user
Generate different influences.So corresponding the bought quotient of inquiry before being determined with inquiry before with the degree of correlation of current queries
The correlation of product and current buying intention, it is specific to indicate to be that the attention weight that obtained before is indicated multiplied by corresponding commodity
To obtain the short-term preference based on attention mechanism.
Before building is based on the short-term preference pattern of attention mechanism, all comment informations of current commodity and one
The inquiry of text projects to together all by a unified PV-DM model come the expression of learn commodity and inquiry, then by this expression
In one sub-spaces.
(2) long-term preference pattern constructs module, is configured as：Construct the long-term preference pattern based on attention mechanism.
The long-term shopping preferences of user are more stable for short-term, but also will be slow variation.With active user institute
Some purchase commodity slowly update it to indicate long-term preference.
The long-term preference pattern building module specifically includes：
(2.1) long-term preference initialization and building submodule, are configured as：It is indicated using m commodity of initial purchase
Initialize its long-term preference, and using and short-term preference pattern of the building based on attention mechanism same principle, obtain base
In the long-term preference pattern of attention mechanism；
(2.2) long-term preference updates submodule, is configured as：It is every later to pass through m commodity, just update once current use
The long-term preference at family obtains the updated long-term preference pattern based on attention mechanism；
(3) representation module again is inquired, is configured to：It merges the short-term preference pattern based on attention mechanism, be based on paying attention to
The long-term preference pattern and current queries of power mechanism, learn the interactive relation between three by multilayer fully-connected network,
The inquiry recombinated indicates, and the degree of correlation of all commodity and current queries is defined using a distance function；
In the inquiry again representation module, the distance function is COS distance or dot product or Euclidean distance or graceful
Hatton's distance.
(4) commodity return module is configured to：After the training multilayer fully-connected network, each user is submitted
New inquiry, obtain the distance value of all commodity and current queries, then all distance values are ranked up from high to low, and will before
The corresponding commodity of n distance value return to user, wherein n is positive integer.
Such as：N is 10 or 20.
In another embodiment, the personalized commercial retrieval process device further includes：
Model training module is configured as：The multilayer is trained to connect entirely using the learning method of " based in pairs "
Network.
It can be improved the robustness of the multilayer fully-connected network to learn in this way, wherein BPR (Bayes's personalized ordering) damage
The loss that function is used for the end is lost, the SGD (stochastic gradient descent) with momentum is used optimization method.
In order to alleviate the problems faced in existing method, the present invention can be in conjunction with the metastable long-term of user
Shopping preferences and personalized retrieval is carried out to commodity to the short-term shopping preferences of time-sensitive, finally improves the accurate of retrieval
Degree, to improve the retrieval experience of user.
The inquiry knot that the present invention can effectively be submitted the long-term and short-term shopping preferences of user and active user
It closes, and indicates the shopping need of user again.
Two respective attention mechanism in the modeling of long-term and short-term shopping preferences that the present invention passes through user, can dash forward
Factor relevant to current queries in two preferences out.
The present invention improves the accuracy of personalized commercial retrieval, to retain to a certain extent for e-commerce website
More users and raising income.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer program
Product.Therefore, the shape of hardware embodiment, software implementation or embodiment combining software and hardware aspects can be used in the present invention
Formula.Moreover, the present invention, which can be used, can use storage in the computer that one or more wherein includes computer usable program code
The form for the computer program product implemented on medium (including but not limited to magnetic disk storage and optical memory etc.).
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage medium
In, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic
Dish, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random
AccessMemory, RAM) etc..
Above-mentioned, although the foregoing specific embodiments of the present invention is described with reference to the accompanying drawings, not protects model to the present invention
The limitation enclosed, those skilled in the art should understand that, based on the technical solutions of the present invention, those skilled in the art are not
Need to make the creative labor the various modifications or changes that can be made still within protection scope of the present invention.
Claims (8)
1. a kind of personalized commercial search method based on deep learning attention mechanism, which is characterized in that including：
Step 1：Construct the short-term preference pattern based on attention mechanism；The step 1 specifically includes：
Current inquiry is given, the degree of correlation before being estimated using neural network between m inquiry and current queries, and use
Different attention weight indicates；Wherein, m is positive integer；
The attention weight of acquisition is indicated multiplied by corresponding commodity, is input in RNN model, output indicates to be based on attention machine
M vector of the short-term preference of system；
Step 2：Construct the long-term preference pattern based on attention mechanism；The step 2 specifically includes：
It is indicated using m commodity of initial purchase to initialize its long-term preference, and utilized with building based on attention mechanism
The same principle of short-term preference pattern, obtains the long-term preference pattern based on attention mechanism；
It is every later to pass through m commodity, the long-term preference of an active user is just updated, is obtained updated based on attention mechanism
Long-term preference pattern；
Step 3：Inquiry indicates again；Its process is：It merges the short-term preference pattern based on attention mechanism, be based on attention mechanism
Long-term preference pattern and current queries, learn the interactive relation between three by multilayer fully-connected network, obtain weight
The inquiry of group indicates, and the degree of correlation of all commodity and current queries is defined using a distance function；
Step 4：After the training multilayer fully-connected network, to the new inquiry that each user submits, obtain all commodity with
The distance value of current queries, then all distance values are ranked up from high to low, and the corresponding commodity of preceding n distance value are returned
To user, wherein n is positive integer.
2. a kind of personalized commercial search method based on deep learning attention mechanism as described in claim 1, feature
It is, this method further includes：
The multilayer fully-connected network is trained using the learning method of " based in pairs ".
3. a kind of personalized commercial search method based on deep learning attention mechanism as described in claim 1, feature
Be, in the step 1, using one two layers neural network it is related between current queries come m inquiry before estimating
Degree.
4. a kind of personalized commercial search method based on deep learning attention mechanism as described in claim 1, feature
It is, in the step 3, the distance function is COS distance or dot product or Euclidean distance or manhatton distance.
5. a kind of personalized commercial based on deep learning attention mechanism retrieves device, which is characterized in that including personalized quotient
Product examine rope processor, the personalized commercial retrieval process device include：
Short-term preference pattern constructs module, is configured as：Construct the short-term preference pattern based on attention mechanism；It is described short-term
Preference pattern building module specifically includes：
Attention Weight Acquisition submodule, is configured as：Current inquiry is given, is looked into for m before being estimated using neural network
The degree of correlation between current queries is ask, and is indicated using different attention weights；Wherein, m is positive integer；And
Short-term preference indicates submodule, is configured as：The attention weight of acquisition is indicated multiplied by corresponding commodity, is input to
In RNN model, output indicates m vector of the short-term preference based on attention mechanism；
Long-term preference pattern constructs module, is configured as：Construct the long-term preference pattern based on attention mechanism；It is described long-term
Preference pattern building module specifically includes：
Long-term preference initialization and building submodule, are configured as：It is indicated using m commodity of initial purchase to initialize it
Long-term preference, and using the same principle with short-term preference pattern of the building based on attention mechanism, it obtains based on attention machine
The long-term preference pattern of system；
Long-term preference updates submodule, is configured as：It is every later to pass through m commodity, just update the long-term of an active user
Preference obtains the updated long-term preference pattern based on attention mechanism；
Representation module again is inquired, is configured to：Merge short-term preference pattern based on attention mechanism, based on attention mechanism
Long-term preference pattern and current queries, are learnt the interactive relation between three by multilayer fully-connected network, are recombinated
Inquiry indicate, and define using a distance function degree of correlation of all commodity and current queries；
Commodity return module, is configured to：After the training multilayer fully-connected network, newly look into what each user submitted
Ask, obtain the distance value of all commodity and current queries, then all distance values are ranked up from high to low, and by preceding n away from
Commodity corresponding from value return to user, wherein n is positive integer.
6. a kind of personalized commercial based on deep learning attention mechanism as claimed in claim 5 retrieves device, feature
It is, the personalized commercial retrieval process device further includes：
Model training module is configured as：The multilayer fully-connected network is trained using the learning method of " based in pairs ".
7. a kind of personalized commercial based on deep learning attention mechanism as claimed in claim 5 retrieves device, feature
Be, in the attention Weight Acquisition submodule, using one two layers neural network come m inquiry before estimating with currently
Degree of correlation between inquiry.
8. a kind of personalized commercial based on deep learning attention mechanism as claimed in claim 5 retrieves device, feature
It is, in the inquiry again representation module, the distance function is COS distance or dot product or Euclidean distance or Manhattan
Distance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810645528.2A CN108897823B (en) | 2018-06-21 | 2018-06-21 | Personalized commodity retrieval method and device based on deep learning attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810645528.2A CN108897823B (en) | 2018-06-21 | 2018-06-21 | Personalized commodity retrieval method and device based on deep learning attention mechanism |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108897823A true CN108897823A (en) | 2018-11-27 |
| CN108897823B CN108897823B (en) | 2020-03-17 |
Family
ID=64345384
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810645528.2A Active CN108897823B (en) | 2018-06-21 | 2018-06-21 | Personalized commodity retrieval method and device based on deep learning attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108897823B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110321473A (en) * | 2019-05-21 | 2019-10-11 | 山东省计算中心（国家超级计算济南中心） | Diversity preference information method for pushing, system, medium and equipment based on multi-modal attention |
| CN111047394A (en) * | 2019-11-20 | 2020-04-21 | 泰康保险集团股份有限公司 | Commodity recommendation method and device, electronic equipment and computer readable medium |
| CN111435378A (en) * | 2019-01-14 | 2020-07-21 | 中国人民大学 | Query result sorting method and device, electronic equipment and storage medium |
| CN112700274A (en) * | 2020-12-29 | 2021-04-23 | 华南理工大学 | Advertisement click rate estimation method based on user preference |
| CN113826119A (en) * | 2019-05-23 | 2021-12-21 | 谷歌有限责任公司 | Pure attention computer vision |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170127016A1 (en) * | 2015-10-29 | 2017-05-04 | Baidu Usa Llc | Systems and methods for video paragraph captioning using hierarchical recurrent neural networks |
| US9830315B1 (en) * | 2016-07-13 | 2017-11-28 | Xerox Corporation | Sequence-based structured prediction for semantic parsing |
| CN108090801A (en) * | 2017-11-29 | 2018-05-29 | 维沃移动通信有限公司 | Method of Commodity Recommendation, mobile terminal and server |
-
2018
- 2018-06-21 CN CN201810645528.2A patent/CN108897823B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170127016A1 (en) * | 2015-10-29 | 2017-05-04 | Baidu Usa Llc | Systems and methods for video paragraph captioning using hierarchical recurrent neural networks |
| US9830315B1 (en) * | 2016-07-13 | 2017-11-28 | Xerox Corporation | Sequence-based structured prediction for semantic parsing |
| CN108090801A (en) * | 2017-11-29 | 2018-05-29 | 维沃移动通信有限公司 | Method of Commodity Recommendation, mobile terminal and server |
Non-Patent Citations (1)
| Title |
|---|
| 卢湖川等: "目标跟踪算法综述", 《模式识别与人工智能》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111435378A (en) * | 2019-01-14 | 2020-07-21 | 中国人民大学 | Query result sorting method and device, electronic equipment and storage medium |
| CN111435378B (en) * | 2019-01-14 | 2023-09-05 | 中国人民大学 | Query result ordering method and device, electronic equipment and storage medium |
| CN110321473A (en) * | 2019-05-21 | 2019-10-11 | 山东省计算中心（国家超级计算济南中心） | Diversity preference information method for pushing, system, medium and equipment based on multi-modal attention |
| CN110321473B (en) * | 2019-05-21 | 2021-05-25 | 山东省计算中心（国家超级计算济南中心） | Multi-modal attention-based diversity preference information pushing method, system, medium and device |
| CN113826119A (en) * | 2019-05-23 | 2021-12-21 | 谷歌有限责任公司 | Pure attention computer vision |
| CN111047394A (en) * | 2019-11-20 | 2020-04-21 | 泰康保险集团股份有限公司 | Commodity recommendation method and device, electronic equipment and computer readable medium |
| CN112700274A (en) * | 2020-12-29 | 2021-04-23 | 华南理工大学 | Advertisement click rate estimation method based on user preference |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108897823B (en) | 2020-03-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108897823A (en) | Personalized commercial search method and device based on deep learning attention mechanism | |
| CN112313697B (en) | System and method for generating interpretable description-based recommendations for description angle augmentation | |
| US11216503B1 (en) | Clustering search results | |
| CN103514239B (en) | A kind of integrated user behavior and the recommendation method and system of item contents | |
| CN109189904A (en) | Individuation search method and system | |
| CN102214169B (en) | The offer method and device of key word information and target information | |
| CN104331411B (en) | The method and apparatus of recommended project | |
| CN106649774A (en) | Artificial intelligence-based object pushing method and apparatus | |
| US8972399B2 (en) | Ranking based on social activity data | |
| CN103679462A (en) | Comment data processing method and device and searching method and system | |
| CN109360057A (en) | Information-pushing method, device, computer equipment and storage medium | |
| CN105677780A (en) | Scalable user intent mining method and system thereof | |
| CN102360358A (en) | Keyword recommendation method and system | |
| CN103886001A (en) | Personalized commodity recommendation system | |
| CN109670161A (en) | Commodity similarity calculating method and device, storage medium, electronic equipment | |
| CN109978660A (en) | A kind of recommender system off-line training method based on intensified learning frame | |
| CN110321473B (en) | Multi-modal attention-based diversity preference information pushing method, system, medium and device | |
| KR102376652B1 (en) | Method and system for analazing real-time of product data and updating product information using ai | |
| CN110516033A (en) | A kind of method and apparatus calculating user preference | |
| CN116108267A (en) | Recommendation method and related equipment | |
| CN107077392A (en) | That is changed for visualizing prefetches cache | |
| CN109410001A (en) | A kind of Method of Commodity Recommendation, system, electronic equipment and storage medium | |
| Zhou et al. | Multiple attribute decision making based on hesitant fuzzy Einstein geometric aggregation operators | |
| CN109684561B (en) | Interest point recommendation method based on deep semantic analysis of user sign-in behavior change | |
| Serrano | Smart internet search with random neural networks |
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 |