CN105631031B - A kind of imperial palace dress ornament feature selection approach and device - Google Patents
A kind of imperial palace dress ornament feature selection approach and device Download PDFInfo
- Publication number
- CN105631031B CN105631031B CN201511027272.1A CN201511027272A CN105631031B CN 105631031 B CN105631031 B CN 105631031B CN 201511027272 A CN201511027272 A CN 201511027272A CN 105631031 B CN105631031 B CN 105631031B
- Authority
- CN
- China
- Prior art keywords
- node
- dress ornament
- network
- imperial palace
- merging
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
Landscapes
- Engineering & Computer Science (AREA)
- Library & Information Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention discloses a kind of imperial palace dress ornament feature selection approach, are applied to electronic equipment, and the electronic equipment determines the characteristic quantity to be selected of every imperial palace dress ornament；Network after determining imperial palace dress ornament network and its merging；Calculate network transitions probability matrix after merging；The first matrix is obtained according to transition probability matrix；Determine nodes initial distribution probability after merging；The initial distribution probability vector is multiplied by first matrix, network node shifts distribution probability after being merged；The coefficient of several most relevance collection and the most relevance collection is obtained according to the transfer distribution probability, obtain each most relevance centralized node shared characteristic quantity and its number, it is used in combination the number to be multiplied by the coefficient of the most relevance collection, by the maximum several characteristic quantities of product characteristic quantity as a result.Characteristic quantity is selected from the characteristic quantity of the imperial palace dress ornament with correlation in the embodiment of the present invention, the feature that can represent imperial palace dress ornament can be selected most from a large amount of imperial palace garment ornament.
Description
Technical field
The present invention relates to Data Mining, more particularly to a kind of imperial palace dress ornament feature selection approach and device.
Background technology
Court dress is decorated with a large amount of pattern and retains, while associated specialist scholar writes a large amount of monograph texts about imperial palace dress ornament
It offers, imperial palace dress ornament has been carried out greatly from all various aspects such as historical origin, social cChange, dress ornament content, artistic style, quality material
Measure thoroughgoing and painstaking research work.So for the dress ornament of imperial palace, feature can come from pattern, can be from describing
The word etc. of property.And there are various contacts, such as the pattern different identification of imperial palace dress ornament between the dress ornament of different imperial palaces
Different identity, material different identification grade difference etc..
A large amount of feature can be obtained from the dress ornament of imperial palace, palace can most be represented by needing to select from a large amount of garment ornament
The feature of court of a feudal ruler dress ornament, so that related personnel studies and uses.
Existing Feature Selection has priori (Apriori) algorithm and its serial innovatory algorithm, and substantially process is to wait for
It selects in characteristic set, high word frequency or the feature of statistical measures is exported as feature, but since the sample of imperial palace service is with non-
The characteristics of being independently distributed has caused the prior art to apply when feature selecting is attended in imperial palace, it may appear that selection feature cannot be abundant
The case where reflecting imperial palace dress ornament object.
Invention content
The embodiment of the invention discloses a kind of imperial palace dress ornament feature selection approach, can be selected from a large amount of garment ornament
Go out most represent the feature of imperial palace dress ornament.
In order to achieve the above objectives, the embodiment of the invention discloses a kind of imperial palace dress ornament feature selection approach, are applied to electronics
Equipment, the method includes the steps：
Determine the characteristic quantity to be selected of the preset imperial palace dress ornament pond imperial palaces Zhong Meijian dress ornament；
Using every imperial palace dress ornament as node, the characteristic quantity to be selected having using the dress ornament is appointed as the degree of corresponding node
Often there are one identical characteristic quantities to be selected for tool between meaning imperial palace dress ornament, then generate a line between corresponding node, obtain imperial palace
Dress ornament network；
Each node in the imperial palace dress ornament network is merged by the degree of node, network after being merged；
Calculate after the merging each node in network to each node including itself a transition probability,
Obtain transition probability matrix；
Calculate the k power of the transition probability matrix, wherein k values be since 2 ing, successively add 1 integer, until obtaining
The transition probability matrix k power in all off diagonal elements be less than preset first threshold, take the transfer general
The k-1 power of rate matrix is as the first matrix；
The initial distribution probability for determining each node in network after the merging, obtains initial distribution probability vector；
The initial distribution probability vector is multiplied by first matrix, the transfer of each node of network point after being merged
Cloth probability；
By transition probability between every group in network after merging a pair of of node with connection relation, at least one is more than first
The threshold value and set to the node transfer distribution probability node that is all higher than second threshold is determined as most relevance collection, obtains more
A most relevance collection；
The transfer distribution probability for each node that each most relevance is concentrated is added as corresponding most relevance collection
Coefficient, obtain the shared characteristic quantity of each most relevance centralized node, and it is all in the most relevance collection to count this feature amount
The number occurred in node is used in combination the number to be multiplied by the coefficient of the most relevance collection, obtains corresponding weighted value；
The weighted value is ranked up from big to small, takes feature of the sequence corresponding to the weighted value of preceding preset quantity
It measures, as a result characteristic quantity.
Preferably, the characteristic quantity to be selected of the determination preset imperial palace dress ornament pond imperial palaces Zhong Meijian dress ornament, including：
For every imperial palace dress ornament is preserved in the preset imperial palace dress ornament pond, if local preserve the court dress
Corresponding description text is adornd, then natural language processing tool is utilized to obtain the participle mark of the corresponding description text of the imperial palace dress ornament
Note, using the participle mark as the feature to be selected of the imperial palace dress ornament；If local preserve the corresponding image of imperial palace dress ornament,
Using image processing tool, texture, background colour or the contrast of the corresponding image of imperial palace dress ornament are obtained, as the imperial palace dress ornament
Characteristic quantity to be selected.
Preferably, described each node that calculates after the merging in network is to one of each node including itself
Secondary transition probability obtains transition probability matrix, including：
For each node in network after the merging, using the node as the node that sets out, set out node described in definition
It is 0 to the transition probability of itself, the session number of set out described in definition node to specific other nodes accounts for the node itself that sets out
The ratio of total session number sets out node to the transition probability of specific other nodes as described in, obtains transition probability matrix.
Preferably, after the determination merging in network each node initial distribution probability, it is general to obtain initial distribution
Rate vector, including：
Determine after the merging number of nodes of each node before merging in network；
Total node number in the imperial palace dress ornament network is accounted for the number of nodes before each node merges in network after the merging
Ratio, it is as the initial distribution probability of the node, the initial distribution of each node in network after the obtained merging is general
Rate is combined into the initial distribution probability vector.
Preferably, described by transition probability at least one between every group in network after merging a pair of of node with connection relation
It is a to be determined as most relevance more than first threshold and the set to the node transfer distribution probability node that is all higher than second threshold
Collection, obtains multiple most relevance collection, including：
It will be with the main even node using the node as main even node for each node in network after the merging
Other nodes of connection are used as by even node, for each by even node, judge that the main even node is transferred to this and is even saved
The transition probability of point or the connected node are transferred to the main transition probability for connecting node, and whether at least one is more than first
Threshold value；If so, judging the main even node and whether being both greater than the second threshold by the transfer distribution probability of even node；If
Be, then by it is described it is main even node with by even node be determined as have be associated with access；
To all there is the node of association access to take out between arbitrary two node, obtain multiple most relevance collection.
The embodiment of the invention also discloses a kind of imperial palace dress ornament feature selecting devices, are applied to electronic equipment, described device
Including：
Characteristic quantity determining module, the characteristic quantity to be selected for determining the preset imperial palace dress ornament pond imperial palaces Zhong Meijian dress ornament；
Dress ornament network module is used for using every imperial palace dress ornament as node, the characteristic quantity to be selected having with the dress ornament
For the degree of corresponding node, often there are one identical characteristic quantities to be selected for tool between arbitrary imperial palace dress ornament, then are generated between corresponding node
One line obtains imperial palace dress ornament network；
Merging module is closed for merging each node in the imperial palace dress ornament network by the degree of node
And rear network；
Transition probability matrix generation module, for calculating after the merging in network each node to including itself
Each node a transition probability, obtain transition probability matrix；
First matrix generation module, the k power for calculating the transition probability matrix, wherein k values be since 2,
Successively plus 1 integer, until all off diagonal elements in the k power of the obtained transition probability matrix be less than it is preset
First threshold takes the k-1 power of the transition probability matrix as the first matrix；
Initial distribution probability vector generation module, the initial distribution for determining each node in network after the merging are general
Rate obtains initial distribution probability vector；
Transfer distribution probability determining module is obtained for the initial distribution probability vector to be multiplied by first matrix
The transfer distribution probability of each node of network after merging；
Most relevance collection determining module, for being shifted between every group of a pair of of node with connection relation in network after merging
Probability at least one be more than first threshold and the set to the node transfer distribution probability node that is all higher than second threshold is true
It is set to most relevance collection, obtains multiple most relevance collection；
The transfer distribution probability of weighting block, each node for concentrating each most relevance is added conduct pair
The coefficient for answering most relevance collection obtains the shared characteristic quantity of each most relevance centralized node, and count this feature amount this most
The number occurred in all nodes of big incidence set, is used in combination the number to be multiplied by the coefficient of the most relevance collection, obtains corresponding add
Weights；
As a result determining module takes sequence in preceding preset quantity for the weighted value to be ranked up from big to small
Characteristic quantity corresponding to weighted value, as a result characteristic quantity.
Preferably, the characteristic quantity determining module, is specifically used for：
For every imperial palace dress ornament is preserved in the preset imperial palace dress ornament pond, if local preserve the court dress
Corresponding description text is adornd, then natural language processing tool is utilized to obtain the participle mark of the corresponding description text of the imperial palace dress ornament
Note, using the participle mark as the feature to be selected of the imperial palace dress ornament；If local preserve the corresponding image of imperial palace dress ornament,
Using image processing tool, texture, background colour or the contrast of the corresponding image of imperial palace dress ornament are obtained, as the imperial palace dress ornament
Characteristic quantity to be selected.
Preferably, the transition probability matrix generation module, is specifically used for：
For each node in network after the merging, using the node as the node that sets out, set out node described in definition
It is 0 to the transition probability of itself, the session number of set out described in definition node to specific other nodes accounts for the node itself that sets out
The ratio of total session number sets out node to the transition probability of specific other nodes as described in, obtains transition probability matrix.
Preferably, the initial distribution probability vector generation module, including：
Number of nodes determination sub-module, for determining after the merging number of nodes of each node before merging in network；
Submodule is combined, for accounting for the imperial palace dress ornament with the number of nodes before each node merges in network after the merging
The ratio of total node number in network will each be saved as the initial distribution probability of the node in network after the obtained merging
The initial distribution probabilistic combination of point is at the initial distribution probability vector.
Preferably, the most relevance collection determining module, is specifically used for：
It will be with the main even node using the node as main even node for each node in network after the merging
Other nodes of connection are used as by even node, for each by even node, judge that the main even node is transferred to this and is even saved
The transition probability of point or the connected node are transferred to the main transition probability for connecting node, and whether at least one is more than first
Threshold value；If so, judging the main even node and whether being both greater than the second threshold by the transfer distribution probability of even node；If
Be, then by it is described it is main even node with by even node be determined as have be associated with access；
To all there is the node of association access to take out between arbitrary two node, obtain multiple most relevance collection.
As seen from the above technical solutions, an embodiment of the present invention provides a kind of imperial palace dress ornament feature selection approach, applications
In electronic equipment, the electronic equipment determines the characteristic quantity to be selected of the preset imperial palace dress ornament pond imperial palaces Zhong Meijian dress ornament；It will be described
For every imperial palace dress ornament as node, the characteristic quantity to be selected having using the dress ornament is every between arbitrary imperial palace dress ornament as the degree of corresponding node
There are one identical characteristic quantities to be selected for tool, then generate a line between corresponding node, obtain imperial palace dress ornament network；It will be described
Each node in the dress ornament network of imperial palace is merged by the degree of node, network after being merged；Calculate network after the merging
In each node to a transition probability of each node including itself, obtain transition probability matrix；Described in calculating
The k power of transition probability matrix, wherein k values be since 2 ing, successively add 1 integer, until the obtained transition probability square
All off diagonal elements in the k power of battle array are less than preset first threshold, and the k-1 power of the transition probability matrix is taken to make
For the first matrix；The initial distribution probability for determining each node in network after the merging, obtains initial distribution probability vector；It will
The initial distribution probability vector is multiplied by first matrix, the transfer distribution probability of each node of network after being merged；It will
After merging in network between every group of a pair of of node with connection relation transition probability at least one be more than first threshold and this is right
The set for the node that node transfer distribution probability is all higher than second threshold is determined as most relevance collection, obtains multiple most relevances
Collection；The transfer distribution probability for each node that each most relevance is concentrated is added and is as corresponding most relevance collection
Number obtains the shared characteristic quantity of each most relevance centralized node, and counts this feature amount in all nodes of most relevance collection
The number of middle appearance is used in combination the number to be multiplied by the coefficient of the most relevance collection, obtains corresponding weighted value；By the weighted value
It is ranked up from big to small, takes characteristic quantity of the sequence corresponding to the weighted value of preceding preset quantity, as a result characteristic quantity.By
The correlation between the dress ornament of imperial palace is determined using the transition probability size of imperial palace dress ornament, and in the embodiment of the present invention from phase
The characteristic quantity for selecting most to represent imperial palace dress ornament essence in the characteristic quantity of the imperial palace dress ornament of relevance, can be from a large amount of court dress
The feature that can most represent imperial palace dress ornament is selected in decorations feature.
Description of the drawings
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
Obtain other attached drawings according to these attached drawings.
Fig. 1 is a kind of flow diagram of imperial palace dress ornament feature selection approach provided in an embodiment of the present invention；
Fig. 2 is dress ornament network diagram in imperial palace provided by the invention；
Fig. 3 is corresponding to network diagram after the merging of Fig. 2；
Fig. 4 is the imperial palace dress ornament network signal in a kind of imperial palace dress ornament feature selection approach specific example provided by the invention
Figure；
Fig. 5 is the structural schematic diagram of device provided in an embodiment of the present invention.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation describes, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other without creative efforts
Embodiment shall fall within the protection scope of the present invention.
Below by specific embodiment, the present invention is described in detail.
Fig. 1 is a kind of flow diagram of imperial palace dress ornament feature selection approach provided in an embodiment of the present invention, the method
Applied to electronic equipment, this method may include step：
S101：Determine the characteristic quantity to be selected of the preset imperial palace dress ornament pond imperial palaces Zhong Meijian dress ornament.
The relevant information of imperial palace dress ornament to be analyzed, final purpose of the present invention are stored in the preset imperial palace dress ornament pond
The relevant information according to these imperial palace dress ornaments is sought to, determines the feature of imperial palace dress ornament.For the preset imperial palace dress ornament pond
In preserve every imperial palace dress ornament, preserve the corresponding description text of the imperial palace dress ornament if local, utilize natural language
Handling implement obtains the participle mark of the corresponding description text of the imperial palace dress ornament, using participle mark as the imperial palace dress ornament
Feature to be selected；If local preserve the corresponding image of imperial palace dress ornament, image processing tool is utilized, the imperial palace dress ornament pair is obtained
Texture, background colour or the contrast for the image answered, the characteristic quantity to be selected as the imperial palace dress ornament.
S102：Using every imperial palace dress ornament as node, the characteristic quantity to be selected having using the dress ornament is corresponding node
It spends, often there are one identical characteristic quantities to be selected for tool between arbitrary imperial palace dress ornament, then generate a line between corresponding node, obtain
Imperial palace dress ornament network.
The imperial palace dress ornament network is the network with line between node and node, wherein one palace of each node on behalf
Court of a feudal ruler dress ornament, how many line between each two node, it is meant that how many is identical to be selected between this two pieces imperial palace dress ornament
Characteristic quantity.Imperial palace dress ornament network as shown in Figure 2, the network have 4 nodes, respectively node 1, node 2, node 3 and node
4, wherein there are two identical characteristic quantity to be selected between node 1 and node 2, have between node 1 and node 33 it is identical to be selected
Characteristic quantity has 1 identical characteristic quantity to be selected between node 1 and node 4, have between node 2 and node 32 it is identical to be selected
Characteristic quantity has 2 identical characteristic quantities to be selected, does not have identical spy to be selected between node 3 and node 4 between node 2 and node 4
Sign amount.Wherein, the degree of node 1 is 16, indicates that node 1 has 16 characteristic quantities to be selected in total, the degree of node 2 is 15, indicates section
Point 2 has 15 characteristic quantities to be selected in total, and the degree of node 3 is 14, indicates that node 3 has 14 characteristic quantities to be selected, node 4 in total
Degree be 16, indicate node 4 in total have 16 characteristic quantities to be selected.
S103：Each node in the imperial palace dress ornament network is merged by the degree of node, network after being merged.
In the imperial palace dress ornament network will there is mutually unison node to merge into a node, the section being replaced after merging
Point is all established by the node after merging with the connection relation of other nodes.As shown in Fig. 2, the degree of node 1 and node 4 is all
It is 16, node 1 and node 4 merge, and obtain network as shown in Figure 3, and in Fig. 3, node 1 ' is Fig. 2 interior joints 1 and node 4
Node after merging, the node have 4 lines with node 2, have 3 lines with node 3.
S104：Calculate primary transfer of each node to each node including itself in network after the merging
Probability obtains transition probability matrix.
For each node in network after the merging, using the node as the node that sets out, set out node described in definition
It is 0 to the transition probability of itself, the session number of set out described in definition node to specific other nodes accounts for the node itself that sets out
The ratio of total session number sets out node to the transition probability of specific other nodes as described in, obtains transition probability matrix.Such as
In Fig. 3, the transition probability of node 1 ' to node 2 is 4/7, and the transition probability of node 1 ' to node 3 is 3/7, and node 1 ' arrives itself
Transition probability be 0；The transition probability of node 2 to node 1 ' is 4/6, and the transition probability of node 2 to node 3 is 2/6, node 2
It is 0 to the transition probability of itself；The transition probability of node 3 to node 1 ' is 3/5, and the transition probability of node 3 to node 2 is 2/5,
It is 0 that node 3, which arrives the transition probability of itself,；In this manner it is possible to obtain the transition probability matrix：
S105：Calculate the k power of the transition probability matrix, wherein k values be since 2 ing, successively add 1 integer, directly
It is less than preset first threshold to all off diagonal elements in the k power of the obtained transition probability matrix, takes described
The k-1 power of transition probability matrix is as the first matrix.
Calculate 2 power of the transition probability matrix, 4 power of the transition probability matrix that 3 power ..., hypothesis obtain
In all off diagonal elements be less than preset first threshold, then using 3 power of the transition probability matrix as the first square
Battle array.
S106：The initial distribution probability for determining each node in network after the merging, obtains initial distribution probability vector.
Specifically, determining after the merging number of nodes of each node before merging in network.For example, can be according to merging
Afterwards in network each node the number of degrees, in the imperial palace dress ornament network statistics with the number of degrees node number to get
Number of nodes of each node before merging in network after to the merging.
S107：The initial distribution probability vector is multiplied by first matrix, each node of network after being merged
Shift distribution probability.
S108：By transition probability between every group in network after merging a pair of of node with connection relation, at least one is more than
The first threshold and set to the node transfer distribution probability node that is all higher than second threshold is determined as most relevance collection, obtains
To multiple most relevance collection.
It will be with the main even node using the node as main even node for each node in network after the merging
Other nodes of connection are used as by even node, for each by even node, judge that the main even node is transferred to this and is even saved
The transition probability of point or the connected node are transferred to the main transition probability for connecting node, and whether at least one is more than first
Threshold value；If so, judging the main even node and whether being both greater than the second threshold by the transfer distribution probability of even node；If
Be, then by it is described it is main even node with by even node be determined as have be associated with access；
To there is the node of association access to take out between arbitrary two node, obtain multiple most relevance collection.
As shown in Figure 3, it is assumed that the distribution probability of 3 nodes shown in Fig. 3 is all higher than 1/4, and assumes that first threshold is 1/
2, second threshold 1/4.The transfer of the 1st step is considered first：Consider from node 1, the transition probability to node 2 is 4/7>1/
2, the transition probability to node 3 is 3/7<1/2, therefore node 1 and node 2 belong to same incidence set, node 1 cannot be true with node 3
Surely belong to same incidence set；Consider from node 2s, the transition probability to node 1 is 2/3>1/2, the transfer to node 3 is general
Rate is 1/3<1/2, therefore node 1 and node 2 belong to same incidence set (when the first step has verified that node 2 belongs to the incidence set of node
Afterwards, the step is negligible), node 1 not can determine that with node 3 belongs to same incidence set；Consider from node 3s, to node 1
Transition probability is 3/5>1/2, the transition probability to node 2 is 2/5<1/2, in conjunction with node 1,2 conclusion of node, it is known that node 3 with
Node 1 belongs to same incidence set, and node 3 is not belonging to same incidence set with node 2.In conjunction with above-mentioned conclusion, it is known that node 1, node 2
Belong to same incidence set, node 3, node 1 belong to same incidence set.
Node 1 belongs to 2 incidence sets simultaneously at this time, illustrates node 1 while there are two the features of incidence set for tool.
S109：The transfer distribution probability for each node that each most relevance is concentrated is added as corresponding most high point
The coefficient for joining collection obtains the shared characteristic quantity of each most relevance centralized node, and counts this feature amount in the most relevance collection
The number occurred in all nodes is used in combination the number to be multiplied by the coefficient of the most relevance collection, obtains corresponding weighted value.
S110：The weighted value is ranked up from big to small, takes sequence corresponding to the weighted value of preceding preset quantity
Characteristic quantity, characteristic quantity as a result.
In order to which the content of each steps of S103 to S110 is better described, illustrated using following examples.
Text chunk is described for different imperial palace dress ornaments：
Section 1：" dragon " is the symbol of the Chinese nation, and dragon becomes the symbol for representing power and grade after entering class society
Number, imperial robe becomes the symbol of emperor, and dragon becomes when representing the royal rich and honour symbol with power, and looks also transfer to become prestige
Sternly, splendid.
Section 2：Emperor's, imperial gown azurite color embroider five pawls front gold dragon four, two shoulders and each one front and back, with the five colours；Supreme Being
The dragon of clothes is mended, and is that a positive dragon is embroidered in front, that is, embroiders tap front, imperial body then spirals agglomerating, it appears that and it is the symbol for occupying stably rivers and mountains, meaning
Justice rises dragon pattern than the successive dynasties seems honorable.
Section 3：In Qing Dynasty's imperial palace trapping patterns, mainly there are positive dragon, row dragon, rolling dragon grain pattern, be during which decorated with multicolored Yunlong.One
In terms of being moulding, imperial moulded form is curved shape, the dynamic dignified and sovereign power by emperor of this rhythm
Power shows incisively and vividly.
Verbal description shown in section 1 passes through semantic analysis Algorithm Analysis, 16 characteristic quantities to be selected is obtained, by word shown in section 2
Semantic analysis Algorithm Analysis is passed through in description, obtains 15 characteristic quantities to be selected, and verbal description shown in section 3 is passed through semantic analysis algorithm
Analysis, obtains 14 characteristic quantities to be selected.Wherein, same characteristic features amount has 2 between section 1 and section 2, there is same characteristic features amount between section 1 and section 3
3, same characteristic features amount has 2 between section 2 and section 3.
Imperial palace dress ornament network according to Fig.4, can be obtained, and the transition probability of node 1 to node 2 is 2/5, node 1 to section
The transition probability of point 3 is 3/5, and node 1 to the transition probability of itself is 0；The transition probability of node 2 to node 1 is 1/2, node 2
Transition probability to node 3 is 1/2, and node 2 to the transition probability of itself is 0；The transition probability of node 3 to node 1 is 3/5,
The transition probability of node 3 to node 2 is 2/5, and node 3 to the transition probability of itself is 0；It can obtain the transition probability of the network
Matrix is：
The probability transfer matrix is subjected to square operation, obtains the second matrix：
If preset first threshold is 1/2, in addition to diagonal entry, remaining element is respectively less than described second matrix
First threshold, it is determined that the transition probability matrix of the network is first matrix.
If preset second threshold is 1/3, because in the element of the transfer distribution probability vector, 11/30>1/3,4/
15<1/3, it is determined that node 1 and node 3 constitute most relevance collection 1, and node 2 constitutes most relevance collection 2.
Since the most relevance concentrates 1 node for node 1 and node 3, then by the transfer distribution probability 11/ of node 1
30 are added with the transfer distribution probability of node 3, obtain being 11/15 corresponding to the coefficient of the most relevance collection 1；Due to it is described most
Node in big incidence set 2 only has node 2, then obtains being 4/15 corresponding to the coefficient of the most relevance collection 2.By most relevance
Collect 1 interior joint 1 and node 3 all characteristic quantities it is nondistinctive put together from the point of view of, the characteristic quantity repeated has：" dragon ", weight
It appears again and has showed 11 times；" power " has repeated 2 times, " prestige ", has repeated 2 times；In 2 only section of most relevance collection
In point 2, the characteristic quantity repeated is：" wealth and rank " has repeated 2 times.
The weighted value for then calculating " dragon " is：11*11/15=121/15；The weighted value of " power " is 2*11/15=22/15；
The weighted value of " prestige " is 2*11/15=22/15；The weighted value of " wealth and rank " is 2*4/15=8/15, if taking the weighted value most
The big corresponding characteristic quantity of preceding 3 weighted values, can access " dragon ", " power " and " prestige ", i.e., by " dragon ", " power " and
The imperial palace garment ornament for the text description corresponding to section 1, section 2 and section 3 that " prestige " goes out as final choice.
Fig. 5 is a kind of imperial palace dress ornament feature selecting device provided in an embodiment of the present invention, is applied to electronic equipment, the dress
It sets and may include：
Characteristic quantity determining module 501, the characteristic quantity to be selected for determining the preset imperial palace dress ornament pond imperial palaces Zhong Meijian dress ornament；
Dress ornament network module 502 is used for using every imperial palace dress ornament as node, the feature to be selected having with the dress ornament
Amount is the degree of corresponding node, and often there are one identical characteristic quantities to be selected for tool between arbitrary imperial palace dress ornament, then raw between corresponding node
At a line, imperial palace dress ornament network is obtained；
Merging module 503 is obtained for merging each node in the imperial palace dress ornament network by the degree of node
Network after merging；
Transition probability matrix generation module 504, for calculating after the merging in network each node to including itself
Transition probability of each node inside, obtains transition probability matrix；
First matrix generation module 505, the k power for calculating the transition probability matrix, wherein k values are to be opened from 2
Beginning, the integer for adding 1 successively, until all off diagonal elements in the k power of the obtained transition probability matrix are less than in advance
If first threshold, take the k-1 power of the transition probability matrix as the first matrix；
Initial distribution probability vector generation module 506, initial point for determining after the merging each node in network
Cloth probability obtains initial distribution probability vector；
Distribution probability determining module 507 is shifted, for the initial distribution probability vector to be multiplied by first matrix, is obtained
The transfer distribution probability of each node of network after to merging；
Most relevance collection determining module 508, between every group of a pair of of node with connection relation in network after merging
Transition probability at least one be more than first threshold and this to node transfer distribution probability be all higher than second threshold node collection
Conjunction is determined as most relevance collection, obtains multiple most relevance collection；
Weighting block 509, the transfer distribution probability of each node for concentrating each most relevance, which is added, to be made
For the coefficient of corresponding most relevance collection, the shared characteristic quantity of each most relevance centralized node is obtained, and count this feature amount and exist
The number occurred in all nodes of most relevance collection, is used in combination the number to be multiplied by the coefficient of the most relevance collection, is corresponded to
Weighted value；
As a result determining module 510 take sequence in preceding preset quantity for the weighted value to be ranked up from big to small
Weighted value corresponding to characteristic quantity, characteristic quantity as a result.
Further, the characteristic quantity determining module 501, is specifically used for：
For every imperial palace dress ornament is preserved in the preset imperial palace dress ornament pond, if local preserve the court dress
Corresponding description text is adornd, then natural language processing tool is utilized to obtain the participle mark of the corresponding description text of the imperial palace dress ornament
Note, using the participle mark as the feature to be selected of the imperial palace dress ornament；If local preserve the corresponding image of imperial palace dress ornament,
Using image processing tool, texture, background colour or the contrast of the corresponding image of imperial palace dress ornament are obtained, as the imperial palace dress ornament
Characteristic quantity to be selected.
Further, the transition probability matrix generation module 504, is specifically used for：
For each node in network after the merging, using the node as the node that sets out, set out node described in definition
It is 0 to the transition probability of itself, the session number of set out described in definition node to specific other nodes accounts for the node itself that sets out
The ratio of total session number sets out node to the transition probability of specific other nodes as described in, obtains transition probability matrix.
Further, the initial distribution probability vector generation module 506, including：
Number of nodes determination sub-module (not shown), for determining, each node is merging in network after the merging
Preceding number of nodes；
Submodule (not shown) is combined, for being accounted for the number of nodes before each node merges in network after the merging
The ratio of total node number in the imperial palace dress ornament network, as the initial distribution probability of the node, after the obtained merging
The initial distribution probabilistic combination of each node is at the initial distribution probability vector in network.
Further, the most relevance collection determining module 508, is specifically used for：
It will be with the main even node using the node as main even node for each node in network after the merging
Other nodes of connection are used as by even node, for each by even node, judge that the main even node is transferred to this and is even saved
The transition probability of point or the connected node are transferred to the main transition probability for connecting node, and whether at least one is more than first
Threshold value；If so, judging the main even node and whether being both greater than the second threshold by the transfer distribution probability of even node；If
Be, then by it is described it is main even node with by even node be determined as have be associated with access；
To all there is the node of association access to take out between arbitrary two node, obtain multiple most relevance collection.
An embodiment of the present invention provides inventive embodiments to provide a kind of imperial palace dress ornament feature selection approach and device, application
In electronic equipment, the electronic equipment determines the characteristic quantity to be selected of the preset imperial palace dress ornament pond imperial palaces Zhong Meijian dress ornament；It will be described
For every imperial palace dress ornament as node, the characteristic quantity to be selected having using the dress ornament is every between arbitrary imperial palace dress ornament as the degree of corresponding node
There are one identical characteristic quantities to be selected for tool, then generate a line between corresponding node, obtain imperial palace dress ornament network；It will be described
Each node in the dress ornament network of imperial palace is merged by the degree of node, network after being merged；Calculate network after the merging
In each node to a transition probability of each node including itself, obtain transition probability matrix；Described in calculating
The k power of transition probability matrix, wherein k values be since 2 ing, successively add 1 integer, until the obtained transition probability square
All off diagonal elements in the k power of battle array are less than preset first threshold, and the k-1 power of the transition probability matrix is taken to make
For the first matrix；The initial distribution probability for determining each node in network after the merging, obtains initial distribution probability vector；It will
The initial distribution probability vector is multiplied by first matrix, the transfer distribution probability of each node of network after being merged；It will
After merging in network between every group of a pair of of node with connection relation transition probability at least one be more than first threshold and this is right
The set for the node that node transfer distribution probability is all higher than second threshold is determined as most relevance collection, obtains several most relevances
Collection；The transfer distribution probability for each node that each most relevance is concentrated is added and is as corresponding most relevance collection
Number obtains the shared characteristic quantity of each most relevance centralized node, and counts this feature amount in all nodes of most relevance collection
The number of middle appearance is used in combination the number to be multiplied by the coefficient of the most relevance collection, obtains corresponding weighted value；Take weighted value maximum
Preset quantity weighted value corresponding to characteristic quantity, characteristic quantity as a result.Due to utilizing court dress in the embodiment of the present invention
The transition probability size of decorations determines the correlation between the dress ornament of imperial palace, and from the characteristic quantity of the imperial palace dress ornament with correlation
Selection can most represent the characteristic quantity of imperial palace dress ornament essence, and can be selected from a large amount of imperial palace garment ornament can most represent imperial palace
The feature of dress ornament.
For systems/devices embodiment, since it is substantially similar to the method embodiment, so the comparison of description is simple
Single, the relevent part can refer to the partial explaination of embodiments of method.
It should be noted that herein, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also include other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in process, method, article or equipment including the element.
One of ordinary skill in the art will appreciate that all or part of step in realization above method embodiment is can
It is completed with instructing relevant hardware by program, the program can be stored in computer read/write memory medium,
The storage medium designated herein obtained, such as：ROM/RAM, magnetic disc, CD etc..
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the scope of the present invention.It is all
Any modification, equivalent replacement, improvement and so within the spirit and principles in the present invention, are all contained in protection scope of the present invention
It is interior.
Claims (10)
1. a kind of imperial palace dress ornament feature selection approach, which is characterized in that it is applied to electronic equipment, the method includes the steps：
Determine the characteristic quantity to be selected of the preset imperial palace dress ornament pond imperial palaces Zhong Meijian dress ornament；
Using every imperial palace dress ornament as node, the characteristic quantity to be selected having using the dress ornament is the degree of corresponding node, arbitrary palace
Often there are one identical characteristic quantities to be selected for tool between the dress ornament of the court of a feudal ruler, then generate a line between corresponding node, obtain imperial palace dress ornament
Network；
Each node in the imperial palace dress ornament network is merged by the degree of node, network after being merged；
Calculate after the merging that each node is obtained to a transition probability of each node including itself in network
Transition probability matrix；
Calculate the k power of the transition probability matrix, wherein k values be since 2 ing, successively add 1 integer, until obtained institute
All off diagonal elements stated in the k power of transition probability matrix are less than preset first threshold, take the transition probability square
The k-1 power of battle array is as the first matrix；
The initial distribution probability for determining each node in network after the merging, obtains initial distribution probability vector；
The initial distribution probability vector is multiplied by first matrix, the transfer distribution of each node of network is general after being merged
Rate；
By transition probability between every group in network after merging a pair of of node with connection relation at least one be more than first threshold,
And the set that the node that distribution probability is all higher than second threshold is shifted to node is determined as most relevance collection, obtains multiple maximums
Incidence set；
The transfer distribution probability for each node that each most relevance is concentrated is added and is as corresponding most relevance collection
Number obtains the shared characteristic quantity of each most relevance centralized node, and counts this feature amount in all nodes of most relevance collection
The number of middle appearance is used in combination the number to be multiplied by the coefficient of the most relevance collection, obtains corresponding weighted value；
The weighted value is ranked up from big to small, takes characteristic quantity of the sequence corresponding to the weighted value of preceding preset quantity,
Characteristic quantity as a result.
2. according to the method described in claim 1, it is characterized in that, every court dress in the determination preset imperial palace dress ornament pond
The characteristic quantity to be selected of decorations, including：
For every imperial palace dress ornament is preserved in the preset imperial palace dress ornament pond, if local preserve the imperial palace dress ornament pair
The description text answered then utilizes the participle that natural language processing tool obtains the corresponding description text of the imperial palace dress ornament to mark, with
To be selected feature of the participle mark as the imperial palace dress ornament；If local preserve the corresponding image of imperial palace dress ornament, utilize
Image processing tool obtains texture, background colour or the contrast of the corresponding image of imperial palace dress ornament, as waiting for for the imperial palace dress ornament
Select characteristic quantity.
3. according to the method described in claim 1, it is characterized in that, each node arrives in network after the calculating merging
Transition probability of each node including itself, obtains transition probability matrix, including：
For each node in network after the merging, using the node as the node that sets out, set out described in definition node to from
The transition probability of body is 0, and the session number of set out described in definition node to specific other nodes accounts for the node itself that sets out and always connects
The ratio of line number sets out node to the transition probability of specific other nodes as described in, obtains transition probability matrix.
4. according to the method described in claim 1, it is characterized in that, after the determination merging in network each node just
Beginning distribution probability obtains initial distribution probability vector, including：
Determine after the merging number of nodes of each node before merging in network；
Number of nodes before being merged with each node in network after the merging accounts for the ratio of total node number in the imperial palace dress ornament network
Value, as the initial distribution probability of the node, by the initial distribution probability group of each node in network after the obtained merging
Synthesize the initial distribution probability vector.
5. according to the method described in claim 1, it is characterized in that, described have connection relation by every group in network after merging
Between a pair of of node transition probability at least one be more than first threshold and this to node transfer distribution probability be all higher than second threshold
The set of node be determined as most relevance collection, obtain multiple most relevance collection, including：
It will be connect with the main even node using the node as main even node for each node in network after the merging
Other nodes be used as by even node, for each by even node, judge that the main even node is transferred to this and is connected node
Transition probability or the connected node are transferred to the main transition probability for connecting node, and whether at least one is more than the first threshold
Value；If so, judging the main even node and whether being both greater than the second threshold by the transfer distribution probability of even node；If
Be, then by it is described it is main even node with by even node be determined as have be associated with access；
To all there is the node of association access to take out between arbitrary two node, obtain multiple most relevance collection.
6. a kind of imperial palace dress ornament feature selecting device, which is characterized in that be applied to electronic equipment, described device includes：
Characteristic quantity determining module, the characteristic quantity to be selected for determining the preset imperial palace dress ornament pond imperial palaces Zhong Meijian dress ornament；
Dress ornament network module, for being pair with the characteristic quantity to be selected that the dress ornament has using every imperial palace dress ornament as node
Answer the degree of node, often there are one identical characteristic quantities to be selected for tool between arbitrary imperial palace dress ornament, then one is generated between corresponding node
Line obtains imperial palace dress ornament network；
Merging module, for merging each node in the imperial palace dress ornament network by the degree of node, after obtaining merging
Network；
Transition probability matrix generation module, for calculating after the merging in network each node to every including itself
Transition probability of a node, obtains transition probability matrix；
First matrix generation module, the k power for calculating the transition probability matrix, wherein k values are since 2, successively
The integer for adding 1, until all off diagonal elements in the k power of the obtained transition probability matrix are less than preset first
Threshold value takes the k-1 power of the transition probability matrix as the first matrix；
Initial distribution probability vector generation module, the initial distribution probability for determining each node in network after the merging,
Obtain initial distribution probability vector；
Transfer distribution probability determining module is merged for the initial distribution probability vector to be multiplied by first matrix
The transfer distribution probability of each node of network afterwards；
Most relevance collection determining module, for transition probability between every group of a pair of of node with connection relation in network after merging
At least one is more than first threshold and the set that the node that distribution probability is all higher than second threshold is shifted to node is determined as
Most relevance collection obtains multiple most relevance collection；
Weighting block, the transfer distribution probability of each node for concentrating each most relevance are added to be used as and correspond to most
The coefficient of big incidence set obtains the shared characteristic quantity of each most relevance centralized node, and counts this feature amount in the most high point
Connection collects the number occurred in all nodes, is used in combination the number to be multiplied by the coefficient of the most relevance collection, obtains corresponding weighted value；
As a result determining module takes sequence in the weighting of preceding preset quantity for the weighted value to be ranked up from big to small
It is worth corresponding characteristic quantity, as a result characteristic quantity.
7. device according to claim 6, which is characterized in that the characteristic quantity determining module is specifically used for：
For every imperial palace dress ornament is preserved in the preset imperial palace dress ornament pond, if local preserve the imperial palace dress ornament pair
The description text answered then utilizes the participle that natural language processing tool obtains the corresponding description text of the imperial palace dress ornament to mark, with
To be selected feature of the participle mark as the imperial palace dress ornament；If local preserve the corresponding image of imperial palace dress ornament, utilize
Image processing tool obtains texture, background colour or the contrast of the corresponding image of imperial palace dress ornament, as waiting for for the imperial palace dress ornament
Select characteristic quantity.
8. device according to claim 6, which is characterized in that the transition probability matrix generation module is specifically used for：
For each node in network after the merging, using the node as the node that sets out, set out described in definition node to from
The transition probability of body is 0, and the session number of set out described in definition node to specific other nodes accounts for the node itself that sets out and always connects
The ratio of line number sets out node to the transition probability of specific other nodes as described in, obtains transition probability matrix.
9. device according to claim 6, which is characterized in that the initial distribution probability vector generation module, including：
Number of nodes determination sub-module, for determining after the merging number of nodes of each node before merging in network；
Submodule is combined, for accounting for the imperial palace dress ornament network with the number of nodes before each node merges in network after the merging
The ratio of middle total node number, as the initial distribution probability of the node, by each node in network after the obtained merging
Initial distribution probabilistic combination is at the initial distribution probability vector.
10. device according to claim 6, which is characterized in that the most relevance collection determining module is specifically used for：
It will be connect with the main even node using the node as main even node for each node in network after the merging
Other nodes be used as by even node, for each by even node, judge that the main even node is transferred to this and is connected node
Transition probability or the connected node are transferred to the main transition probability for connecting node, and whether at least one is more than the first threshold
Value；If so, judging the main even node and whether being both greater than the second threshold by the transfer distribution probability of even node；If
Be, then by it is described it is main even node with by even node be determined as have be associated with access；
To all there is the node of association access to take out between arbitrary two node, obtain multiple most relevance collection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201511027272.1A CN105631031B (en) | 2015-12-30 | 2015-12-30 | A kind of imperial palace dress ornament feature selection approach and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201511027272.1A CN105631031B (en) | 2015-12-30 | 2015-12-30 | A kind of imperial palace dress ornament feature selection approach and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105631031A CN105631031A (en) | 2016-06-01 |
| CN105631031B true CN105631031B (en) | 2018-09-18 |
Family
ID=56045964
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201511027272.1A Active CN105631031B (en) | 2015-12-30 | 2015-12-30 | A kind of imperial palace dress ornament feature selection approach and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105631031B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101278304A (en) * | 2005-09-02 | 2008-10-01 | 皇家创新有限公司 | Feature selection |
| CN101814082A (en) * | 2010-01-20 | 2010-08-25 | 中国人民解放军总参谋部第六十三研究所 | Method for automatic feature weighting and selection in detection of similar and duplicate record based on ant colony optimization |
| CN103593674A (en) * | 2013-11-19 | 2014-02-19 | 太原理工大学 | Cervical lymph node ultrasonoscopy feature selection method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101300247B1 (en) * | 2011-11-11 | 2013-08-26 | 경희대학교 산학협력단 | Markov chain hidden conditional random fields model based pattern recognition method |
-
2015
- 2015-12-30 CN CN201511027272.1A patent/CN105631031B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101278304A (en) * | 2005-09-02 | 2008-10-01 | 皇家创新有限公司 | Feature selection |
| CN101814082A (en) * | 2010-01-20 | 2010-08-25 | 中国人民解放军总参谋部第六十三研究所 | Method for automatic feature weighting and selection in detection of similar and duplicate record based on ant colony optimization |
| CN103593674A (en) * | 2013-11-19 | 2014-02-19 | 太原理工大学 | Cervical lymph node ultrasonoscopy feature selection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105631031A (en) | 2016-06-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gereffi | Paths of industrialization: An overview | |
| CN106021364B (en) | Foundation, image searching method and the device of picture searching dependency prediction model | |
| Adler | Stardom and talent | |
| Aspers et al. | Encounters in the global fashion business: Afterword | |
| CN109657156A (en) | A kind of personalized recommendation method generating confrontation network based on circulation | |
| CN108320187A (en) | A kind of recommendation method based on depth social networks | |
| CN110727813B (en) | Self-adaptive trending index ordering method for commodity pictures | |
| CN106874355A (en) | The collaborative filtering method of social networks and user's similarity is incorporated simultaneously | |
| CN110245228A (en) | The method and apparatus for determining text categories | |
| CN105787662A (en) | Mobile application software performance prediction method based on attributes | |
| CN106202151A (en) | One is used for improving the multifarious method of personalized recommendation system | |
| CN106855851A (en) | Knowledge extraction method and device | |
| CN109725972A (en) | A kind of navigation display method and system | |
| CN106886559A (en) | The collaborative filtering method of good friend's feature and similar users feature is incorporated simultaneously | |
| CN110321473A (en) | Diversity preference information method for pushing, system, medium and equipment based on multi-modal attention | |
| Zakaria et al. | Developing apparel sizing system using anthropometric data: body size and shape analysis, key dimensions, and data segmentation | |
| CN107248023A (en) | A kind of screening technique and device to mark enterprise list | |
| CN107391577A (en) | A kind of works label recommendation method and system based on expression vector | |
| CN207367205U (en) | Intelligent style commending system | |
| CN105631031B (en) | A kind of imperial palace dress ornament feature selection approach and device | |
| CN104376585B (en) | Non-convex compressed sensing image reconstructing method based on picture block structure property policy | |
| CN104766091B (en) | Space and spectrum synergetic structure and multiple dimensioned understanding method, the system of remote sensing image | |
| CN114821202B (en) | Clothing recommendation method based on user preference | |
| CN106919647A (en) | A kind of network structure similitude based on cluster recommends method | |
| CN113449808B (en) | Multi-source image-text information classification method and corresponding device, equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |