CN109189936A - A kind of label semanteme learning method measured based on network structure and semantic dependency - Google Patents
A kind of label semanteme learning method measured based on network structure and semantic dependency Download PDFInfo
- Publication number
- CN109189936A CN109189936A CN201810914904.3A CN201810914904A CN109189936A CN 109189936 A CN109189936 A CN 109189936A CN 201810914904 A CN201810914904 A CN 201810914904A CN 109189936 A CN109189936 A CN 109189936A
- Authority
- CN
- China
- Prior art keywords
- label
- network
- text
- semantic
- random walk
- 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
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000005295 random walk Methods 0.000 claims abstract description 27
- 230000006399 behavior Effects 0.000 claims abstract description 9
- 238000013138 pruning Methods 0.000 claims description 7
- 230000002708 enhancing effect Effects 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000003672 processing method Methods 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000009510 drug design Methods 0.000 abstract description 2
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 9
- 239000010931 gold Substances 0.000 description 9
- 229910052737 gold Inorganic materials 0.000 description 9
- 239000011449 brick Substances 0.000 description 8
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000013508 migration Methods 0.000 description 4
- 230000005012 migration Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000006116 polymerization reaction Methods 0.000 description 2
- 239000000047 product Substances 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 1
- 239000006227 byproduct Substances 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013450 outlier detection Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2216/00—Indexing scheme relating to additional aspects of information retrieval not explicitly covered by G06F16/00 and subgroups
- G06F2216/03—Data mining
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
The present invention relates to a kind of label semanteme learning methods measured based on network structure and semantic dependency, including obtain truth labels network G based on user behavior fact init Tag network;Label network G after constructing specification according to truth labels network GR;In label network GRThe random walk strategy of upper application enhancements constructs the label network G based on random walk strategyC;Label network G is constructed based on the relevant text information of labelT;To label network GC, label network GTIt is normalized, learning label semantic vector by random walk strategy and term vector learning method indicates.The present invention has rational design, not only make full use of network topology structure, the relevant textual information that node includes is taken into account again, can within a short period of time from topological structure and text representation study to be easy to operation, confidence level be high, expression sufficiently, the label semantic vector of low noise, can be widely applied to the label network study and the study of label semanteme of the text collection containing label.
Description
Technical field
The invention belongs to network representation learning art fields, especially a kind of to be measured based on network structure and semantic dependency
Label semanteme learning method.
Background technique
In network representation learning art, text semantic study is mainly to the character representation of text, i.e., by target text
Object (word, sentence, section, a piece) is expressed as numerical value (single value, vector or matrix) form.In view of the need calculated and semantic modeling is applied
It asks, currently used model is latent semantic analysis LSA (the Latent Semantic based on Singular Value Decomposition Using
Analysis), the potential Di Li Cray based on probabilistic model is distributed LDA (Latent Dirichlet Allocation), is based on
The term vector of Neural Networks Solution indicates model NNLM (Neural Network Language Model) and word2vec etc.,
These models are mainly used for long text.In face of high sparse, high noisy short text, researcher is proposed using external corpus,
Such as WordNet, Mesh and Wikipedia are formed to expand original short text corpus, or according to certain rule polymerization short text
The method of " pseudo- document ", still, both methods has obvious drawback, the former is limited to external corpus, and the latter is limited to manually
Rule.By the user behavior analysis to short text, the information extension that user largely uses label to expand short text is found, for
Short text containing label, researcher using label propose Tag-LDA, TWTM (Tag-Weighted Topic Model),
TWDA (Tag-Weighted Dirichlet Allocation), Labeled LDA, MB-LDA and HGTM (Hashtag-
Graph Based Topic Model) the methods of carry out theme semantic modeling, obtain preferable effect.At the same time, model
The vector table for also outputing label is shown as byproduct.
Classification, cluster, outlier detection, link prediction in network research require to link in network sparse
Property, network node is expressed as vector, then can effectively avoid the problem.Important branch in network representation study is to learn net
The vector of network node indicates.It is (Graph Factorization) and special based on Laplce that conventional method mainly has figure to decompose
Levy the method (Laplacian Eigenmaps) etc. of vector.Random walk is that one kind is widely used in progress node phase in figure
Like the method for inquiry learning, Perozzi et al. learns the linear order into network topology structure based on random walk, uses for reference
Mikolov Skip-Gram term vector learning method and learning network node indicate, however the method for Deepwalk is based only upon section
Connection between point is learnt, and the content information that node may include is not accounted for.Yang et al. demonstrates Deepwalk reality
It is equivalent to decompose the transition probability matrix of overlapping network in matter, on this basis, be divided jointly by introducing node diagnostic
Solution, proposes the associated depth migration model of content (Text Associated Deep Walk).Tang et al. is proposed to big
Type network carries out the model LINE (Large-scale Information Network Embedding) of learning model building, but should
Model only considers the once node with two degree, to a large amount of beta prunings of the transmitting of information on network, and three steps four between different nodes
Step or even farther association are the important components of the global structure of network, are had great importance.Based on this, Cao et al.
GraRep is proposed by defining different transfer matrixes, defining different loss functions for different step numbers sufficiently to learn node
Between related information.
In conclusion existing algorithm is difficult to solve original label network from the unconfined collaboration fact of user, information
Height is made an uproar the problem of changeable, theme boundary is unintelligible or topic drift, semantic model fail.
Summary of the invention
It is an object of the invention to overcome the deficiencies in the prior art, propose a kind of based on network structure and semantic dependency degree
The label semanteme learning method of amount, utilize text information and network topology structure, excavate it is stable, label theme can be embodied
Related information obtains the core path of the semantic transmitting of user's common recognition, based on the semantic expressiveness of this study label, solves original
Label network is made an uproar unintelligible changeable, theme boundary or topic drift and language from user's unconfined collaboration fact, information height
The problem of adopted model fails.
The present invention solves its technical problem and adopts the following technical solutions to achieve:
A kind of label semanteme learning method measured based on network structure and semantic dependency, comprising the following steps:
Step 1 is based on user behavior fact init Tag network, obtains truth labels network G;
Step 2 constructs the label network G after specification according to truth labels network GR;
Step 3, in label network GRThe random walk strategy of upper application enhancements constructs the label based on random walk strategy
Network GC;
Step 4 constructs label network G based on the relevant text information of labelT;
Step 5, to label network GC, label network GTIt is normalized, passes through random walk strategy and term vector
Learning method, which learns label semantic vector, to be indicated.
Further, the truth labels network G are as follows:
The label network G={ V, E } based on co-occurrence inside text is defined, V is all labels in entire text collection D;Such as
Any label i of fruit and label j are appeared in simultaneously in a text d, then have a line between them, be denoted as eij;Define the network
The weight g on middle sideijAre as follows:
Wherein, Di,jFor the text collection for simultaneously including label i and j, hdFor the tag set of document d.
Further, the processing method of the step 2 are as follows:
Firstly, considering the associated random noise of semantic node: subtracting branch and remove the stronger weak rigidity of randomness, have to network
Effect side is constrained with noise reduction;Incidence matrix after enabling beta pruning is expressed as T, then wherein element tijAs label i and label j beta pruning
Relating value afterwards, is expressed as follows:
Wherein, δ be can be truncated 20% low frequency side phase threshold, gijFor the weight on side in network;
Then, consider that label node degree of divergence is different, for the side in any one network, according to two, side endpoint institute
The weight of the different adjustment web tab incidence relations of associated endpoint number, enhancing label theme association:
t′ij=tij*log(N/N(i))*log(N/N(j))。
Wherein, t 'ijFor GRThe weight on side in network, number of the N by network comprising node, N (i), N (j) are label i, j
Out-degree on network, logN/N (i) are that current label and the related Probability p (i) of nodes=N (i)/N are reciprocal
It is logarithmic.
Further, the improved random walk strategy of the step 3 be the complex network structures of high noisy are sampled to it is a plurality of
Linear order obtains the microcosmic description information of network part with breadth-first search method, is obtained with Depth Priority Searching
The global macroscopic information of network is slided on linear order using window according to the linear order of sampling and obtains new label pass
Connection relationship, to obtain new side right weight.
Further, the processing method of the step 4 are as follows:
Firstly, there are the lexical set W of cooccurrence relation with label i for definitioni, use wijIndicate time of label i and vocabulary j co-occurrence
Number;
Then, w is weighted using inverse document frequency IDF, word tiIdf value calculating method it is as follows:
Wherein, | D | it is the number of files in set, | { j:ti∈dj| to include word tiNumber of files;
Secondly, calculating w againijWith idfjProduct, obtain the text representation vector of label;
Finally, calculating label text two-by-two indicates the cosine similarity of vector, definitionFor the low frequency side that can be truncated 80%
Interceptive value removes cosine similarity and is less thanSide, retain side weight be labeled as cosine similarity value.
Further, the step 5 introduces figure sampling preference parameter, in label network G in network random walk processT
With label network GCBetween switch, be associated with benefiting from network structure with two kinds of text information;In semantic vector renewal process,
It is regarded sentence by the linear order obtained using network random walk, semantic based on left and right context study label.
The advantages and positive effects of the present invention are:
1, the present invention combine text semantic study and network representation learning art, by multistage noise reduction, network specification, with
Machine migration, Text similarity computing strategy, reduce network noise, obtain core network, and on this basis learn label to
Amount indicates, not only makes full use of network topology structure, and can take into account the relevant textual information that node includes, can be when shorter
In from topological structure and text representation study to be easy to operation, confidence level be high, expression sufficiently, the label semanteme of low noise to
Amount, can be widely applied to the text collection containing label label network study and label semanteme study and text mining it is each
Kind application.
2, the present invention has rational design, and on the basis of factural information, the core for carrying out label related network is abstract, on figure
Side be sampled with weight update, help to obtain stable label network structure, enable model to noise jamming not
Sensitivity promotes modeling generalization.
Detailed description of the invention
Fig. 1 is overall architecture schematic diagram of the invention.
Specific embodiment
The embodiment of the present invention is further described below in conjunction with attached drawing.
Design philosophy of the invention is: being based primarily upon statistical machine learning theory and Text Mining Technology, utilizes user's
Behavioral data constructs label network with a high credibility to learn the pure semantic expressiveness of label.It is true just first with user behavior
Beginningization label network secondly based on specification technology and improves random walk technology reengineering label network, related using label again
Text similarity building label network, finally the label network based on reconstruct and the label network based on text similarity study
Label semantic vector indicates.Network reconfiguration can regard the automatic strong theme association discovery of multipath label as in the present invention
Filter, weaker Partial filtration will be associated with and fallen, the noise associated influence unrelated with theme is reduced, thus help to find with
Enhancing the semantic more consistent and close label network of theme indicates, to obtain the label semantic vector of theme enhancing.
Based on above-mentioned design philosophy, the label semanteme study side of the invention measured based on network structure and semantic dependency
Method, as shown in Figure 1, comprising the following steps:
Step 1: being based on user behavior fact init Tag network, obtain the label network G based on the fact the user behavior.
In internet text this application field, in essence, " label=theme ".Here theme can be fine granularity
, such as specific event, time, place, it is also possible to coarseness, such as news, the abstract concept of polymerization etc..Label
Appearance be that generation is write by the text of user, the relationship between label is that behavior based on user is true.Therefore, naturally
Ground, there are cooccurrence relations with text for label.
Most direct relationship is the cooccurrence relation in single text between label, defines the mark based on co-occurrence inside text
Network G={ V, E } is signed, V is all labels in entire text collection D.If any label i and label j appears in one simultaneously
In text d, then there is a line between them, be denoted as eij.Define the weight g on side in the networkijFor simultaneously comprising the two labels
Textual data.I.e.
Wherein, Di,jFor the text collection for simultaneously including label i and j, hdFor the tag set of document d.Intuitively, two
Label appears in the more text of quantity simultaneously, and the weight of the two line is bigger, illustrates that the semantic association of the two is closer.
So far, the label network based on the user behavior fact has been obtained.
Other than the cooccurrence relation in text described above, this step further comprises making for indirect utilization user and label
The label fact network constructed with relationship or the cooccurrence relation of external linkage and label.
Step 2: based on the fact that the label network G after label network G building specificationR。
The label network noise for being directly based upon true building is larger, these noises mostly come from the randomness that user uses
And otherness.It is arbitrarily that randomness, which refers to that user adds label for text, it is more likely that there is a situation where that more words are synonymous, example
Such as " Tsinghua University " and " Tsing-Hua University ";Otherness refers to that the difference of the intension to one text expression understands, so that add different
Label, such as different user are " can wear and how much wear how many " addition label " wrapping up in wadded jacket " or " bikini ".
In order to agglomerate semanteme, this step is based on four simple hypothesis.(1) two label degree of incidence is more, between label
Theme association it is closer.The identical neighbour of (2) two labels is more, and the theme association between label is closer.(3) same to theme
Tag set have the building-up effect of similar block, community.(4) label relevant to a large amount of labels is associated with the theme of other labels
It is not close instead.
The associated random noise of semantic node is considered first.Subtract branch and remove the stronger weak rigidity of randomness, has to network
Effect side is constrained with noise reduction.Incidence matrix after enabling beta pruning is expressed as T, then wherein element tijAs label i and label j beta pruning
Relating value afterwards.
Wherein, δ be can be truncated 20% low frequency side phase threshold.
Secondly consider that label node degree of divergence is different, for the side in any one network, according to two, side endpoint institute
The weight of the different adjustment web tab incidence relations of associated endpoint number, enhancing label theme association:
t′ij=tij*log(N/N(i))*log(N/N(j))。
Herein, t 'ijFor GRThe weight on side in network, number of the N by network comprising node, N (i), N (j) are label i, j
Out-degree on network, i.e., and the associated number of edges in the vertex.LogN/N (i) is that current label is relevant with nodes
Reciprocal logarithmic of Probability p (i)=N (i)/N, embody the information content of label i in a network, can be described as inverse association frequency
(IRF, Inverse Relation Frequency), the value is bigger, illustrates that theme semantic space indexes better, theme semantic association
Middle effective information is more.
Step 3: the label network G after step 2 obtains specificationRUpper application enhancements random walk strategy building based on
The label network G of machine migration strategyC。
This step is based on improved random walk strategy and samples the relevant node of each node.In step 2, we obtain
The weight t ' of network topij, defined herein two degree of transfer weights instruct to shift.It defines after from label t migration to label i, under
One step can arrive the not normalized transition probability of label j are as follows:
WhereinFor regulatory factor, it is defined as follows:
dtjThe shortest distance on the figure of label t to label j is defined, only defines the node that distance is up to 2 herein.Label t
For from preposition node when i random walk.Work as dtjWhen=0, label t i.e. label j, π is referred toi t jIt refers to walking from label j
Travel back to the probability of label j after to label i from label i again.
According to definitionIt after this normalization, carries out m n and walks random walk, obtain the path that m step-length is n,
The sliding that window size is s is carried out in these paths, between the label of co-occurrence in window plus a line, weight add 1.It is all sliding
After the completion of dynamic, weight is the sum of label co-occurrence window, obtains new label network G according to thisC。
Step 4: label network G is constructed based on the relevant text information of labelT。
This step utilizes the associated text of label, i.e., constructs label network using the content characteristic of network node itself.This
The label and text vocabulary of invention have cooccurrence relation, and co-occurrence number is more, illustrate that vocabulary can more embody meaning represented by label.
There are the lexical set W of cooccurrence relation with label i for definitioni, use wijIndicate the number of label i and vocabulary j co-occurrence.
For example, text collection includes three texts, it is respectively as follows:
" South Africa # gold brick is met #[current political news eye Shu gold brick and is met: the city " clapping gold brick " of gold] "
" South Africa # gold brick meets # economic community, and science and technology is not broken up the family "
" South Africa # gold brick meets # gold brick country, and friendship often exists;With wind resistance rain, win-win is shared.Wish our motherland more
It is powerful, it is happy and peace."
Wherein, word " gold brick " the weight w of label " South Africa # gold brick meets # " is 3.
After text representation, using inverse document frequency IDF (Inverse Document Frequency) to w into
Row weighting.Word tiIdf value calculating method it is as follows:
Wherein, | D | it is the number of files in set, | { j:ti∈dj| to include word tiNumber of files, calculate w again laterijWith
idfjProduct, obtain the text representation vector of label.
Calculating label text two-by-two indicates the cosine similarity of vector, similar step 2, definitionFor can be truncated 80% it is low
The interceptive value on frequency side.Remove cosine similarity to be less thanSide.The weight for retaining side is labeled as the value of cosine similarity.
Step 5: the label network G that step 3, step 4 are obtainedC, label network GTIt is normalized, passes through random walk
Strategy and term vector learning method study label semantic vector indicate.
The G that step 3 obtainsCEmbody the label association based on network structure, the G that step 4 obtainsTIt embodies based on semanteme
The label of correlation is associated with.This step will merge GCAnd GTTwo networks learn the semantic table of label on fused network
Show.
Since the proportion range of two networks is different, the weight of two networks is uniformly arrived between [0,1] first.Use line
Property function is normalized, and function is as follows:
Equal proportion scaling, x are carried out to initial datanormFor the value after x normalization, xminAnd xmaxRespectively side in network
The minimum value and maximum value of weight.Network after note normalization is respectively GC-normAnd GT-norm.Then carried out on two networks
Random walk obtains neighbour's sequence of each node, and word2vec study is carried out in the sequence, uses SkipGram method
The vector for updating label node indicates.The calculation process of label semantic vector Φ is as follows:
Wherein, Shuffle () reordering for label node, to avoid preference caused by operation successively.
RandomWalk(GC-norm, j, t) and it completes in figure GC-normOn from j, random walk t walk, obtain a length be t label
Sequence node Wj, RandomWalk (GT-norm, j, t) and similar.SkipGram(Φ,Wj, d) and it is term vector learning method, according to section
Point sequence WjUpdate the label semantic vector Φ that dimension is d.
It is emphasized that embodiment of the present invention be it is illustrative, without being restrictive, therefore packet of the present invention
Include and be not limited to embodiment described in specific embodiment, it is all by those skilled in the art according to the technique and scheme of the present invention
The other embodiments obtained, also belong to the scope of protection of the invention.
Claims (6)
1. a kind of label semanteme learning method measured based on network structure and semantic dependency, it is characterised in that: including following
Step:
Step 1 is based on user behavior fact init Tag network, obtains truth labels network G;
Step 2 constructs the label network G after specification according to truth labels network GR;
Step 3, in label network GRThe random walk strategy of upper application enhancements constructs the label network based on random walk strategy
GC;
Step 4 constructs label network G based on the relevant text information of labelT;
Step 5, to label network GC, label network GTIt is normalized, passes through random walk strategy and term vector study side
Calligraphy learning label semantic vector indicates.
2. a kind of label semanteme learning method measured based on network structure and semantic dependency according to claim 1,
It is characterized by: the truth labels network G are as follows:
The label network G={ V, E } based on co-occurrence inside text is defined, V is all labels in entire text collection D;If appointed
Meaning label i and label j are appeared in simultaneously in a text d, then have a line between them, be denoted as eij;Define side in the network
Weight gijAre as follows:
Wherein, Di,jFor the text collection for simultaneously including label i and j, hdFor the tag set of document d.
3. a kind of label semanteme learning method measured based on network structure and semantic dependency according to claim 1,
It is characterized by: the processing method of the step 2 are as follows:
Firstly, considering the associated random noise of semantic node: subtracting branch and remove the stronger weak rigidity of randomness, to the effective edge of network
It is constrained with noise reduction;Incidence matrix after enabling beta pruning is expressed as T, then wherein element tijAfter as label i and label j beta pruning
Relating value is expressed as follows:
Wherein, δ be can be truncated 20% low frequency side phase threshold, gijFor the weight on side in network;
Then, consider that label node degree of divergence is different, for the side in any one network, according to associated by the endpoint of two, side
The weight of the different adjustment web tab incidence relations of endpoint number, enhancing label theme association:
t′ij=tij*log(N/N(i))*log(N/N(j))。
Wherein, t 'ijFor GRThe weight on side in network, number of the N by network comprising node, N (i), N (j) are label i, j in net
Out-degree on network figure, logN/N (i) are current label and the related Probability p (i) of nodes=N (i)/N inverse pair
Numerical expression.
4. a kind of label semanteme learning method measured based on network structure and semantic dependency according to claim 1,
It is characterized by: the improved random walk strategy of step 3 be the complex network structures of high noisy are sampled to it is a plurality of linear
Sequence obtains the microcosmic description information of network part with breadth-first search method, obtains network with Depth Priority Searching
Global macroscopic information slided on linear order using window according to the linear order of sampling and obtain new label association and close
System, to obtain new side right weight.
5. a kind of label semanteme learning method measured based on network structure and semantic dependency according to claim 1,
It is characterized by: the processing method of the step 4 are as follows:
Firstly, there are the lexical set W of cooccurrence relation with label i for definitioni, use wijIndicate the number of label i and vocabulary j co-occurrence;
Then, w is weighted using inverse document frequency idf, word tiIdf value calculating method it is as follows:
Wherein, | D | it is the number of files in set, | { j:ti∈dj| to include word tiNumber of files;
Secondly, calculating w againijWith idfjProduct, obtain the text representation vector of label;
Finally, calculating label text two-by-two indicates the cosine similarity of vector, definitionFor can be truncated 80% low frequency side truncation
Threshold value is removed cosine similarity and is less thanSide, retain side weight be labeled as cosine similarity value.
6. a kind of label semanteme learning method measured based on network structure and semantic dependency according to claim 1,
It is characterized by: the step 5 in network random walk process, introduces figure sampling preference parameter, in label network GTAnd mark
Sign network GCBetween switch, be associated with benefiting from network structure with two kinds of text information;In semantic vector renewal process, utilize
It is regarded sentence by the linear order that network random walk obtains, semantic based on left and right context study label.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810914904.3A CN109189936B (en) | 2018-08-13 | 2018-08-13 | Label semantic learning method based on network structure and semantic correlation measurement |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810914904.3A CN109189936B (en) | 2018-08-13 | 2018-08-13 | Label semantic learning method based on network structure and semantic correlation measurement |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109189936A true CN109189936A (en) | 2019-01-11 |
CN109189936B CN109189936B (en) | 2021-07-27 |
Family
ID=64921583
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810914904.3A Active CN109189936B (en) | 2018-08-13 | 2018-08-13 | Label semantic learning method based on network structure and semantic correlation measurement |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109189936B (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109948066A (en) * | 2019-04-16 | 2019-06-28 | 杭州电子科技大学 | A kind of point of interest recommended method based on Heterogeneous Information network |
CN110413865A (en) * | 2019-08-02 | 2019-11-05 | 知者信息技术服务成都有限公司 | Semantic expressiveness model and its method based on alternating binary coding device characterization model |
CN110851491A (en) * | 2019-10-17 | 2020-02-28 | 天津大学 | Network link prediction method based on multiple semantic influences of multiple neighbor nodes |
CN110889001A (en) * | 2019-11-25 | 2020-03-17 | 浙江财经大学 | Big image sampling visualization method based on image representation learning |
CN111460118A (en) * | 2020-03-26 | 2020-07-28 | 聚好看科技股份有限公司 | Artificial intelligence conflict semantic recognition method and device |
CN111611498A (en) * | 2020-04-26 | 2020-09-01 | 北京科技大学 | Network representation learning method and system based on domain internal semantics |
CN111723301A (en) * | 2020-06-01 | 2020-09-29 | 山西大学 | Attention relation identification and labeling method based on hierarchical theme preference semantic matrix |
CN112131569A (en) * | 2020-09-15 | 2020-12-25 | 上海交通大学 | Risk user prediction method based on graph network random walk |
CN112182511A (en) * | 2020-11-27 | 2021-01-05 | 中国人民解放军国防科技大学 | Complex semantic enhanced heterogeneous information network representation learning method and device |
CN112529621A (en) * | 2020-12-10 | 2021-03-19 | 中山大学 | Advertisement audience basic attribute estimation method based on heterogeneous graph embedding technology |
CN112580916A (en) * | 2019-09-30 | 2021-03-30 | 深圳无域科技技术有限公司 | Data evaluation method and device, computer equipment and storage medium |
CN112688813A (en) * | 2020-12-24 | 2021-04-20 | 中国人民解放军战略支援部队信息工程大学 | Routing node importance ordering method and system based on routing characteristics |
CN112989209A (en) * | 2021-05-10 | 2021-06-18 | 腾讯科技(深圳)有限公司 | Content recommendation method, device and storage medium |
WO2021190091A1 (en) * | 2020-03-26 | 2021-09-30 | 深圳壹账通智能科技有限公司 | Knowledge map construction method and device based on knowledge node belonging degree |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105956197A (en) * | 2016-06-15 | 2016-09-21 | 杭州量知数据科技有限公司 | Social media graph representation model-based social risk event extraction method |
CN106055604A (en) * | 2016-05-25 | 2016-10-26 | 南京大学 | Short text topic model mining method based on word network to extend characteristics |
CN106874931A (en) * | 2016-12-30 | 2017-06-20 | 东软集团股份有限公司 | User portrait grouping method and device |
CN107122455A (en) * | 2017-04-26 | 2017-09-01 | 中国人民解放军国防科学技术大学 | A kind of network user's enhancing method for expressing based on microblogging |
CN107291803A (en) * | 2017-05-15 | 2017-10-24 | 广东工业大学 | A kind of network representation method for merging polymorphic type information |
WO2018118546A1 (en) * | 2016-12-21 | 2018-06-28 | Microsoft Technology Licensing, Llc | Systems and methods for an emotionally intelligent chat bot |
CN108255809A (en) * | 2018-01-10 | 2018-07-06 | 北京海存志合科技股份有限公司 | Consider the method for calculating the theme corresponding to document of Words similarity |
-
2018
- 2018-08-13 CN CN201810914904.3A patent/CN109189936B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106055604A (en) * | 2016-05-25 | 2016-10-26 | 南京大学 | Short text topic model mining method based on word network to extend characteristics |
CN105956197A (en) * | 2016-06-15 | 2016-09-21 | 杭州量知数据科技有限公司 | Social media graph representation model-based social risk event extraction method |
WO2018118546A1 (en) * | 2016-12-21 | 2018-06-28 | Microsoft Technology Licensing, Llc | Systems and methods for an emotionally intelligent chat bot |
CN106874931A (en) * | 2016-12-30 | 2017-06-20 | 东软集团股份有限公司 | User portrait grouping method and device |
CN107122455A (en) * | 2017-04-26 | 2017-09-01 | 中国人民解放军国防科学技术大学 | A kind of network user's enhancing method for expressing based on microblogging |
CN107291803A (en) * | 2017-05-15 | 2017-10-24 | 广东工业大学 | A kind of network representation method for merging polymorphic type information |
CN108255809A (en) * | 2018-01-10 | 2018-07-06 | 北京海存志合科技股份有限公司 | Consider the method for calculating the theme corresponding to document of Words similarity |
Non-Patent Citations (1)
Title |
---|
李志宇,梁循,周小平,张海燕,马跃峰: "一种大规模网络中基于节点结构特征映射的链接预测方法", 《计算机学报》 * |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109948066A (en) * | 2019-04-16 | 2019-06-28 | 杭州电子科技大学 | A kind of point of interest recommended method based on Heterogeneous Information network |
CN110413865A (en) * | 2019-08-02 | 2019-11-05 | 知者信息技术服务成都有限公司 | Semantic expressiveness model and its method based on alternating binary coding device characterization model |
CN112580916A (en) * | 2019-09-30 | 2021-03-30 | 深圳无域科技技术有限公司 | Data evaluation method and device, computer equipment and storage medium |
CN112580916B (en) * | 2019-09-30 | 2024-05-28 | 深圳无域科技技术有限公司 | Data evaluation method, device, computer equipment and storage medium |
CN110851491A (en) * | 2019-10-17 | 2020-02-28 | 天津大学 | Network link prediction method based on multiple semantic influences of multiple neighbor nodes |
CN110851491B (en) * | 2019-10-17 | 2023-06-30 | 天津大学 | Network link prediction method based on multiple semantic influence of multiple neighbor nodes |
CN110889001A (en) * | 2019-11-25 | 2020-03-17 | 浙江财经大学 | Big image sampling visualization method based on image representation learning |
CN110889001B (en) * | 2019-11-25 | 2021-11-05 | 浙江财经大学 | Big image sampling visualization method based on image representation learning |
CN111460118A (en) * | 2020-03-26 | 2020-07-28 | 聚好看科技股份有限公司 | Artificial intelligence conflict semantic recognition method and device |
CN111460118B (en) * | 2020-03-26 | 2023-10-20 | 聚好看科技股份有限公司 | Artificial intelligence conflict semantic recognition method and device |
WO2021190091A1 (en) * | 2020-03-26 | 2021-09-30 | 深圳壹账通智能科技有限公司 | Knowledge map construction method and device based on knowledge node belonging degree |
CN111611498A (en) * | 2020-04-26 | 2020-09-01 | 北京科技大学 | Network representation learning method and system based on domain internal semantics |
CN111611498B (en) * | 2020-04-26 | 2024-01-02 | 北京科技大学 | Network representation learning method and system based on field internal semantics |
CN111723301B (en) * | 2020-06-01 | 2022-05-27 | 山西大学 | Attention relation identification and labeling method based on hierarchical theme preference semantic matrix |
CN111723301A (en) * | 2020-06-01 | 2020-09-29 | 山西大学 | Attention relation identification and labeling method based on hierarchical theme preference semantic matrix |
CN112131569A (en) * | 2020-09-15 | 2020-12-25 | 上海交通大学 | Risk user prediction method based on graph network random walk |
CN112131569B (en) * | 2020-09-15 | 2024-01-05 | 上海交通大学 | Risk user prediction method based on graph network random walk |
CN112182511B (en) * | 2020-11-27 | 2021-02-19 | 中国人民解放军国防科技大学 | Complex semantic enhanced heterogeneous information network representation learning method and device |
CN112182511A (en) * | 2020-11-27 | 2021-01-05 | 中国人民解放军国防科技大学 | Complex semantic enhanced heterogeneous information network representation learning method and device |
CN112529621A (en) * | 2020-12-10 | 2021-03-19 | 中山大学 | Advertisement audience basic attribute estimation method based on heterogeneous graph embedding technology |
CN112688813A (en) * | 2020-12-24 | 2021-04-20 | 中国人民解放军战略支援部队信息工程大学 | Routing node importance ordering method and system based on routing characteristics |
CN112989209A (en) * | 2021-05-10 | 2021-06-18 | 腾讯科技(深圳)有限公司 | Content recommendation method, device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN109189936B (en) | 2021-07-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109189936A (en) | A kind of label semanteme learning method measured based on network structure and semantic dependency | |
CN107133213B (en) | Method and system for automatically extracting text abstract based on algorithm | |
CN111581983A (en) | Method for predicting social concern hotspots in network public opinion events based on group analysis | |
Li et al. | Image sentiment prediction based on textual descriptions with adjective noun pairs | |
Dutta et al. | A graph based clustering technique for tweet summarization | |
CN103544242A (en) | Microblog-oriented emotion entity searching system | |
JP2021508866A (en) | Promote area- and client-specific application program interface recommendations | |
JP2021508391A (en) | Promote area- and client-specific application program interface recommendations | |
CN112559747A (en) | Event classification processing method and device, electronic equipment and storage medium | |
CN116975271A (en) | Text relevance determining method, device, computer equipment and storage medium | |
Fersini et al. | A probabilistic relational approach for web document clustering | |
Yong et al. | A new emotion analysis fusion and complementary model based on online food reviews | |
Li et al. | A coarse-to-fine collective entity linking method for heterogeneous information networks | |
CN116467461A (en) | Data processing method, device, equipment and medium applied to power distribution network | |
Xiao et al. | Research on multimodal emotion analysis algorithm based on deep learning | |
Wang et al. | A tutorial and survey on fault knowledge graph | |
Ravindranath et al. | Inferring structure and meaning of semi-structured documents by using a gibbs sampling based approach | |
CN113297854A (en) | Method, device and equipment for mapping text to knowledge graph entity and storage medium | |
Yun et al. | Combining vector space features and convolution neural network for text sentiment analysis | |
Zhou et al. | Spatiotemporal data cleaning and knowledge fusion | |
Lee et al. | Exploiting online social data in ontology learning for event tracking and emergency response | |
Fan et al. | Topic modeling methods for short texts: A survey | |
Tian et al. | Label importance ranking with entropy variation complex networks for structured video captioning. | |
Ben-Lhachemi et al. | Hashtag recommendation using word sequences’ embeddings | |
Jiang et al. | Bidirectional LSTM-CRF models for keyword extraction in Chinese sport news |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20240403 Address after: Room 1518B, Unit 2, 12th Floor, Huizhi Building, No. 9 Xueqing Road, Haidian District, Beijing, 100080 Patentee after: Beijing contention Technology Co.,Ltd. Country or region after: China Address before: No.9, 13th Street, economic and Technological Development Zone, Binhai New Area, Tianjin Patentee before: TIANJIN University OF SCIENCE AND TECHNOLOGY Country or region before: China |
|
TR01 | Transfer of patent right |