• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    Similarity measurement method of high-dimensional data based on normalized net lattice subspace①

    2017-06-27 08:09:22LiWenfa李文法WangGongmingLiKeHuangSu
    High Technology Letters 2017年2期
    關鍵詞:文法

    Li Wenfa (李文法), Wang Gongming, Li Ke, Huang Su

    (*Beijing Key Laboratory of Information Service Engineering,Beijing Union University, Beijing 100101, P.R.China) (**National Laboratory of Biomacromolecules, Institute of Biophysics, Chinese Academy of Sciences, Beijing 100101, P.R.China)

    Similarity measurement method of high-dimensional data based on normalized net lattice subspace①

    Li Wenfa (李文法)②*, Wang Gongming**, Li Ke*, Huang Su*

    (*Beijing Key Laboratory of Information Service Engineering,Beijing Union University, Beijing 100101, P.R.China) (**National Laboratory of Biomacromolecules, Institute of Biophysics, Chinese Academy of Sciences, Beijing 100101, P.R.China)

    The performance of conventional similarity measurement methods is affected seriously by the curse of dimensionality of high-dimensional data. The reason is that data difference between sparse and noisy dimensionalities occupies a large proportion of the similarity, leading to the dissimilarities between any results. A similarity measurement method of high-dimensional data based on normalized net lattice subspace is proposed. The data range of each dimension is divided into several intervals, and the components in different dimensions are mapped onto the corresponding interval. Only the component in the same or adjacent interval is used to calculate the similarity. To validate this method, three data types are used, and seven common similarity measurement methods are compared. The experimental result indicates that the relative difference of the method is increasing with the dimensionality and is approximately two or three orders of magnitude higher than the conventional method. In addition, the similarity range of this method in different dimensions is [0, 1], which is fit for similarity analysis after dimensionality reduction.

    high-dimensional data, the curse of dimensionality, similarity, normalization, subspace, NPsim

    0 Introduction

    A similarity measurement can determine similarity degree between two data, or distance between two points, which is the basis of data-mining methods such as clustering, classification, nearest neighbor search, and association analysis. Conventional similarity measurement methods include Euclidean distance, Jaccard coefficient[1], and Pearson coefficient[2]. These methods can satisfy the similarity measurement requirement in low-dimensional space (less than 16)[3]. However, with the increasing spatial dimensionalities, the distance between a query point and its nearest neighbor point tends to be equal to the distance from the query point to its farthest neighbor point. When the distance between any two points is equal everywhere, the similarity is pointless; this is called the isometrics in high-dimensional space[4]. The root cause of this phenomenon is the curse of dimensionality that is derived from properties of sparsity and empty space in a high-dimensional space. Thus, the performances of many similarity measurements are positively affected in the low-dimensional space, yet decrease sharply in the high-dimensional space.

    In recent years, a series of methods have been proposed for similarity measurement of high-dimensional data; these includeHsim(X,Y)[5],HDsim(X,Y)[6],Gsim(X,Y)[7],Close(X,Y)[8],andEsim(X,Y)[9].However,thesemethodsignoretherelativedifferenceinproperty,noisedistribution,weight,andareonlyvalidforcertaindatatypes[10].ThePsim(X,Y)functionconsiderstheabove-mentionedfactors[10]andisapplicabletoavarietyofdatatypes;however,itisunabletocomparesimilarityunderdifferentdimensionsbecauseitsrangedependsonthespatialdimensionality.

    Tosolvethisproblem,asimilaritymeasurementmethodofhigh-dimensionaldatabasedonnormalizednetlatticesubspaceisproposed.Thesimilarityrangeisnolongerlimitedbythespatialdimensionality.

    1 Relatedwork

    1.1 Curse of dimensionality

    This is a ubiquitous phenomenon in the application field of high-dimensional data, and occurs because of the sparsity and empty space in high-dimensional space.

    1.1.1 Sparsity

    There is ad-dimensional data setDinahypercubeunitΨ=[0,1]d,anddataelementsaredistributeduniformly.Theprobabilityofapointfallingintoonehypercubewithlengthsissd,whichdecreaseswiththeincreaseofsbecauses<1.Thatis,itisverylikelythatthereisnopointinalargerange[11].Forexample,approximatelyonly0.59%dataexistsinahypercubewithlength0.95whendimensions=100.

    1.1.2Emptyspacephenomenon

    Anormaldistributiondatasetcanbeexpressedbyitscenterpointandstandarddeviation.ThedistancesbetweenthedatapointsandthecenterpointobeytheGaussdistribution;however,theirrelativeorientationcanbeselectedrandomly.Inaddition,thenumberofpossibledirectionsrelativetoacenterpointisincreasedexponentiallyandthedistancebetweenthemisincreasedwiththeincreaseofdimensionality.Fromtheviewpointofthedensityofadataset,amaximumvalueexistsatthecenterpoint,althoughtheremaynotbeapointclosetothecenterpoint.Thisphenomenonofahigh-dimensionalspaceiscalled“emptyspace.”

    1.1.3 Isometry

    The volume of unit sphere in ad-dimensional space is described as follows.

    (1)

    V(d)decreasesgraduallywiththeincreaseofdimensionalityd.Fig.1showsthatV(d)→0ifd>16.

    Fig.1 Variation trend of unit sphere volume with

    With the increase in dimensionality, the number of corners increases and the volume of unit sphere gradually decreases because the volume of the unit hyperspace does not change. Thus, most of the data will be distributed in the hyperspace corner. This phenomenon is shown in Fig.2 from left to right; the three subgraphs show the distributions of super-space data with dimensionality of 2, 3, and 8, respectively. In eight-dimensional space, 98% data is distributed in 2^8 = 256 corners. Moreover, the maximum and minimum Euclidean distances between the data and center point are both the same. When the dimensionality tends to infinity, the difference between the maximum and minimum Euclidean distance of the sample points to the center point tends toward 0.

    Fig.2 Data distribution in different dimensions

    Therefore, with the increase in dimensionality, the Euclidean distance between any data tends to remain the same, and no longer has the measurement function. The corresponding data-mining methods, such as clustering, classification, and nearest neighbor, would lose their effect.

    1.2 Conventional high-dimensional data similarity measurement methods

    In recent years, a similarity measurement problem in high-dimensional space has been studied to a certain extent but the research is insufficient. TheHsim(X,Y)functionwasproposedbyYang[5],whichisbetterthantheconventionalmethodbutneglectstherelativedifferenceandnoisedistribution.Inaddition,itisnotsuitableformeasuringthesimilarityofcategorical-attributedata.Next,theGsim(X,Y)function[7]wasproposedaccordingtotherelativedifferenceofpropertiesindifferentdimensions;however,itignorestheweightdiscrepancy.Zhaointroducedthepiecewisefunctionδ(X,Y)intoHsim(X,Y)andproposedtheHsimc(X,Y)function[12],whichcomprisesafunctionofmeasuringcategorical-attributedata.However,similaritybetweenapairofpointswhosecomponentsarecomplementaryineverydimensionisinconsistentwiththeactualresult.Thepiecewisefunctionδ(X,Y)offunctionXiemodifiedHsimc(X,Y)andproposedtheHDsim(X,Y)function[6],whichcansolvetheproblemderivedfromacomplementarypropertyineverydimension.However,theattributedifferenceandnoisedistributionproblemareneglected.TheClose(X,Y)function[8]basedonthemonotonousdecreaseofe-xcanovercometheinfluencefromcomponentsinsomedimensionswithlargevariancebutdoesnotconsidertherelativedifference,whichwouldbeaffectedbynoise.TheEsim(X,Y)[9]functionwasproposedbyimprovingHsim(X,Y)andClose(X,Y)functionsandcombiningtheinfluenceofpropertyonsimilarity.Ineverydimension,theEsim(X,Y)componentshowsapositivecorrelationtothevalueinthisdimension.Alldimensionsaredividedintotwoparts:normalandnoisydimensions.Inanoisydimension,thenoiseoccupiesmajority.Whennoiseissimilarorlargerthantheoneinanormaldimension,thismethodisinvalid.Thesecondarymeasurementmethod[13]isusedtocalculatethesimilaritybyvirtueofpropertydistribution,spacedistance,etc.;however,itneglectsthenoisedistributionandweight.Inaddition,itistime-consuming.TheconceptofnearestneighborprojectionwasproposedbyHinneburg[14],whichwascombinedwithdimensionalityreductiontosolvetheprobleminhigh-dimensionalspace.However,thismethodcomplicatesthedeterminationofasuitablequalitycriterionfunction.Thus,anextensiontheorywasintroducedintosimilaritycalculation[15],inwhich,thehigh-dimensionaldataisexpressedasanorderedthreetuplebyvirtueofmatterelement,andthedeviation(theintervallengthofattributevalueineverydimension)isaddedintofunctionA. However, this method is too complicated, and the result validation of the high-dimensional data was not described in the corresponding paper. Yi[10]determined that in a high-dimensional space, the difference in a noisy dimension is larger than in a sparse dimension, no matter how similar the data is. This difference occupies a large amount of the similarity calculation, leading to the calculation results of any objects being similar. Therefore, thePsim(X,Y)function[12]wasproposedtoeliminatethenoisyinfluencebyanalyzingthedifferenceamongalldimensions.Theexperimentalresultsindicatethatthismethodissuitableforavarietyofdata.However,itsrangeis[0,n],wherenisthedimensionality.Thus,thesimilaritiesindifferentdimensionscannotbecompared.

    2 Similaritymeasurementmethodbasedonnormalizednetlatticesubspace

    2.1 Sparse and noisy dimensions

    With increasing dimensionality, the similarities based on theLdnormbetweenanydatabecomethesame.TherootcauseisthattheLdnormdependsonthedimensiontoomuchwhichhaslargelydifferentcomponents.Inotherwords,whencalculatingsimilaritybetweenXandY,thelargerthevalueofXi-Yionthei-th dimension, the greater the contribution of thei-th dimension toXandY.AlthoughbothXandYareverysimilarinotherdimensionsexceptthei-th dimension, the overall similarity ofXandYisverysmall.Thisi-th dimension is called sparse or noisy dimension.

    Owing to the existence of sparsity and noise in the high-dimensional space, no matter how similar the two records are there will always be a different dimension. The difference in these dimensions occupies a large proportion of the whole similarity, leading to any record in the high-dimensional space being dissimilar[16].

    To solve this problem, the data range in every dimension can be divided into several intervals, and the components can be mapped onto corresponding intervals. When calculating the similarity between two points, only the dimensions that fall into the same interval are used. The other dimensions are regarded as sparse or noisy dimensions, and are not included in the calculation.

    2.2 Meshing of high-dimensional data space

    Let the dimension of dataset bed,andthenumberofdataobjectbeM.Then,everydataobjectisexpressedasxk(1≤k≤M).Inaddition,everydimensionisdividedinton=[θd]continuousintervals,andθistherealnumberbetween0and1.Thus,thenumberofpointsineveryintervalisG=[M/n].

    Inthei-th dimension, all components are sorted in an ascending order. Thek-th sorted value isVal[k](1≤k≤M).Rijisthej-th interval in thei-th dimension, whose lower and upper bounds areLRijandURij,respectively.ItcanbeseenthatLRij=Val[(j-1)G+1]andURij=Val[jG].

    (2)

    (3)

    Forxkandyl,thesetofdimensionsinwhichcomponentsfallintothesameintervalis

    (4)

    Ifthei-th components ofxkandylfallintotheadjacentintervals,andthedistancebetweenthemislessthantheaveragelengthofthetwoadjacentintervals,thetwopointsareregardedasclosetoeachother,andincludedinthesimilaritycalculation.Thesetofthesedimensionsisshownas

    (5)

    ThesetofdimensionsincludedinthesimilaritycalculationistheunionofS1andS2:

    S=S1US2

    (6)

    2.3Similaritymeasurement

    ThePsim(X,Y)functionproposedbyYiissuitableforavarietyofdatatypes[10];however,itsrangeisdependentonthespatialdimensionality,andthusthecomparisonofsimilarityindifferentdimensionsisnotpossible.Underthecircumstanceofmaintainingeffects,Psim(X,Y)iscorrectedas

    (7)

    whereXandYareanytwopointsinthed-dimensional space, andXjandYjarecomponentsinthei-th dimension. Moreover,δ(Xj,Yj)isthediscriminantfunction.IfXjandYjareinthesameinterval[LRj,URj],δ(Xj,Yj)=1,otherwiseδ(Xj,Yj)=0.E(X,Y)representsthenumberofintervalsinwhichcomponentsofXandYareallthesame.TherangeofNPsim(X,Y)isobservedtobein[0, 1].TheaboveistheoutlineofNPsim,andthedetailedintroductioncanbefoundinRef.[10].

    3 Experiment

    Tovalidatethismethod,threedatatypeswithdifferentdistributionsweregeneratedthroughMatlab.Next,thesimilaritiesindifferentdimensionswerecalculatedusingtheproposedmethod,andwerecomparedwiththeresultobtainedfromcalculatingManhattandistance,Euclideandistance,Hsim(X,Y),Gsim(X,Y),Close(X,Y),Esim(X,Y),andPsim(X,Y).

    3.1Datadescription

    Thefollowingthreedatatypeswereusedintheexperiment[10].

    (1)Independentandidenticallydistributed(IID):Here,allvariablesobeythesamedatadistributionfunctionbutareindependentofeachother.TheIIDdataZisgeneratedbyZ=(Z1,…,ZM),andZifollowsthedistributionofZi~F(0,1).

    (3)Dependentandidenticallydistributed(DID):Allvariablesobeythesamedatadistributionbutarenotindependent.Inaddition,twodimensionsareindependentofeachothercalled“freedimensions”;theotherdimensionsarerelatedtothem.TheDIDdataZisgeneratedasfollows.First,twod×1randomvariablesAandBobeyingthedistributionofF(0,1)aregenerated.Second,two1×MrandomvariablesUandVobeyingthedistributionofF(-1, 1)areproduced.Third,Z1(2≤i≤M)isgeneratedthroughZi=A×Ui+B×Vi.Atlast,theDIDdataZisproducedasZ=(Z1,…,ZM).

    3.2Relativedifference

    Tovalidatethismethod,IID,RID,andDIDdataaregeneratedusinganormrnd()functionofMatlab[10].Thedimensionofeverydatatypeisasfollows: 10, 60, 110, 160, 210, 260, 310, 360,and410.Thenumberofdataineverydimensionis1000.Inaddition,therelativedifferencebetweenthefarthestandnearestneighborsiscalculatedasfollows[17]:

    (8)

    whereDmaxn,Dminn,andDavgnaremaximal,minimal,andaveragesimilaritiesinthed-dimensional space, respectively. The relative difference results are shown in Figs 3~5.

    According to the characteristics of the results, similarity measurement methods are divided into two types: the first includes Manhattan distance, Euclidean distance,Hsim(X,Y),Gsim(X,Y),Close(X,Y),andEsim(X,Y);andtheothersincludePsim(X,Y)andNPsim(X,Y).Therelativedifferenceofthesecondtypeofmethodsistwoorthreemagnitudeslargerthanthatofthefirsttypeofmethods.Therefore,theperformanceadvantageofthesecondmethodtypeisobvious.

    TherelativedifferenceofPsim(X,Y)andNPsim(X,Y)hasnodifferentiationdegree.Thus,thestatisticalanalysisneedstobestudiedfurther.

    Fig.3 Relative difference of various similarity measurement methods for IID data

    Fig.4 Relative difference of various similarity measurement methods for RID data

    3.3 Statistical analysis

    To compare the effect ofPsim(X,Y)andNPsim(X,Y),themaximum,minimum,andaverageofDIDdataindifferentdimensionsarecalculated,asshowninFig.6.TheexperimentalresultsindicatethatthesimilarityrangeofPsim(X,Y)increaseswiththedimension.Thus,thefunctionisnotsuitableforthesimilaritycomparisonindifferentdimensions.However,theproblemdoesnotexistinNPsim(X,Y).Table1liststhenumbersofPsim(X,Y)whosevalueisgreaterthan1indifferentdimensions.Thenumberof

    Fig.5 Relative difference of various similarity measurement methods for DID data

    Fig.6 Statistical value of various similarity measurement methods for DID data

    Dimension1060110160210Number1686041203731132481045284672Dimension260310360410260Number9842963024720155885198429

    Psim(X,Y)ineverydimensionis1000×1000=1,000,000.Inaddition,the5%~17%resultismorethan1,andthusthecomparisonofsimilarityindifferentdimensionsisnotpossible.Therefore,NPsim(X,Y)cansatisfytherequirementofsimilaritycomparisonindifferentdimensions.

    4 Conclusion

    Thesimilaritymeasurementisthebasisofdata-miningalgorithms,suchasclustering,classification,andnearestneighbor.However,owingtothecurseofdimensionality,themeasurementalwaysfailsinhigh-dimensionalspace.Asimilaritymeasurementmethodofhigh-dimensionaldatabasedonanormalizednetlatticesubspaceisproposed.Inthismethod,datarangeofeachdimensionisdividedintoseveralintervals,andthecomponentsaremappedontothecorrespondingintervals.Duringsimilaritycalculation,onlythecomponentinthesameoradjacentintervalisused.Thismethodcanavoidthesimilarityeffectthatgeneratedfromthesparseornoisydimension.Tovalidatetheproposedalgorithm,threetypesofdistributiondataareusedintheexperiment,andanothersevenmethodtypesarecompared.Theexperimentalresultsshowthattheproposedmethodissuitableforsimilaritymeasurementinhigh-dimensiondata.

    Inthefuture,theweightcalculationindifferentdimensions,andtheautomaticupdatingstrategyofaspatialgridwillbestudied.Inaddition,theproposedmethodwillapplyarelateddata-miningalgorithm,suchasclusteringorcorrelationanalysis.

    [ 1] Tan P N, Michael S, Vipin K. Introduction to Data Mining. Boston: Addison-Wesley Publishing Company, 2005. 25-36

    [ 2] Chen J N. The Research and Application of Key Technologies in Knowledge Discovery of High-dimensional Clustering. Beijing: Publishing House of Electronics Industry, 2011. 120-128(In Chinese)

    [ 3] Aggarwal C C. Re-designing distance functions and distance based applications for high dimensional data.ACMSIGMODRecord, 2001, 33(1):117-128

    [ 4] Warren B P. Approximate Dynamic Programming: Solving the Curses of Dimensionality (2nd Edition). Hoboken, New Jersey: John Wiley & Sons Press, 2011. 124-161

    [ 5] Yang F Z, Zhu Y Y. An efficient method for similarity search on quantitative transaction data.JournalofComputerResearchandDevelopment, 2004, 41(2):361-368

    [ 6] Xie M X, Guo J Z, Zhang H B, et al. Research on the similarity measurement of high dimensional data.ComputerEngineeringandScience, 2010, 32(5):92-96(In Chinese)

    [ 7] Huang S D, Chen Q M. On clustering algorithm of high dimensional data based on similarity measurement.ComputerApplicationsandSoftware, 2009, 26(9):102-105(In Chinese)

    [ 8] Shao C S, Lou W, Yan L M. Optimization of algorithm of similarity measurement in high-dimensional data.ComputerTechnologyandDevelopment, 2011, 21(2):1-4(In Chinese)

    [ 9] Wang X Y, Zhang H Y, Shen L Z, et al. Re-search on high dimensional clustering algorithm based on similarity measurement.ComputerTechnologyandDevelopment, 2013, 23(5):30-33(In Chinese)

    [10] Yi L H. Research on clustering algorithm for high dimensional data:[Ph.D dissertation]. Qinhuangdao: Institute of Information Science and Engineering, Yanshan University, 2011. 28-30(In Chinese)

    [11] Ericson K, Pallickara S. On the performance of high dimensional data clustering and classification algorithms.FutureGenerationComputerSystems, 2013, 29(4):1024-1034

    [12] Zhao H. Study on Some Issues of Data Clustering in Data Mining:[Ph.D dissertation]. Xi’an: School of Electronic Engineering, Xidian University, 2005. 35-42(In Chinese)

    [13] Jia X Y. A high dimensional data clustering algorithm based on twice similarity.JournalofComputerApplications, 2005, 25(B12):176-177

    [14] Alexander H, Charu A C, Keim D A. What is the nearest neighbor in high dimensional spaces? In: Proceedings of the 26th International Conference on Very Large Data Bases, Cairo, Egypt, 2000. 506-515

    [15] Yuan R P, Shi M R. Research on the similarity of high dimensional big data based on extenics.OperationsResearchandManagementScience, 2015, 24(5):184-188

    [16] Kriegel H P, Kr?ger P, Zimek A. Clustering high-dimensional data: a survey on subspace clustering, pattern-based clustering, and correlation clustering.ACMTransactionsonKnowledgeDiscoveryfromData, 2009, 3(1):1-58

    [17] Charu C, Aggarwal, Yu P S. The IGrid index: reversing the dimensionality curse for similarity indexing in high dimensional space. In: Proceedings of the 6th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Boston, USA, 2000. 119-129

    10.3772/j.issn.1006-6748.2017.02.009

    ①Supported by the National Natural Science Foundation of China (No. 61502475) and the Importation and Development of High-Caliber Talents Project of the Beijing Municipal Institutions (No. CIT & TCD201504039).

    ②To whom correspondence should be addressed. E-mail: liwenfa@buu.edu.cn

    on Dec. 10, 2016

    ?? born in 1974. He received his Ph.D. degree in Graduate University of Chinese Academy of Sciences in 2009. He also received his B.S. and M.S. degrees from PLA Information Engineering University in 1998 and 2003 respectively. His research interests include information security, data analysis and mining, etc.

    猜你喜歡
    文法
    從絕響到轉型:近現(xiàn)代“文法”概念與“文法學”
    關于1940 年尼瑪抄寫的《托忒文文法》手抄本
    中國石油大學勝利學院文法與經(jīng)濟管理學院簡介
    西夏文銅鏡的真言文法與四臂觀音像研究
    西夏學(2018年2期)2018-05-15 11:24:00
    LL(1)文法分析器的研究與分析
    科技風(2017年25期)2017-05-30 15:40:44
    A nearest neighbor search algorithm of high-dimensional data based on sequential NPsim matrix①
    25年呵護患病妻子不離不棄
    兵團工運(2016年9期)2016-11-09 05:46:13
    基于領域文法的微博輿情分析方法及其應用
    基于單向點格自動機的UPG文法識別并行算法
    文法有道,為作文注入音樂美
    學生天地(2016年26期)2016-06-15 20:29:39
    在线观看国产h片| 美女国产高潮福利片在线看| 亚洲欧美日韩另类电影网站| 中文字幕制服av| 久久久久久人人人人人| 老汉色∧v一级毛片| 曰老女人黄片| 热re99久久精品国产66热6| 亚洲三区欧美一区| 女性被躁到高潮视频| 国产精品99久久99久久久不卡 | 一级,二级,三级黄色视频| 丝袜美腿诱惑在线| 热99久久久久精品小说推荐| 在线观看三级黄色| 精品人妻熟女毛片av久久网站| 亚洲视频免费观看视频| 涩涩av久久男人的天堂| 99re6热这里在线精品视频| 久热久热在线精品观看| 啦啦啦视频在线资源免费观看| 黄色视频在线播放观看不卡| 亚洲精品国产av蜜桃| 在线观看一区二区三区激情| 人妻一区二区av| 肉色欧美久久久久久久蜜桃| 五月伊人婷婷丁香| 桃花免费在线播放| 啦啦啦在线免费观看视频4| 女性被躁到高潮视频| 亚洲伊人色综图| 国产伦理片在线播放av一区| 美女脱内裤让男人舔精品视频| 大香蕉久久网| 国产白丝娇喘喷水9色精品| 国产成人精品一,二区| 超色免费av| 免费女性裸体啪啪无遮挡网站| 99国产精品免费福利视频| 涩涩av久久男人的天堂| 亚洲精品中文字幕在线视频| 精品少妇一区二区三区视频日本电影 | 丝袜喷水一区| 熟女少妇亚洲综合色aaa.| 国产亚洲av片在线观看秒播厂| 男的添女的下面高潮视频| 麻豆av在线久日| 69精品国产乱码久久久| 午夜福利网站1000一区二区三区| 一区二区三区精品91| 久久久久精品久久久久真实原创| 亚洲精品乱久久久久久| 丰满乱子伦码专区| 丰满迷人的少妇在线观看| 黑人欧美特级aaaaaa片| 黄色毛片三级朝国网站| av在线播放精品| 一区二区三区乱码不卡18| 少妇熟女欧美另类| 伊人久久国产一区二区| 日本欧美视频一区| 日本爱情动作片www.在线观看| 男女啪啪激烈高潮av片| 国产极品粉嫩免费观看在线| 啦啦啦在线观看免费高清www| 18禁裸乳无遮挡动漫免费视频| 成人漫画全彩无遮挡| 亚洲 欧美一区二区三区| 久久 成人 亚洲| 午夜日韩欧美国产| 最近的中文字幕免费完整| 九九爱精品视频在线观看| 99国产精品免费福利视频| 国产精品国产三级专区第一集| 女人精品久久久久毛片| av网站在线播放免费| 啦啦啦视频在线资源免费观看| 人妻 亚洲 视频| xxx大片免费视频| 一本—道久久a久久精品蜜桃钙片| 两个人看的免费小视频| xxxhd国产人妻xxx| 三级国产精品片| 伊人久久国产一区二区| 亚洲第一青青草原| 纵有疾风起免费观看全集完整版| 少妇的丰满在线观看| 搡老乐熟女国产| 国产在视频线精品| 人妻少妇偷人精品九色| 亚洲三区欧美一区| 久久久a久久爽久久v久久| 日韩,欧美,国产一区二区三区| 亚洲精品自拍成人| 男女高潮啪啪啪动态图| 天天躁日日躁夜夜躁夜夜| 下体分泌物呈黄色| 狠狠精品人妻久久久久久综合| 国产日韩欧美在线精品| 热99久久久久精品小说推荐| 交换朋友夫妻互换小说| 如日韩欧美国产精品一区二区三区| 成人国产av品久久久| av在线老鸭窝| 国产精品不卡视频一区二区| 成人亚洲欧美一区二区av| 亚洲精品久久久久久婷婷小说| 又粗又硬又长又爽又黄的视频| xxx大片免费视频| 性色avwww在线观看| 日韩一区二区三区影片| 精品视频人人做人人爽| 亚洲视频免费观看视频| 国产福利在线免费观看视频| 热99久久久久精品小说推荐| 黑丝袜美女国产一区| 最近最新中文字幕大全免费视频 | 国产成人免费观看mmmm| 欧美精品av麻豆av| videossex国产| 麻豆精品久久久久久蜜桃| 夜夜骑夜夜射夜夜干| 水蜜桃什么品种好| 精品人妻一区二区三区麻豆| 波多野结衣av一区二区av| 精品少妇黑人巨大在线播放| 免费观看在线日韩| 在线看a的网站| 中文字幕人妻熟女乱码| 在线观看三级黄色| 黄色怎么调成土黄色| 国语对白做爰xxxⅹ性视频网站| 一本久久精品| 91成人精品电影| 伦精品一区二区三区| 永久免费av网站大全| 国产国语露脸激情在线看| 久久热在线av| 国产 精品1| 精品亚洲乱码少妇综合久久| 久久久欧美国产精品| 最近中文字幕2019免费版| 国产一区二区三区av在线| 少妇人妻 视频| 大片电影免费在线观看免费| 国产av国产精品国产| 人人妻人人添人人爽欧美一区卜| 国产无遮挡羞羞视频在线观看| 蜜桃国产av成人99| 在线观看免费视频网站a站| 亚洲精品aⅴ在线观看| 视频区图区小说| 男女边吃奶边做爰视频| 亚洲,一卡二卡三卡| 亚洲欧美日韩另类电影网站| 成年动漫av网址| 成人亚洲精品一区在线观看| 久久久国产精品麻豆| 国产视频首页在线观看| 精品一区在线观看国产| 中文字幕最新亚洲高清| 色婷婷久久久亚洲欧美| av在线app专区| 国产免费又黄又爽又色| 在线亚洲精品国产二区图片欧美| 免费看不卡的av| 国产免费福利视频在线观看| 爱豆传媒免费全集在线观看| 丰满少妇做爰视频| 精品人妻一区二区三区麻豆| 亚洲成av片中文字幕在线观看 | 男女下面插进去视频免费观看| av福利片在线| 在线观看美女被高潮喷水网站| 99久久精品国产国产毛片| 亚洲欧美日韩另类电影网站| 精品久久蜜臀av无| 伊人亚洲综合成人网| a级毛片在线看网站| 国产国语露脸激情在线看| 日本爱情动作片www.在线观看| 亚洲五月色婷婷综合| 亚洲三级黄色毛片| 精品国产一区二区三区久久久樱花| 少妇被粗大猛烈的视频| xxxhd国产人妻xxx| 一边摸一边做爽爽视频免费| 国产淫语在线视频| 日本-黄色视频高清免费观看| 少妇人妻久久综合中文| 热re99久久国产66热| 中文字幕亚洲精品专区| www.av在线官网国产| 亚洲美女黄色视频免费看| 男女国产视频网站| 国产 精品1| 亚洲精品美女久久av网站| 一级毛片 在线播放| 久久久久国产精品人妻一区二区| 欧美激情极品国产一区二区三区| 国产精品免费大片| 十八禁高潮呻吟视频| 久久久久久久久久久久大奶| 午夜日韩欧美国产| 伦精品一区二区三区| 亚洲一级一片aⅴ在线观看| 秋霞在线观看毛片| 免费观看在线日韩| 久久久久久久久免费视频了| 青春草视频在线免费观看| 午夜av观看不卡| 午夜福利影视在线免费观看| 欧美日本中文国产一区发布| 高清av免费在线| 纯流量卡能插随身wifi吗| 永久免费av网站大全| 国产av国产精品国产| 男女无遮挡免费网站观看| 哪个播放器可以免费观看大片| 我要看黄色一级片免费的| 久久亚洲国产成人精品v| 久久久久久伊人网av| 亚洲精品第二区| 伦理电影大哥的女人| 亚洲精品国产一区二区精华液| 香蕉精品网在线| 亚洲一码二码三码区别大吗| 国产av国产精品国产| h视频一区二区三区| 美女主播在线视频| 青春草国产在线视频| 美女脱内裤让男人舔精品视频| 97在线视频观看| 香蕉精品网在线| av网站免费在线观看视频| 中文字幕av电影在线播放| 亚洲三级黄色毛片| 亚洲av.av天堂| 国产乱来视频区| 老司机影院毛片| 久久ye,这里只有精品| 国产成人91sexporn| 少妇人妻精品综合一区二区| 亚洲av日韩在线播放| 啦啦啦中文免费视频观看日本| 天天躁狠狠躁夜夜躁狠狠躁| 99香蕉大伊视频| 曰老女人黄片| 国产有黄有色有爽视频| 少妇猛男粗大的猛烈进出视频| 日韩欧美一区视频在线观看| 亚洲精品乱久久久久久| 亚洲av综合色区一区| 午夜福利网站1000一区二区三区| 性色avwww在线观看| 亚洲在久久综合| 777米奇影视久久| 天天躁夜夜躁狠狠躁躁| 日本av免费视频播放| 亚洲精品日本国产第一区| 九草在线视频观看| 久久热在线av| 成年动漫av网址| 精品久久久久久电影网| 中文字幕亚洲精品专区| 两性夫妻黄色片| 在线 av 中文字幕| 90打野战视频偷拍视频| 天堂8中文在线网| 国产精品蜜桃在线观看| h视频一区二区三区| 成年人午夜在线观看视频| 老汉色∧v一级毛片| 午夜福利,免费看| 国产极品粉嫩免费观看在线| 一级毛片我不卡| 一个人免费看片子| 久久亚洲国产成人精品v| 制服丝袜香蕉在线| 欧美另类一区| 国产精品国产三级国产专区5o| 涩涩av久久男人的天堂| 国产伦理片在线播放av一区| 一区二区三区精品91| 亚洲欧洲精品一区二区精品久久久 | 1024香蕉在线观看| 国产免费福利视频在线观看| 国产亚洲欧美精品永久| 亚洲欧美一区二区三区国产| 亚洲精品一二三| 日本午夜av视频| 一级毛片我不卡| 日韩精品免费视频一区二区三区| 日韩av不卡免费在线播放| 亚洲情色 制服丝袜| 18禁动态无遮挡网站| 黑丝袜美女国产一区| 久久久久国产一级毛片高清牌| 亚洲一区中文字幕在线| 日韩一区二区视频免费看| 日本色播在线视频| 欧美日韩国产mv在线观看视频| 99久久人妻综合| 欧美精品一区二区免费开放| 韩国av在线不卡| 满18在线观看网站| 五月天丁香电影| 免费高清在线观看日韩| 国产精品久久久久久av不卡| 大香蕉久久网| 黄色毛片三级朝国网站| 亚洲国产日韩一区二区| 亚洲 欧美一区二区三区| 亚洲精华国产精华液的使用体验| xxx大片免费视频| 十八禁高潮呻吟视频| 欧美日韩一级在线毛片| 国产欧美日韩一区二区三区在线| 婷婷色综合大香蕉| 久久久久久免费高清国产稀缺| 亚洲精品视频女| 在线 av 中文字幕| 国产成人精品无人区| av电影中文网址| 国产免费现黄频在线看| 国产免费视频播放在线视频| 尾随美女入室| 国产精品女同一区二区软件| 1024香蕉在线观看| 欧美国产精品一级二级三级| 人妻 亚洲 视频| 777米奇影视久久| 久久久a久久爽久久v久久| 国产精品秋霞免费鲁丝片| 久久久久久人人人人人| 免费观看a级毛片全部| 国产亚洲一区二区精品| 国产精品成人在线| 日韩中文字幕欧美一区二区 | 国产男女超爽视频在线观看| 精品少妇内射三级| 性高湖久久久久久久久免费观看| 高清不卡的av网站| 亚洲,欧美精品.| 黑人猛操日本美女一级片| 日韩视频在线欧美| 2018国产大陆天天弄谢| 日韩电影二区| 午夜久久久在线观看| 国产成人精品一,二区| av免费在线看不卡| 午夜福利视频在线观看免费| 中国三级夫妇交换| 国产xxxxx性猛交| 人人妻人人澡人人看| 国产黄色视频一区二区在线观看| 女人久久www免费人成看片| 国产精品香港三级国产av潘金莲 | 中文精品一卡2卡3卡4更新| 18禁动态无遮挡网站| 亚洲精品日韩在线中文字幕| 我要看黄色一级片免费的| 99国产精品免费福利视频| 男人操女人黄网站| 丰满人妻熟妇乱又伦精品不卡| 老熟妇仑乱视频hdxx| 亚洲中文av在线| 亚洲一区高清亚洲精品| 久久婷婷成人综合色麻豆| 夜夜爽天天搞| 99久久99久久久精品蜜桃| 国产欧美日韩一区二区三| 国产97色在线日韩免费| 中文亚洲av片在线观看爽| 精品久久久久久久毛片微露脸| 黑人猛操日本美女一级片| 日韩精品青青久久久久久| 一夜夜www| 正在播放国产对白刺激| 97碰自拍视频| 久久久久久久精品吃奶| 精品一区二区三区av网在线观看| 丁香欧美五月| 波多野结衣一区麻豆| 欧美不卡视频在线免费观看 | 亚洲国产欧美一区二区综合| 免费在线观看日本一区| 国产精品久久久av美女十八| 波多野结衣一区麻豆| 手机成人av网站| 淫秽高清视频在线观看| 国产精品一区二区在线不卡| 午夜a级毛片| 久久九九热精品免费| 香蕉久久夜色| 18禁国产床啪视频网站| 国产aⅴ精品一区二区三区波| 无遮挡黄片免费观看| 欧美一级毛片孕妇| 精品人妻在线不人妻| 成熟少妇高潮喷水视频| 一进一出好大好爽视频| 夜夜躁狠狠躁天天躁| 成年人免费黄色播放视频| 免费在线观看日本一区| 90打野战视频偷拍视频| 两性午夜刺激爽爽歪歪视频在线观看 | 正在播放国产对白刺激| 人人妻人人爽人人添夜夜欢视频| 免费一级毛片在线播放高清视频 | 很黄的视频免费| 免费看a级黄色片| www国产在线视频色| 久久狼人影院| 天堂影院成人在线观看| 久久久水蜜桃国产精品网| 国产1区2区3区精品| 欧美日韩中文字幕国产精品一区二区三区 | 欧美黄色淫秽网站| 99re在线观看精品视频| 国产aⅴ精品一区二区三区波| 午夜免费观看网址| 国产成人av激情在线播放| 日韩欧美三级三区| 50天的宝宝边吃奶边哭怎么回事| 午夜福利免费观看在线| 老司机深夜福利视频在线观看| 久久人妻熟女aⅴ| 亚洲午夜精品一区,二区,三区| 老鸭窝网址在线观看| 亚洲人成网站在线播放欧美日韩| 真人一进一出gif抽搐免费| 国产精品综合久久久久久久免费 | www.精华液| 黄色女人牲交| 国产精品亚洲av一区麻豆| 老司机靠b影院| 日韩精品青青久久久久久| 美女高潮到喷水免费观看| 黑人猛操日本美女一级片| 亚洲成人免费av在线播放| 久9热在线精品视频| 在线观看www视频免费| 国产黄色免费在线视频| 精品久久久精品久久久| 久久天堂一区二区三区四区| 久久精品国产99精品国产亚洲性色 | 在线观看午夜福利视频| 亚洲av成人一区二区三| 久久亚洲精品不卡| 久久久国产成人精品二区 | 又大又爽又粗| 国产免费男女视频| 国产亚洲欧美精品永久| 国产亚洲精品第一综合不卡| 在线av久久热| 日韩欧美一区二区三区在线观看| 久久人人爽av亚洲精品天堂| 国产亚洲欧美在线一区二区| 又黄又粗又硬又大视频| 亚洲黑人精品在线| 黄网站色视频无遮挡免费观看| 欧美日韩一级在线毛片| 久久国产乱子伦精品免费另类| 欧美中文综合在线视频| 岛国视频午夜一区免费看| 麻豆一二三区av精品| 欧美激情高清一区二区三区| 久热爱精品视频在线9| 制服诱惑二区| 最好的美女福利视频网| 精品久久久久久成人av| 亚洲久久久国产精品| 午夜日韩欧美国产| 亚洲aⅴ乱码一区二区在线播放 | 日本撒尿小便嘘嘘汇集6| 久久国产精品男人的天堂亚洲| 国内久久婷婷六月综合欲色啪| 亚洲精品美女久久久久99蜜臀| 91精品国产国语对白视频| 欧美日韩亚洲综合一区二区三区_| www日本在线高清视频| 一级片免费观看大全| 老汉色∧v一级毛片| aaaaa片日本免费| 午夜福利欧美成人| 精品久久久久久久久久免费视频 | 色综合欧美亚洲国产小说| av视频免费观看在线观看| 人妻丰满熟妇av一区二区三区| 国产av精品麻豆| 欧美国产精品va在线观看不卡| 老鸭窝网址在线观看| 999精品在线视频| 亚洲免费av在线视频| 在线观看一区二区三区| 国产人伦9x9x在线观看| 亚洲精品粉嫩美女一区| 国产精品乱码一区二三区的特点 | 我的亚洲天堂| 亚洲九九香蕉| 在线观看66精品国产| 女警被强在线播放| 一a级毛片在线观看| 黄色怎么调成土黄色| 俄罗斯特黄特色一大片| 欧美成人午夜精品| av福利片在线| 天天躁狠狠躁夜夜躁狠狠躁| 亚洲情色 制服丝袜| 嫩草影院精品99| 久久久久国产精品人妻aⅴ院| 青草久久国产| 久久久久亚洲av毛片大全| 性欧美人与动物交配| 精品一品国产午夜福利视频| 精品一区二区三区av网在线观看| 五月开心婷婷网| 好看av亚洲va欧美ⅴa在| 日韩视频一区二区在线观看| av电影中文网址| 久久久久久久久久久久大奶| 亚洲专区中文字幕在线| 久久久国产一区二区| 国产精品二区激情视频| 国产免费男女视频| 午夜精品在线福利| 神马国产精品三级电影在线观看 | 国产精品久久久av美女十八| 天堂影院成人在线观看| 黑丝袜美女国产一区| 亚洲欧美精品综合一区二区三区| 欧美人与性动交α欧美精品济南到| 欧美av亚洲av综合av国产av| 人人妻人人添人人爽欧美一区卜| 免费女性裸体啪啪无遮挡网站| 一二三四在线观看免费中文在| bbb黄色大片| 久久亚洲精品不卡| 麻豆一二三区av精品| a级毛片在线看网站| 18禁观看日本| 丰满人妻熟妇乱又伦精品不卡| 欧美日韩中文字幕国产精品一区二区三区 | 国产99久久九九免费精品| 日本a在线网址| 亚洲一区二区三区色噜噜 | 999精品在线视频| 在线av久久热| 国产精品二区激情视频| 在线十欧美十亚洲十日本专区| 久久性视频一级片| 亚洲美女黄片视频| 久久久久久久久免费视频了| 99国产综合亚洲精品| 精品少妇一区二区三区视频日本电影| 美女 人体艺术 gogo| 两个人免费观看高清视频| 一级毛片精品| 满18在线观看网站| 午夜视频精品福利| 黑人猛操日本美女一级片| 亚洲av电影在线进入| 亚洲熟女毛片儿| 亚洲免费av在线视频| 欧美黑人欧美精品刺激| 色婷婷久久久亚洲欧美| bbb黄色大片| 一区福利在线观看| 精品国产国语对白av| 黄片大片在线免费观看| 国产亚洲精品第一综合不卡| 黄片播放在线免费| 在线看a的网站| 免费在线观看影片大全网站| 欧美另类亚洲清纯唯美| 久久精品人人爽人人爽视色| 国产乱人伦免费视频| 无人区码免费观看不卡| 亚洲第一青青草原| 超碰成人久久| 久9热在线精品视频| 精品人妻在线不人妻| 91在线观看av| 亚洲色图 男人天堂 中文字幕| 精品一区二区三卡| 色综合婷婷激情| 久久久久九九精品影院| 久久国产精品男人的天堂亚洲| 中文字幕av电影在线播放| 一边摸一边抽搐一进一小说| 91成人精品电影| 国产99久久九九免费精品| 国产精品久久视频播放| 午夜免费成人在线视频| 国产精品一区二区免费欧美| 99国产精品一区二区蜜桃av| 日韩精品青青久久久久久| 黑人欧美特级aaaaaa片| 国产成人精品久久二区二区91| 黑人巨大精品欧美一区二区蜜桃| 欧美久久黑人一区二区| 国产成人精品无人区| 人人妻,人人澡人人爽秒播| 最近最新中文字幕大全电影3 | 在线观看免费视频网站a站| 国产成人精品久久二区二区免费| 国产精品偷伦视频观看了| 啦啦啦免费观看视频1| 色播在线永久视频| 两性夫妻黄色片| 久久亚洲精品不卡| 一本综合久久免费| 波多野结衣一区麻豆| 精品人妻在线不人妻|