• <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 ?

    Feature combination via importance-inhibition analysis

    2013-09-17 05:59:58YangSichunGaoChaoYaoJiaminDaiXinyuChenJiajun
    關(guān)鍵詞:省份河流

    Yang Sichun Gao Chao Yao Jiamin Dai Xinyu Chen Jiajun

    (1State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210093, China)

    (2School of Computer Science, Anhui University of Technology, Maanshan 243032, China)

    (3School of Computer Science and Information Engineering, Chuzhou University, Chuzhou 239000, China)

    A utomatic question answering(QA)[1]is a hot research direction in the field of natural language processing(NLP)and information retrieval(IR),which allows users to ask questions in natural language,and returns concise and accurate answers.QA systems include three major modules, namely question analysis, paragraph retrieval and answer extraction.As a crucial component of question analysis,question classification classifies questions into several semantic categories which indicate the expected semantic type of answers to questions.The semantic category of a question helps to filter out irrelevant answer candidates,and determine the answer selection strategies.

    In current research on question classification,the method based on machine learning is widely used,and features are the key to building an accurate question classifier[2-10].Li et al.[2-3]presented a hierarchical classifier based on the sparse network of winnows(SNoW)architecture, and made use of rich features, such as words,parts of speech, named entity, chunk, head chunk, and class-specific words.Zhang et al.[4]proposed a tree kernel support vector machine classifier,and took advantage of the structural information of questions.Huang et al.[5-6]extracted head word features and presented two approaches to augment hypernyms of such head words using WordNet.However, when used to train question classifiers,these features were almost combined incrementally via importance analysis(IA)which is based on the importance of individual features.This method is effective when using only a few features,but for very rich features,it may prevent question classification from further improvement due to the problem of ignoring the inhibition among features.

    In order to alleviate this problem,this paper proposes a new method for combining features via importance-inhibition analysis(IIA).By taking into account the inhibition among features as well as the importance of individual features,the IIA method more objectively depicts the process of combining features,and can further improve the performance of question classification.Experimental results on the Chinese questions set show that the IIA method performs more effectively than the IA method on the whole,and achieves the same highest accuracy as the one by the exhaustive method.

    1 Feature Extraction

    We use an open and free available language technology platform(LTP)(http://ir.hit.edu.cn/demo/ltp)which integrates ten key Chinese processing modules on morphology, word sense, syntax, semantics and other document analysis,and take the question“中國(guó)哪一條河流經(jīng)過(guò)的省份最多?(Which river flows through most provinces in China?)”as an example.The result of word segmentation, POS tagging, named entity recognition and dependency parsing of the sample question is presented in Fig.1.

    We extract bag-of-words(BOW),part-of-speech(POS), word sense(WSD,WSDm), named entity(NE),dependency relation(R)and parent word(P)as basic features.Here, WSD is the 3-layer coding, i.e.,coarse,medium and finegrained categoriesin the semantic dictionary “TongYiCiCiLin”, while WSDm is the 2-layer, i.e., coarse and medium grained word category.Tab.1 gives the features and their values of the sample question.

    Fig.1 Analysis result of the sample question with LTP platform

    Tab.1 Features and their values of the sample question

    2 Combining Features via Importance-inhibition Analysis

    The basic features described above belong to different syntactic and semantic categories,and contribute to question classification from various levels of language knowledge.We combine these basic features to further improve the performance of question classification.Since the BOW feature is the basis of other features,it is always combined with other features.For example, the POS feature follows the BOW feature when these two types of features are combined.

    With respect to the methods for combining features,the most intuitive one is the exhaustive method which lists all the feature combinations one by one.The exhaustive method is inefficient and not feasible in practical applications.In existing literature, combining features is conducted just on the basis of the importance of the features.However,this method may prevent it from further improvement on question classification due to the problem of ignoring the inhibition among features.For example,the dependency relation feature R and the POS feature belong to the same syntactic category,and they both contribute to question classification.However, since R covers POS to a large extent in syntactic expression,R will inhibit POS when they appear in the same feature combination.Similarly,the word sense features WSD and WSDm belong to the same semantic category,since the difference between WSD and WSDm is not obvious,they will inhibit each other when they are present at the same feature combination.From the above discussions, we find that an effective method for combining features should take into account the inhibition among features as well as the importance of individual features.

    In this paper,we propose a new method for combining features via importance-inhibition analysis.Before introducing the IIA method in detail,we should specify some notations.In our importance-inhibition analysis setting,the feature set is a basic concept following the common feature combination.

    A feature setFconsists of each featurefiextracted from a question, i.e.F={fii=1,2, …};F'is a subset ofF,and consists of each featuref(i)which has side effects for feature combinations, i.e.F'={f(i)i=1,2,…};F(ij)denotes thej-th one in thei-th round of feature combination,and it is a subset ofF;F*idenotes a feature combination with the highest accuracy in thei-th round,and it is also a subset ofF.

    Now we can give some formal definitions.

    Definition 1(importance) Given featuresfiandfj,fiis more important thanfjif the accuracy offiis higher than that offj.

    Definition 2(inhibition) Given a featurefiand a feature combinationF(ij),there exists inhibition betweenFi(j)andfiif the accuracy of the feature combinationF(ij)∪{fi}is lower than that ofF(ij)orfi.

    Definition 3(k_ary combination) Given a feature set F(ij),it is ak_ary feature combination in whichkfeatures are contained.

    Definition 4(bestk_ary combination) Given a(k-1)_ary combinationF(ij)and a candidate featurefi,F(xiàn)(ij)∪{fi}is the bestk_ary combination if it has the highest accuracy in the current round of feature combinations.

    Now let us move to the details of the IIA method.From the above definitions, we can easily see that, given featuresfi,fjand a feature combinationF(ij),the accuracy ofF(ij)∪{fi}is not always higher than that ofF(ij)∪{fj}whenfiis more important thanfj.By taking into account the inhibition among features,we combine features via a heuristic algorithm.First,choose BOW as the best 1_ary feature combination,and combine each candidate feature from the rest with BOW to form 2_ary feature combinations.Then choose the one with the highest accuracy as the best 2_ary feature combination,and filter out those features lower than the best 1_ary feature combination.Finally,repeat the above steps until the current candidate feature set is empty or all the feature combinations are no longer higher than the highest in the previous round.

    Algorithm 1 gives the implement of the IIA method.

    Algorithm 1Importance-inhibition analysis algorithm

    The IIA method is on the basis of the(k-1)_ary feature combination to obtain the bestk_ary one,so compared with the exhaustive method,it can significantly improve the efficiency of feature combination.In addition,since the IIA method takes into account the inhibition among features as well as the importance of individual features, compared with the IA method, it can more objectively depict the process of combining features and ensure a better performance of question classification.

    3 Experimental Results and Analysis

    3.1 Data set and evaluation

    In our experiments,we use the Chinese questions set provided by IRSC lab of HIT(http://ir.hit.edu.cn),which contains 6 266 questions belonging to 6 categories and 77 classes.

    The open and free available Liblinear-1.4(http://www.csie.ntu.edu.tw/~ cjlin/liblinear/)which is a linear classifier for data with millions of instances and features which is used to be the classifier.We use 10-fold cross validation on the total question set to evaluate the performance of the question classifications.

    3.2 Combining features via IIA

    According to the IIA method,we take BOW as the initial feature,and combine POS,NE,WSD,WSDm,R and P features gradually to form feature combinations,such as 2_ary,3_ary,4_ary and so on.The accuracies of individual features are presented in Fig.2(a).Figs.2(b)to(d)list all the accuracies of 2_ary,3_ary and 4_ary feature combinations respectively, where Base1, Base2 and Base3 stand for the corresponding best 1_ary, 2_ary,3_ary feature combinations.

    Fig.2 Accuracies of n_ary feature combinations.(a)1_ary;(b)2_ary;(c)3_ary;(d)4_ary

    In Fig.2(b)and Fig.2(c), the P feature has the highest classification accuracy among all the candidates,but the accuracies of Base1+P and Base2+P are not the highest in all the 2_ary and 3_ary feature combinations,respectively.In particular, the accuracy of Base1+P is the last but one in all the 2_ary feature combinations.

    In Fig.2(b), the accuracy of Base1+NE is lower than that of Base1,so NE is no longer considered in subsequent rounds.Similarly, in Fig.2(d), the accuracies of Base3+POS and Base3+WSDm are both lower than that of Base3,so POS and WSDm are not considered in subsequent rounds.This is greatly convenient for filtering noise features.

    In Fig.2(c)and Fig.2(d), the accuracies of Base1+NE,Base3+POS,Base3+WSDm are lower than those of Base1 and Base3, respectively.The reason is that R covers POS to a large extent in syntactic expression,and the difference between WSD and WSDm is very small.As a result,there exists the inhibition among features when they are in the same feature combination.

    3.3 Performance comparison with IA

    In order to verify the efficiency and effectiveness of IIA,we conduct performance comparison with IA.Tab.2 shows the accuracies of the feature combinations via IIA and IA,respectively,where the“2_ary”column means 2_ary combinations, the “Base”row denotes the best(n-1)_ary combinations, “+POS”row means the feature combined with its baseline,the accuracy in bold means the maximum ofn_ary combinations,and the one in bold with underline shows the maximum of all the combinations.

    Tab.2 Accuracies of feature combinations via IIA and IA %

    Fig.3 conducts the comparison of average and maximum accuracies between IIA and IA,where theXaxis denotesn_ary feature combinations,theYaxis denotes classification accuracies.

    Fig.3 Performance comparison between IIA and IA

    From Fig.3, we can see that IIA shows a gradual increase in average and maximum accuracies in all the feature combinations,while IA shows a slight decline in accuracy at the 4_ary and 7_ary ones.The reason is that IIA is based on the best previous feature combination to obtain the current one.In addition, IIA performs as well as IA in average accuracy at 3_ary feature combinations,and achieves a great improvement over IA in average and maximum accuracies at 2_ary and 4_ary feature combinations.In particular, IIA achieves 0.813 9% and 0.829 9%higher than IA in average and maximum accuracies at 4_ary feature combinations,so we can draw a conclusion that IIA performs significantly better than IA on the whole.

    In order to further verify the efficiency and effectiveness of IIA,we conduct performance comparison with the exhaustive method.Experimental results show that the exhaustive method carries on 6 rounds for acquiring 63 feature combinations,while IIA does 3 rounds with 13 feature combinations gained.This demonstrates that IIA is much more efficient and feasible than the exhaustive method in practical applications.Furthermore, IIA gets the accuracy of 82.413%which is the highest one gained by the exhaustive method.

    4 Conclusion

    In this paper,we propose a new method called IIA to combine features via importance-inhibition analysis.The method takes into account the inhibition among various features as well as the importance of individual features.Experimental results on the Chinese question set show that the IIA method performs more effectively than the IA method on the whole,and achieves the same highest accuracy as the one gained by the exhaustive method.

    The IIA method is a heuristic one in nature,and may be faced with the problem of a local optimum.In our further work,we will make great efforts to achieve more efficient and effective optimization for combining features.

    Acknowlegement We would like to thank the IRSC laboratory of Harbin Institute of Technology for their free and available LTP platform.

    [1]Zhang Z C, Zhang Y, Liu T, et al.Advances in opendomain question answering [J].Acta Electronica Sinica,2009,37(5):1058-1069.(in Chinese)

    [2]Li X, Roth D.Learning question classifiers[C]//Proc of the19th International Conference on Computational Linguistics.Taipei,China, 2002:1-7.

    [3]Li X, Roth D.Learning question classifiers:the role of semantic information[J].Journal of Natural Language Engineering, 2006, 12(3):229-250.

    [4]Zhang D, Lee W.Question classification using support vector machines[C]//Proc of the26th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval.Toronto, Canada, 2003:26-32.

    [5]Huang Z H,Thint M,Qin Z C.Question classification using head words and their hypernyms[C]//Proc of the2008Conference on Empirical Methods in Natural Language Processing.Honolulu, Hawaii, USA, 2008:927-936.

    [6]Huang Z H,Thint M,Celikyilmaz A.Investigation of question classifier in question answering[C]//Proc of the2009Conference on Empirical Methods in Natural Language Processing.Singapore, 2009:543-550.

    [7]Li F T,Zhang X,Yuan J H,et al.Classifying what-type questions by head noun tagging[C]//Proc of the22nd InternationalConferenceonComputationalLinguistics.Manchester,UK, 2008:481-488.

    [8]Li X,Huang X J,Wu L D.Combined multiple classifiers based on TBL algorithm and their application in question classification [J].Journal of Computer Research and Development, 2008, 45(3):535-541.(in Chinese)

    [9]Sun J G,Cai D F,Lu D X,et al.HowNet based Chinese question automatic classification [J].Journal of Chinese Information Processing, 2007, 21(1):90-95.(in Chinese)

    [10]Zhang Z C, Zhang Y, Liu T, et al.Chinese question classification based on identification of cue words and extension of training set[J].Chinese High Technology Letters, 2009, 19(2):111-118.(in Chinese)

    猜你喜歡
    省份河流
    誰(shuí)說(shuō)小龍蝦不賺錢?跨越四省份,暴走萬(wàn)里路,只為尋找最會(huì)養(yǎng)蝦的您
    河流
    16省份上半年GDP超萬(wàn)億元
    流放自己的河流
    河流
    河流
    22個(gè)省
    決策探索(2017年11期)2017-06-23 18:41:32
    當(dāng)河流遇見(jiàn)海
    因地制宜地穩(wěn)妥推進(jìn)留地安置——基于對(duì)10余省份留地安置的調(diào)研
    靜靜的河流
    雕塑(2000年2期)2000-06-22 16:13:30
    韩国高清视频一区二区三区| 日韩在线高清观看一区二区三区| 好男人视频免费观看在线| 少妇人妻 视频| 成年人免费黄色播放视频 | 下体分泌物呈黄色| 精品国产乱码久久久久久小说| 嘟嘟电影网在线观看| 一级片'在线观看视频| 三级国产精品欧美在线观看| 亚洲欧美成人精品一区二区| 久久久久精品久久久久真实原创| 18+在线观看网站| 黄色视频在线播放观看不卡| 成人特级av手机在线观看| 国产精品国产三级国产av玫瑰| 哪个播放器可以免费观看大片| 五月开心婷婷网| 成年av动漫网址| 亚洲精品久久午夜乱码| 久久99一区二区三区| 国产在线视频一区二区| 妹子高潮喷水视频| 成人无遮挡网站| 欧美日本中文国产一区发布| 我的女老师完整版在线观看| 成人免费观看视频高清| 日韩一区二区视频免费看| 亚洲,一卡二卡三卡| 日韩精品有码人妻一区| 久久人妻熟女aⅴ| 久久久久久久久久成人| 国产91av在线免费观看| 国产黄色视频一区二区在线观看| 18禁在线无遮挡免费观看视频| 观看av在线不卡| 伊人亚洲综合成人网| 国产伦在线观看视频一区| 18禁动态无遮挡网站| 18禁在线无遮挡免费观看视频| 一级a做视频免费观看| 91aial.com中文字幕在线观看| 日本免费在线观看一区| 中文在线观看免费www的网站| 人妻少妇偷人精品九色| 夜夜骑夜夜射夜夜干| 久久久久视频综合| 欧美精品一区二区大全| 免费黄频网站在线观看国产| 女的被弄到高潮叫床怎么办| 另类精品久久| 少妇人妻精品综合一区二区| 色吧在线观看| 国产成人精品无人区| 色5月婷婷丁香| 人妻一区二区av| 国产伦在线观看视频一区| kizo精华| 妹子高潮喷水视频| 嫩草影院入口| 噜噜噜噜噜久久久久久91| 大片电影免费在线观看免费| 中文精品一卡2卡3卡4更新| 久久久久久久久大av| 三级经典国产精品| 日本黄色日本黄色录像| 成年av动漫网址| 卡戴珊不雅视频在线播放| 久久人人爽av亚洲精品天堂| 一级二级三级毛片免费看| 一区二区三区四区激情视频| 成人影院久久| 女性被躁到高潮视频| av一本久久久久| 中国三级夫妇交换| 亚洲欧洲国产日韩| 精品久久久久久电影网| 午夜激情久久久久久久| 亚洲av成人精品一二三区| 国产精品福利在线免费观看| 你懂的网址亚洲精品在线观看| 麻豆成人av视频| 国产又色又爽无遮挡免| kizo精华| 国产一区有黄有色的免费视频| 晚上一个人看的免费电影| 曰老女人黄片| 人妻人人澡人人爽人人| 最近中文字幕高清免费大全6| 国产精品三级大全| 一本—道久久a久久精品蜜桃钙片| 国产精品秋霞免费鲁丝片| 亚洲第一区二区三区不卡| 欧美日韩一区二区视频在线观看视频在线| 亚洲图色成人| 久久人人爽人人片av| 成人无遮挡网站| 国精品久久久久久国模美| 男女国产视频网站| 成人国产av品久久久| 国产成人精品一,二区| 亚洲国产色片| 人人妻人人添人人爽欧美一区卜| 日韩精品有码人妻一区| 亚洲一级一片aⅴ在线观看| 少妇的逼好多水| 日韩av免费高清视频| 日韩伦理黄色片| 26uuu在线亚洲综合色| 九九久久精品国产亚洲av麻豆| 我的女老师完整版在线观看| 欧美精品人与动牲交sv欧美| 国产一区二区三区综合在线观看 | 日日啪夜夜撸| 免费观看a级毛片全部| 另类精品久久| 国产精品熟女久久久久浪| 亚洲婷婷狠狠爱综合网| 国产亚洲91精品色在线| 精品久久国产蜜桃| 久久综合国产亚洲精品| 亚洲成人手机| 能在线免费看毛片的网站| 亚洲欧美日韩卡通动漫| 一本大道久久a久久精品| 国模一区二区三区四区视频| 丰满人妻一区二区三区视频av| 有码 亚洲区| 人人澡人人妻人| 插阴视频在线观看视频| 国产男人的电影天堂91| 亚洲性久久影院| 久久亚洲国产成人精品v| 少妇高潮的动态图| 国产女主播在线喷水免费视频网站| 久久久久久久久久成人| 熟妇人妻不卡中文字幕| 纯流量卡能插随身wifi吗| 国产精品国产三级国产专区5o| 秋霞在线观看毛片| 好男人视频免费观看在线| 青青草视频在线视频观看| 国产精品一区二区在线不卡| 青春草亚洲视频在线观看| 看非洲黑人一级黄片| 交换朋友夫妻互换小说| 22中文网久久字幕| 久久av网站| 又爽又黄a免费视频| 亚洲精品一区蜜桃| 丝瓜视频免费看黄片| 亚洲精品国产av蜜桃| 嘟嘟电影网在线观看| 免费看日本二区| 美女视频免费永久观看网站| 国产精品久久久久久久久免| 美女视频免费永久观看网站| 香蕉精品网在线| 亚洲精品国产av蜜桃| 九九爱精品视频在线观看| 99热这里只有精品一区| 亚洲人成网站在线播| 美女视频免费永久观看网站| 国产精品国产三级国产专区5o| 国产视频内射| 美女视频免费永久观看网站| 国产精品久久久久久久久免| 久久久久久久大尺度免费视频| 老熟女久久久| 在线观看人妻少妇| 国产精品一二三区在线看| 69精品国产乱码久久久| 成人综合一区亚洲| 高清在线视频一区二区三区| 男女无遮挡免费网站观看| 看免费成人av毛片| 91久久精品电影网| 国产成人精品久久久久久| 自拍偷自拍亚洲精品老妇| 18禁裸乳无遮挡动漫免费视频| 一级a做视频免费观看| 欧美精品国产亚洲| 久久国产乱子免费精品| 一级毛片aaaaaa免费看小| 中文字幕人妻熟人妻熟丝袜美| 久久久精品免费免费高清| av有码第一页| 国产成人精品无人区| 亚洲美女视频黄频| 亚洲熟女精品中文字幕| 免费观看a级毛片全部| 日本vs欧美在线观看视频 | 人妻夜夜爽99麻豆av| 国产黄色免费在线视频| 国产一区亚洲一区在线观看| 亚洲成人手机| 亚洲精品第二区| 亚洲成人一二三区av| 亚洲av成人精品一区久久| 久久久精品免费免费高清| 人妻一区二区av| 黄色日韩在线| 国产一区二区三区av在线| 日韩中文字幕视频在线看片| 国精品久久久久久国模美| 美女cb高潮喷水在线观看| a级片在线免费高清观看视频| 国产白丝娇喘喷水9色精品| 欧美精品一区二区大全| 亚洲综合色惰| 精华霜和精华液先用哪个| 免费播放大片免费观看视频在线观看| 日本vs欧美在线观看视频 | 内地一区二区视频在线| 精品一区二区三卡| 十八禁网站网址无遮挡 | 嫩草影院入口| 狂野欧美白嫩少妇大欣赏| 乱人伦中国视频| 亚洲欧美成人精品一区二区| 国产黄片美女视频| 肉色欧美久久久久久久蜜桃| a级毛片在线看网站| 国产 一区精品| 成人国产av品久久久| 国产一区二区在线观看av| 大片电影免费在线观看免费| 一级二级三级毛片免费看| 六月丁香七月| 国产精品熟女久久久久浪| 国产精品国产三级国产专区5o| 99热这里只有精品一区| 秋霞伦理黄片| 精品人妻熟女毛片av久久网站| 国产色婷婷99| 国产高清三级在线| 亚洲无线观看免费| 在线看a的网站| 搡女人真爽免费视频火全软件| 亚洲精品日韩av片在线观看| 久久久久久伊人网av| 爱豆传媒免费全集在线观看| 久久综合国产亚洲精品| 国产黄色视频一区二区在线观看| 老女人水多毛片| 伦理电影大哥的女人| 男女免费视频国产| 午夜av观看不卡| 青春草亚洲视频在线观看| 亚洲精品第二区| 两个人免费观看高清视频 | 一本色道久久久久久精品综合| 国产一区亚洲一区在线观看| 亚洲欧洲精品一区二区精品久久久 | 看非洲黑人一级黄片| 少妇人妻一区二区三区视频| 亚洲精品乱久久久久久| 最近中文字幕高清免费大全6| 三级国产精品片| 97超碰精品成人国产| 纵有疾风起免费观看全集完整版| 久久久久视频综合| 精品卡一卡二卡四卡免费| 麻豆乱淫一区二区| 精品视频人人做人人爽| 国产黄片美女视频| 国产黄频视频在线观看| 久久人人爽人人爽人人片va| 一级毛片我不卡| 黑人巨大精品欧美一区二区蜜桃 | 免费av中文字幕在线| 哪个播放器可以免费观看大片| 国产成人免费观看mmmm| 91aial.com中文字幕在线观看| 精品一区在线观看国产| 亚洲国产精品成人久久小说| av在线app专区| 色5月婷婷丁香| 国产av码专区亚洲av| 天天躁夜夜躁狠狠久久av| 女人精品久久久久毛片| 精品少妇久久久久久888优播| 狂野欧美激情性bbbbbb| 国产色爽女视频免费观看| 国产成人精品福利久久| 亚洲精品久久久久久婷婷小说| 国产亚洲欧美精品永久| 青春草亚洲视频在线观看| 久久久久网色| 极品少妇高潮喷水抽搐| 欧美激情国产日韩精品一区| 99视频精品全部免费 在线| 婷婷色麻豆天堂久久| 成人国产av品久久久| 国产熟女午夜一区二区三区 | 极品人妻少妇av视频| 男女边吃奶边做爰视频| 一级毛片久久久久久久久女| 日韩欧美精品免费久久| 中文字幕免费在线视频6| 国产精品一区二区三区四区免费观看| 精品久久久久久久久亚洲| 久久97久久精品| 麻豆乱淫一区二区| 国产淫语在线视频| 成年美女黄网站色视频大全免费 | 国产一区二区在线观看av| 热re99久久国产66热| 国产亚洲一区二区精品| 久久精品国产鲁丝片午夜精品| av视频免费观看在线观看| 国产精品一区二区性色av| 色婷婷av一区二区三区视频| 好男人视频免费观看在线| 亚洲av在线观看美女高潮| 国产成人精品福利久久| 伊人亚洲综合成人网| 全区人妻精品视频| 亚洲精品日韩av片在线观看| 免费久久久久久久精品成人欧美视频 | 中文精品一卡2卡3卡4更新| 亚洲欧美中文字幕日韩二区| 在线观看免费视频网站a站| 国产男女超爽视频在线观看| 亚洲精品久久久久久婷婷小说| 97在线视频观看| a 毛片基地| 一级av片app| 日本黄色日本黄色录像| 精品国产乱码久久久久久小说| av卡一久久| 免费黄频网站在线观看国产| 成年美女黄网站色视频大全免费 | av免费观看日本| 99热网站在线观看| 中文精品一卡2卡3卡4更新| 91精品伊人久久大香线蕉| 国产亚洲一区二区精品| 午夜视频国产福利| 在线观看av片永久免费下载| 9色porny在线观看| 欧美xxxx性猛交bbbb| 久久久国产欧美日韩av| 人人妻人人澡人人爽人人夜夜| 精华霜和精华液先用哪个| 91在线精品国自产拍蜜月| 男人狂女人下面高潮的视频| 在线精品无人区一区二区三| 欧美日韩在线观看h| 亚洲精品久久午夜乱码| 韩国高清视频一区二区三区| 人妻一区二区av| 天天躁夜夜躁狠狠久久av| 欧美性感艳星| 热re99久久精品国产66热6| 丁香六月天网| 国产在线视频一区二区| av在线观看视频网站免费| 精品久久久久久久久av| videossex国产| 精品亚洲乱码少妇综合久久| 亚洲三级黄色毛片| 国产成人午夜福利电影在线观看| 欧美xxⅹ黑人| 大码成人一级视频| 亚洲国产精品成人久久小说| 国产av一区二区精品久久| 国产黄色免费在线视频| 有码 亚洲区| 日韩 亚洲 欧美在线| 在线观看免费视频网站a站| 男的添女的下面高潮视频| 在线观看美女被高潮喷水网站| 高清毛片免费看| 欧美xxⅹ黑人| 成人影院久久| 国产精品国产三级国产专区5o| 成人午夜精彩视频在线观看| 少妇人妻一区二区三区视频| videossex国产| 少妇被粗大的猛进出69影院 | 2018国产大陆天天弄谢| 欧美日韩av久久| 成人午夜精彩视频在线观看| 亚洲精品aⅴ在线观看| 狂野欧美激情性xxxx在线观看| 婷婷色av中文字幕| 黄色日韩在线| 九色成人免费人妻av| 亚洲精品456在线播放app| 国产成人精品福利久久| 久久久久久伊人网av| 王馨瑶露胸无遮挡在线观看| 国产亚洲av片在线观看秒播厂| 亚洲第一区二区三区不卡| 亚洲欧洲精品一区二区精品久久久 | 久久精品国产亚洲网站| 日本黄色日本黄色录像| 精品国产乱码久久久久久小说| 午夜福利影视在线免费观看| 97超碰精品成人国产| 日韩伦理黄色片| 亚洲精品国产av成人精品| 99久久精品一区二区三区| 欧美日韩在线观看h| 午夜福利影视在线免费观看| 亚洲精华国产精华液的使用体验| 免费播放大片免费观看视频在线观看| 婷婷色综合www| 18+在线观看网站| 日本与韩国留学比较| 亚洲精品久久午夜乱码| 狂野欧美激情性bbbbbb| 国产伦在线观看视频一区| 亚洲精品日韩av片在线观看| 人人澡人人妻人| av线在线观看网站| 午夜福利在线观看免费完整高清在| 国产精品欧美亚洲77777| 婷婷色av中文字幕| 最近中文字幕2019免费版| 午夜久久久在线观看| 91精品国产九色| 全区人妻精品视频| 国产亚洲最大av| 亚洲人成网站在线播| 大话2 男鬼变身卡| 激情五月婷婷亚洲| 亚洲不卡免费看| 午夜福利网站1000一区二区三区| 九色成人免费人妻av| 久久精品国产亚洲网站| 久久久久国产网址| 99久久中文字幕三级久久日本| 在现免费观看毛片| 久久久欧美国产精品| 观看av在线不卡| 亚洲婷婷狠狠爱综合网| 国产淫语在线视频| 深夜a级毛片| 日韩欧美精品免费久久| 国产伦精品一区二区三区四那| 国产一区二区三区av在线| 久久午夜福利片| 久久国产精品男人的天堂亚洲 | 男人和女人高潮做爰伦理| 熟女av电影| 国内揄拍国产精品人妻在线| 99久国产av精品国产电影| 日韩成人伦理影院| 精品久久久久久电影网| 国产 精品1| 国产欧美亚洲国产| 亚州av有码| 日本黄色片子视频| 欧美日韩综合久久久久久| 欧美精品国产亚洲| 亚洲天堂av无毛| 亚洲综合色惰| 久久精品久久久久久久性| 肉色欧美久久久久久久蜜桃| 色视频在线一区二区三区| 成人美女网站在线观看视频| 亚洲精品日韩av片在线观看| 免费观看的影片在线观看| 欧美高清成人免费视频www| 又大又黄又爽视频免费| 免费黄色在线免费观看| 在线天堂最新版资源| 精品少妇久久久久久888优播| 成年女人在线观看亚洲视频| 亚洲美女视频黄频| 国产精品熟女久久久久浪| 十八禁网站网址无遮挡 | 久久精品国产亚洲网站| 国产精品久久久久久久电影| 9色porny在线观看| 久久精品熟女亚洲av麻豆精品| 日产精品乱码卡一卡2卡三| 人妻人人澡人人爽人人| 免费大片黄手机在线观看| 91久久精品国产一区二区三区| 老女人水多毛片| 日韩中字成人| 亚洲精品第二区| 成人亚洲欧美一区二区av| 狂野欧美激情性xxxx在线观看| 中文字幕人妻熟人妻熟丝袜美| 日本色播在线视频| 亚洲欧美一区二区三区国产| 伦理电影大哥的女人| 免费av中文字幕在线| 精品亚洲成a人片在线观看| av免费观看日本| 国产精品国产三级国产专区5o| 在线看a的网站| 99九九在线精品视频 | 国产成人精品婷婷| 国产深夜福利视频在线观看| 国产亚洲5aaaaa淫片| 午夜福利影视在线免费观看| 日韩熟女老妇一区二区性免费视频| 亚洲av电影在线观看一区二区三区| 在线观看一区二区三区激情| 五月伊人婷婷丁香| 天堂俺去俺来也www色官网| 91午夜精品亚洲一区二区三区| 婷婷色综合www| 免费大片黄手机在线观看| a级毛片免费高清观看在线播放| 丝袜喷水一区| 精品国产乱码久久久久久小说| 99热6这里只有精品| 男女免费视频国产| videossex国产| 黄色配什么色好看| 亚洲国产精品一区三区| 中文字幕久久专区| 七月丁香在线播放| 18禁动态无遮挡网站| 免费人妻精品一区二区三区视频| 精品卡一卡二卡四卡免费| 欧美激情国产日韩精品一区| 91久久精品国产一区二区三区| 在线观看免费视频网站a站| 亚州av有码| 狠狠精品人妻久久久久久综合| 日韩一本色道免费dvd| 夜夜骑夜夜射夜夜干| 亚洲国产日韩一区二区| 激情五月婷婷亚洲| 黄色欧美视频在线观看| 草草在线视频免费看| 男人狂女人下面高潮的视频| 久久热精品热| 亚洲av免费高清在线观看| 久久久久久久久久久免费av| 色视频在线一区二区三区| kizo精华| 日韩av免费高清视频| 又黄又爽又刺激的免费视频.| 亚洲怡红院男人天堂| 亚洲美女搞黄在线观看| 国产一区二区在线观看日韩| 建设人人有责人人尽责人人享有的| 久久99一区二区三区| 好男人视频免费观看在线| 精品少妇黑人巨大在线播放| 丰满乱子伦码专区| 日本欧美视频一区| 桃花免费在线播放| 亚洲精品乱码久久久v下载方式| 最新中文字幕久久久久| 午夜精品国产一区二区电影| 国产精品偷伦视频观看了| 欧美精品亚洲一区二区| 九九在线视频观看精品| 黄色怎么调成土黄色| 亚洲,欧美,日韩| 色视频在线一区二区三区| 尾随美女入室| 丰满迷人的少妇在线观看| 国产精品不卡视频一区二区| 久久久精品94久久精品| 亚洲国产毛片av蜜桃av| 亚洲怡红院男人天堂| 男人狂女人下面高潮的视频| 国产精品福利在线免费观看| 欧美xxxx性猛交bbbb| 久久热精品热| 午夜激情福利司机影院| 日韩av不卡免费在线播放| 少妇 在线观看| 国产极品天堂在线| 日本黄色片子视频| 少妇被粗大猛烈的视频| 99热这里只有精品一区| 一区二区av电影网| 久久久久久人妻| 99热国产这里只有精品6| 国产精品久久久久久久久免| 69精品国产乱码久久久| 中文字幕制服av| 观看av在线不卡| 蜜桃久久精品国产亚洲av| 亚洲欧洲国产日韩| 又爽又黄a免费视频| 亚洲自偷自拍三级| 99久久精品一区二区三区| 丰满乱子伦码专区| 亚洲国产精品999| 亚洲,一卡二卡三卡| 免费av不卡在线播放| 午夜老司机福利剧场| 人人妻人人澡人人爽人人夜夜| 美女内射精品一级片tv| 色网站视频免费| 2021少妇久久久久久久久久久| 国产精品一区二区性色av| 欧美xxxx性猛交bbbb| 亚洲av国产av综合av卡| 建设人人有责人人尽责人人享有的| 中文欧美无线码| 香蕉精品网在线| 精品国产一区二区久久| 91成人精品电影| 国产欧美另类精品又又久久亚洲欧美| 中文字幕免费在线视频6| 亚洲欧美成人综合另类久久久| 国产欧美日韩一区二区三区在线 | 永久网站在线| 久久精品国产亚洲网站| 欧美bdsm另类| 亚洲第一av免费看| 九色成人免费人妻av|