• <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
    久久久精品国产亚洲av高清涩受| 欧美色视频一区免费| 亚洲色图 男人天堂 中文字幕| 在线观看午夜福利视频| 亚洲精华国产精华精| 中文字幕人妻熟女乱码| 国产aⅴ精品一区二区三区波| www日本在线高清视频| 99国产极品粉嫩在线观看| 中文字幕人妻丝袜一区二区| 男人的好看免费观看在线视频 | 国产精华一区二区三区| 国产成人欧美在线观看| tocl精华| 国产私拍福利视频在线观看| 免费少妇av软件| 国产av在哪里看| avwww免费| 99国产综合亚洲精品| 99国产精品一区二区蜜桃av| 免费人成视频x8x8入口观看| 久久久久国内视频| e午夜精品久久久久久久| 黄色丝袜av网址大全| 女人爽到高潮嗷嗷叫在线视频| 国产精品98久久久久久宅男小说| 此物有八面人人有两片| 人人妻,人人澡人人爽秒播| 亚洲av成人一区二区三| 欧美 亚洲 国产 日韩一| 亚洲三区欧美一区| 91成人精品电影| 欧美午夜高清在线| 精品久久久久久久毛片微露脸| 18美女黄网站色大片免费观看| aaaaa片日本免费| 91大片在线观看| 国产成人啪精品午夜网站| 91精品国产国语对白视频| 亚洲中文日韩欧美视频| 国产激情欧美一区二区| 成年版毛片免费区| 757午夜福利合集在线观看| 亚洲专区中文字幕在线| 91麻豆精品激情在线观看国产| 大陆偷拍与自拍| 久久九九热精品免费| 色av中文字幕| 欧美最黄视频在线播放免费| 91九色精品人成在线观看| 欧美日韩精品网址| 999久久久精品免费观看国产| 日韩欧美国产一区二区入口| 美女午夜性视频免费| 免费高清视频大片| а√天堂www在线а√下载| 99在线视频只有这里精品首页| 黑人欧美特级aaaaaa片| 午夜精品国产一区二区电影| 国产亚洲欧美在线一区二区| 欧美色欧美亚洲另类二区 | 亚洲成av片中文字幕在线观看| 一本久久中文字幕| 成人三级做爰电影| 最新美女视频免费是黄的| 中文字幕精品免费在线观看视频| 午夜福利高清视频| 国产欧美日韩一区二区三| 99精品欧美一区二区三区四区| 亚洲av成人av| 亚洲av第一区精品v没综合| 欧美成狂野欧美在线观看| 狂野欧美激情性xxxx| 脱女人内裤的视频| 成年版毛片免费区| 亚洲av成人不卡在线观看播放网| av视频免费观看在线观看| 午夜福利免费观看在线| 久久精品91无色码中文字幕| 国产麻豆69| 亚洲欧洲精品一区二区精品久久久| 国产亚洲精品一区二区www| 精品一区二区三区视频在线观看免费| 99re在线观看精品视频| 中文字幕另类日韩欧美亚洲嫩草| 国产av一区在线观看免费| 国产视频一区二区在线看| 国产精品,欧美在线| 香蕉国产在线看| 中文字幕久久专区| 亚洲全国av大片| 免费av毛片视频| 黄色视频不卡| 久久天躁狠狠躁夜夜2o2o| 日本vs欧美在线观看视频| 国内久久婷婷六月综合欲色啪| 精品卡一卡二卡四卡免费| 亚洲精品在线美女| 村上凉子中文字幕在线| 日本精品一区二区三区蜜桃| 亚洲av第一区精品v没综合| 在线观看免费午夜福利视频| 亚洲国产中文字幕在线视频| 波多野结衣巨乳人妻| 黄色成人免费大全| 国产男靠女视频免费网站| 中文字幕人妻丝袜一区二区| 满18在线观看网站| 国产精品美女特级片免费视频播放器 | 亚洲激情在线av| 在线免费观看的www视频| 久久中文字幕人妻熟女| √禁漫天堂资源中文www| 在线观看免费视频日本深夜| 亚洲人成77777在线视频| 欧美黄色淫秽网站| 久久精品91蜜桃| 日韩三级视频一区二区三区| 成年女人毛片免费观看观看9| 国产乱人伦免费视频| 久久中文看片网| 国产熟女午夜一区二区三区| 久久精品亚洲熟妇少妇任你| 深夜精品福利| 国产极品粉嫩免费观看在线| 亚洲一区中文字幕在线| 国产精品,欧美在线| 在线观看日韩欧美| 色婷婷久久久亚洲欧美| 99精品欧美一区二区三区四区| 人人妻人人澡人人看| 一本综合久久免费| 久久中文看片网| 久99久视频精品免费| 亚洲一区二区三区不卡视频| 亚洲成人国产一区在线观看| 日本vs欧美在线观看视频| www日本在线高清视频| 色综合亚洲欧美另类图片| 男人操女人黄网站| 免费在线观看视频国产中文字幕亚洲| 在线免费观看的www视频| 国产国语露脸激情在线看| 亚洲专区字幕在线| 国产成年人精品一区二区| 国产成人一区二区三区免费视频网站| 777久久人妻少妇嫩草av网站| 18禁国产床啪视频网站| 欧美日韩中文字幕国产精品一区二区三区 | 亚洲久久久国产精品| 亚洲国产精品成人综合色| 97超级碰碰碰精品色视频在线观看| 日日夜夜操网爽| 激情视频va一区二区三区| av视频免费观看在线观看| 91av网站免费观看| 人人妻人人澡人人看| 国产成人免费无遮挡视频| 亚洲成人免费电影在线观看| 亚洲国产看品久久| 精品免费久久久久久久清纯| 亚洲一区高清亚洲精品| 精品高清国产在线一区| 动漫黄色视频在线观看| 国产91精品成人一区二区三区| 母亲3免费完整高清在线观看| 99久久综合精品五月天人人| 亚洲av片天天在线观看| 成人av一区二区三区在线看| 国产免费av片在线观看野外av| 两个人视频免费观看高清| 97超级碰碰碰精品色视频在线观看| 在线国产一区二区在线| 午夜福利高清视频| av有码第一页| 欧美丝袜亚洲另类 | 亚洲精品久久国产高清桃花| 精品高清国产在线一区| 色综合亚洲欧美另类图片| 一进一出抽搐动态| 国产精品亚洲一级av第二区| 成在线人永久免费视频| 婷婷丁香在线五月| 国产欧美日韩一区二区三区在线| 亚洲国产欧美一区二区综合| 在线国产一区二区在线| 国产精品免费一区二区三区在线| 亚洲成a人片在线一区二区| 久热爱精品视频在线9| 99在线人妻在线中文字幕| 夜夜夜夜夜久久久久| 少妇 在线观看| 国产精品日韩av在线免费观看 | 男女床上黄色一级片免费看| 精品久久久久久成人av| 99久久久亚洲精品蜜臀av| а√天堂www在线а√下载| 淫妇啪啪啪对白视频| 91成人精品电影| 国产成人影院久久av| 男男h啪啪无遮挡| av欧美777| 别揉我奶头~嗯~啊~动态视频| 天天一区二区日本电影三级 | 精品久久久精品久久久| 99国产精品免费福利视频| 欧美中文日本在线观看视频| 久久久久久人人人人人| 日韩高清综合在线| 欧美国产精品va在线观看不卡| 99热只有精品国产| 人妻久久中文字幕网| 身体一侧抽搐| 一级片免费观看大全| 黄色毛片三级朝国网站| 18禁国产床啪视频网站| 一卡2卡三卡四卡精品乱码亚洲| 色哟哟哟哟哟哟| 亚洲精品中文字幕一二三四区| 怎么达到女性高潮| 午夜免费观看网址| av片东京热男人的天堂| 在线十欧美十亚洲十日本专区| 麻豆久久精品国产亚洲av| 欧美激情久久久久久爽电影 | 国产亚洲欧美在线一区二区| 成年人黄色毛片网站| 国内久久婷婷六月综合欲色啪| 97人妻天天添夜夜摸| 国产精品香港三级国产av潘金莲| 极品人妻少妇av视频| 黑人巨大精品欧美一区二区mp4| 嫩草影院精品99| 国产日韩一区二区三区精品不卡| av在线播放免费不卡| 一进一出抽搐动态| netflix在线观看网站| 久久精品91蜜桃| 咕卡用的链子| 久久久久久免费高清国产稀缺| 精品乱码久久久久久99久播| 美女扒开内裤让男人捅视频| 亚洲成人精品中文字幕电影| 操出白浆在线播放| 亚洲一区二区三区色噜噜| 日本a在线网址| 可以在线观看毛片的网站| 国产高清激情床上av| 精品国产一区二区久久| 老汉色∧v一级毛片| 啪啪无遮挡十八禁网站| 国产免费男女视频| 国产一区二区三区在线臀色熟女| 国产激情欧美一区二区| www.999成人在线观看| 成人三级做爰电影| 亚洲国产精品sss在线观看| 波多野结衣一区麻豆| 午夜激情av网站| 久久久精品欧美日韩精品| 黑人欧美特级aaaaaa片| 免费av毛片视频| 精品久久蜜臀av无| 国产亚洲精品第一综合不卡| 久久久精品国产亚洲av高清涩受| 日韩免费av在线播放| 一边摸一边抽搐一进一小说| 欧美精品亚洲一区二区| 最近最新中文字幕大全免费视频| 中文字幕高清在线视频| 亚洲熟妇熟女久久| 国产在线精品亚洲第一网站| 他把我摸到了高潮在线观看| 美女扒开内裤让男人捅视频| 99re在线观看精品视频| 精品久久蜜臀av无| 精品久久久久久,| bbb黄色大片| 午夜两性在线视频| 丝袜人妻中文字幕| 又紧又爽又黄一区二区| 亚洲五月天丁香| 老司机深夜福利视频在线观看| 黄色视频不卡| 国产激情欧美一区二区| 亚洲精品美女久久av网站| 亚洲精品国产色婷婷电影| 国产成人系列免费观看| bbb黄色大片| 欧美日韩黄片免| 中国美女看黄片| 在线十欧美十亚洲十日本专区| 亚洲人成77777在线视频| 看免费av毛片| 亚洲精品美女久久av网站| 最新美女视频免费是黄的| 免费av毛片视频| 久久草成人影院| 乱人伦中国视频| 男人操女人黄网站| 精品福利观看| 久久人妻熟女aⅴ| 色综合婷婷激情| 免费高清视频大片| 精品国内亚洲2022精品成人| 天堂动漫精品| 中文字幕久久专区| 变态另类丝袜制服| 亚洲 欧美一区二区三区| 亚洲,欧美精品.| 国产亚洲欧美精品永久| 亚洲美女黄片视频| 首页视频小说图片口味搜索| av天堂在线播放| av福利片在线| 亚洲专区中文字幕在线| 欧美色视频一区免费| 亚洲国产高清在线一区二区三 | 亚洲七黄色美女视频| 欧美乱码精品一区二区三区| 免费看美女性在线毛片视频| 搡老熟女国产l中国老女人| 视频区欧美日本亚洲| 夜夜躁狠狠躁天天躁| 他把我摸到了高潮在线观看| 日韩欧美国产在线观看| 1024香蕉在线观看| 成人18禁高潮啪啪吃奶动态图| 男女做爰动态图高潮gif福利片 | 无遮挡黄片免费观看| 很黄的视频免费| 伊人久久大香线蕉亚洲五| 亚洲一卡2卡3卡4卡5卡精品中文| 少妇粗大呻吟视频| 欧美日韩一级在线毛片| 老司机靠b影院| 国产亚洲av嫩草精品影院| 亚洲欧美激情在线| 高清毛片免费观看视频网站| 精品国产超薄肉色丝袜足j| www.www免费av| 在线观看66精品国产| 亚洲国产欧美一区二区综合| 亚洲五月色婷婷综合| 啪啪无遮挡十八禁网站| 少妇 在线观看| av超薄肉色丝袜交足视频| 久久久久国产精品人妻aⅴ院| 51午夜福利影视在线观看| 又黄又粗又硬又大视频| 满18在线观看网站| 老司机在亚洲福利影院| 久久精品成人免费网站| 精品国产亚洲在线| 国产欧美日韩一区二区精品| 天天躁夜夜躁狠狠躁躁| 免费不卡黄色视频| 美女大奶头视频| 十分钟在线观看高清视频www| 欧美日本亚洲视频在线播放| 欧美中文日本在线观看视频| 日韩 欧美 亚洲 中文字幕| 亚洲欧美精品综合久久99| 国产在线观看jvid| 国产成人啪精品午夜网站| 国产主播在线观看一区二区| 伦理电影免费视频| 欧美av亚洲av综合av国产av| 两性夫妻黄色片| 丝袜美足系列| 欧美午夜高清在线| 久久久久久亚洲精品国产蜜桃av| 老汉色∧v一级毛片| 给我免费播放毛片高清在线观看| 一区二区三区激情视频| 好男人在线观看高清免费视频 | 变态另类丝袜制服| 在线国产一区二区在线| 中文字幕另类日韩欧美亚洲嫩草| 国产精品国产高清国产av| 99国产精品一区二区蜜桃av| 欧美日韩精品网址| 午夜免费激情av| 成人特级黄色片久久久久久久| 成人手机av| 亚洲精品国产区一区二| 三级毛片av免费| 乱人伦中国视频| 夜夜夜夜夜久久久久| 波多野结衣av一区二区av| 国产成人欧美在线观看| 婷婷六月久久综合丁香| 最新美女视频免费是黄的| 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 | 老熟妇乱子伦视频在线观看| 国产精品久久视频播放| 激情视频va一区二区三区| 制服诱惑二区| 男人舔女人的私密视频| 在线av久久热| 欧美黑人精品巨大| 黑人欧美特级aaaaaa片| 琪琪午夜伦伦电影理论片6080| 91精品国产国语对白视频| or卡值多少钱| 又黄又爽又免费观看的视频| 在线观看午夜福利视频| 成人av一区二区三区在线看| 97超级碰碰碰精品色视频在线观看| 亚洲美女黄片视频| 日韩欧美在线二视频| 久久天堂一区二区三区四区| 国产99久久九九免费精品| 国产精品日韩av在线免费观看 | 麻豆av在线久日| 亚洲三区欧美一区| 一进一出抽搐动态| 国产成人系列免费观看| 伦理电影免费视频| 国产黄a三级三级三级人| 精品久久久久久久人妻蜜臀av | 首页视频小说图片口味搜索| ponron亚洲| 99国产精品一区二区蜜桃av| 国产1区2区3区精品| 日本 av在线| e午夜精品久久久久久久| 夜夜看夜夜爽夜夜摸| 免费搜索国产男女视频| 亚洲专区字幕在线| 国产成人精品在线电影| 国产在线精品亚洲第一网站| 久久精品aⅴ一区二区三区四区| av片东京热男人的天堂| 午夜福利高清视频| 久久性视频一级片| 国产精品1区2区在线观看.| а√天堂www在线а√下载| 在线国产一区二区在线| 久久久水蜜桃国产精品网| 在线播放国产精品三级| 国产一区在线观看成人免费| 老司机午夜十八禁免费视频| 亚洲精品国产区一区二| x7x7x7水蜜桃| 美女午夜性视频免费| 亚洲精品国产色婷婷电影| 性欧美人与动物交配| 嫩草影视91久久| www国产在线视频色| 欧美黑人精品巨大| 脱女人内裤的视频| 免费女性裸体啪啪无遮挡网站| 免费无遮挡裸体视频| 国产精华一区二区三区| 久久欧美精品欧美久久欧美| 亚洲av电影不卡..在线观看| 欧美激情 高清一区二区三区| 看免费av毛片| 国产亚洲精品久久久久5区| 9热在线视频观看99| 午夜福利视频1000在线观看 | 91麻豆av在线| 欧美激情 高清一区二区三区| 一级毛片精品| 亚洲av第一区精品v没综合| 亚洲五月色婷婷综合| 精品熟女少妇八av免费久了| 免费少妇av软件| 久久午夜综合久久蜜桃| videosex国产| 9热在线视频观看99| 成年人黄色毛片网站| av天堂在线播放| 午夜久久久在线观看| 国产精品综合久久久久久久免费 | 国产高清视频在线播放一区| 国产成人欧美在线观看| 欧美激情 高清一区二区三区| 精品久久久久久久人妻蜜臀av | 亚洲av成人一区二区三| 久久久久久久久久久久大奶| 午夜视频精品福利| 成年女人毛片免费观看观看9| 亚洲视频免费观看视频| 在线国产一区二区在线| 亚洲国产欧美一区二区综合| 欧美一区二区精品小视频在线| www国产在线视频色| 国产私拍福利视频在线观看| 亚洲av成人一区二区三| 久久精品91无色码中文字幕| 伊人久久大香线蕉亚洲五| 人妻久久中文字幕网| 亚洲久久久国产精品| 国产亚洲精品一区二区www| a级毛片在线看网站| 国产欧美日韩一区二区三区在线| 人妻久久中文字幕网| 中文字幕人妻丝袜一区二区| 国产成人精品久久二区二区免费| av网站免费在线观看视频| 老司机靠b影院| 中文字幕色久视频| 美女高潮喷水抽搐中文字幕| 精品国内亚洲2022精品成人| 亚洲av成人av| 亚洲第一电影网av| 国产私拍福利视频在线观看| 人人澡人人妻人| 久久久国产精品麻豆| 9热在线视频观看99| 亚洲一卡2卡3卡4卡5卡精品中文| 啦啦啦免费观看视频1| 亚洲欧美日韩高清在线视频| 精品免费久久久久久久清纯| 欧美日韩黄片免| 女人被狂操c到高潮| 黄片小视频在线播放| 日日摸夜夜添夜夜添小说| 99国产精品一区二区蜜桃av| 一进一出抽搐gif免费好疼| 婷婷丁香在线五月| 久久久水蜜桃国产精品网| 亚洲国产精品合色在线| 中文字幕人妻熟女乱码| 午夜a级毛片| 日日摸夜夜添夜夜添小说| 日韩av在线大香蕉| 亚洲专区中文字幕在线| 日本三级黄在线观看| 琪琪午夜伦伦电影理论片6080| 免费无遮挡裸体视频| 在线观看免费午夜福利视频| 在线观看免费视频网站a站| 窝窝影院91人妻| cao死你这个sao货| 亚洲 国产 在线| 色尼玛亚洲综合影院| 国产精品爽爽va在线观看网站 | 97碰自拍视频| 级片在线观看| 精品国产乱码久久久久久男人| 精品国产超薄肉色丝袜足j| 国产单亲对白刺激| 天堂√8在线中文| 欧美国产精品va在线观看不卡| 国产欧美日韩一区二区精品| 国产不卡一卡二| 国产1区2区3区精品| 国产精品 国内视频| 大型黄色视频在线免费观看| 在线播放国产精品三级| 国产精品av久久久久免费| 亚洲熟女毛片儿| 亚洲免费av在线视频| 99精品欧美一区二区三区四区| 欧美国产精品va在线观看不卡| 午夜福利视频1000在线观看 | 女警被强在线播放| 久久亚洲精品不卡| 国产亚洲精品综合一区在线观看 | 男人舔女人下体高潮全视频| 国产成人免费无遮挡视频| 中亚洲国语对白在线视频| 亚洲少妇的诱惑av| 久久久久久久久免费视频了| 亚洲av熟女| 自线自在国产av| 级片在线观看| 国产伦人伦偷精品视频| 午夜视频精品福利| 亚洲第一电影网av| 中文字幕另类日韩欧美亚洲嫩草| 悠悠久久av| 日本在线视频免费播放| 丝袜美腿诱惑在线| 亚洲国产欧美日韩在线播放| 亚洲国产中文字幕在线视频| 成在线人永久免费视频| 亚洲精品中文字幕在线视频| 国产亚洲欧美精品永久| 变态另类丝袜制服| 一级黄色大片毛片| 女人精品久久久久毛片| 免费高清视频大片| 一二三四在线观看免费中文在| 国产亚洲欧美98| 十八禁网站免费在线| 国语自产精品视频在线第100页| 性少妇av在线| 免费高清视频大片| av免费在线观看网站| 亚洲一区高清亚洲精品| 黄色视频不卡| 一边摸一边做爽爽视频免费| 婷婷丁香在线五月| 国产精品爽爽va在线观看网站 | 午夜福利18| 日日夜夜操网爽| 日韩高清综合在线| 日本欧美视频一区| 中文字幕高清在线视频| 午夜福利视频1000在线观看 | 亚洲 欧美一区二区三区| 18美女黄网站色大片免费观看| 亚洲国产毛片av蜜桃av| 91精品国产国语对白视频| 亚洲五月婷婷丁香| 国产不卡一卡二| 女人爽到高潮嗷嗷叫在线视频| 欧美人与性动交α欧美精品济南到| 黄色女人牲交| 欧美 亚洲 国产 日韩一|