• <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 ?

    Collaborative Filtering Recommendation Algorithm Based on Multi-Relationship Social Network

    2019-08-13 05:54:54ShengBinGengxinSunNingCaoJinmingQiuhiyonghengGuohuaYangHongyanhaoMengJiangandLinaXu
    Computers Materials&Continua 2019年8期

    Sheng Bin,Gengxin SunNing Cao,Jinming Qiu,Ζhiyong Ζheng,Guohua Yang,Hongyan Ζhao,Meng Jiang and Lina Xu

    Abstract: Recommendation system is one of the most common applications in the field of big data.The traditional collaborative filtering recommendation algorithm is directly based on user-item rating matrix.However,when there are huge amounts of user and commodities data,the efficiency of the algorithm will be significantly reduced.Aiming at the problem,a collaborative filtering recommendation algorithm based on multirelational social networks is proposed.The algorithm divides the multi-relational social networks based on the multi-subnet complex network model into communities by using information dissemination method,which divides the users with similar degree into a community.Then the user-item rating matrix is constructed by choosing the k-nearest neighbor set of users within the community,in this case,the collaborative filtering algorithm is used for recommendation.Thus,the execution efficiency of the algorithm is improved without reducing the accuracy of recommendation.

    Keywords: Big data,recommendation algorithm,social network,recommendation efficiency.

    1 Introduction

    As an effective means to solve the problem of information overload [Liu (2016)],recommendation system is also one of the most common applications in the field of large data.According to the user’s history information,hobby and other characteristics,the recommendation system uses knowledge detection technology to filter users’ interesting information and products,so as to realize personalized recommendation.At present,the most commonly used recommendation system technology is recommendation algorithm based on collaborative filtering [Dong and Ke (2018)].The kind of algorithm calculates the similarity of users or commodities based on the user-item rating matrix,and then recommends the commodities that users may need.In addition,the real meaning of big data is not big but multiple sources [Zhang,Li and Tang (2017)].How to find out the inner relationship of multi-source data in the process of fusion through scientific analysis,and extract a unified,richer and more valuable information than a single data source,which is the real significance of using large data.Therefore,integrating the social relationship among users into the recommendation system can calculate the similarity of users more accurately.To a certain extent,it also solves the problems of sparse data and cold start [Margaris,Vassilakis and Georgiadis (2018)] in traditional recommendation algorithm based on collaborative filtering.Actually,there are many relationships among users of social networks,each of which has a different impact on recommendation.Therefore,simply introducing a certain social relationship will inevitably affect the accuracy of recommendation results.

    In view of the above-mentioned facts,a collaborative filtering algorithm based on multi relational social networks is proposed in the article.Firstly,the multi-relational social network is constructed by the multi-subnet complex network model [Sun and Bin (2018)].Then,the algorithm divides the multi-relational social networks based on the multisubnet complex network model into communities by using information dissemination method,which divides the users with similar degree into a community.Next,the knearest neighbor set is composed of theKusers with the highest similarity with the recommended users in the community,which is used to construct the user-item rating matrix.Finally,the collaborative filtering algorithm is used for recommendation.Because of the sparse relationship between community organizations,the similarity between members of different community organizations is weak.Therefore,it is not necessary to analyze the user-item rating matrix constructed by the whole data set in the recommendation process,thus improving the execution efficiency of the algorithm without reducing the accuracy of recommendation.

    2 Related works

    Collaborative filtering technology is mainly divided into memory-based [Zhao (2013)]collaborative filtering and model-based [Jiang,Qian and Shen (2015)] collaborative filtering.Both of the two methods are based on user-item rating matrix.However,with the increase of the number of users and commodities,the user-item rating matrix will become larger and larger,resulting in the performance of collaborative filtering algorithm is decreasing.In order to improve the efficiency of recommendation algorithm,Roh et al.[Roh,Oh and Han (2003)] use hierarchical clustering algorithm to cluster users and commodities,then it is recommended based on the weighted summation of root nodes to leaf nodes.Zhang et al.[Zhang,Ding and Jiang (2005)] used fuzzy clustering method to classify users who like similar items and items that liked by the same user before recommending.Santhini et al.[Santhini,Balamurugan and Govindaraj (2015)] put forward the basic idea of collaborative filtering recommendation for big data by clustering method,which improves the efficiency of recommendation algorithm to a certain extent.However,clustering technology relies on the user-item rating matrix as the original data source to reflect personal preferences.It does not take into account the social relationships such as friendship and trust between users,nor does it consider the existence of implicit relationships between the scored commodities or items.Ignoring these important relationships will inevitably lead to the distortion of recommendation results.Accordingly,the introduction of social networks into the recommendation system to improve the rating matrix of the social recommendation system [Li,Li,Chen et al.(2018)] came into being.Chang et al.[Chang,Diaz and Hung (2015)] fused the trust relationship in social networks in the recommendation algorithm,and predicted the score by using trust between users as an alternative to the vacancy values in the user-item matrix.Pan et al.[Pan,Zhang and Wang (2016)] alleviated the problem of the sparse data and cold start by using the trust relationship in social network,which proved that social network can greatly improve the performance of recommendation system.As a typical complex network [Sun and Bin (2017)],social networks have the characteristics of community structure,which can divide the network members into groups with close relationship within the group and sparse relationship between groups [Jiang,Sun and Bin(2018)].Wang et al.[Wang,Li and Zhang (2016)] used weighted label dissemination algorithm to divide the citation cooperative network into communities,and then used recommendation algorithm to recommend academic papers in the same community.Zhang et al.[Zhang,Zhen and Zhang (2018)] mapped the user-item rating matrix to the user similarity network and combined with the community in the network to recommend the sorted Top-n commodities to users,which improved the accuracy and efficiency of the recommendation.Although the existing recommendation algorithms based on community partitioning in complex networks have made some progress in the accuracy and efficiency of recommendation,they are all based on single-relational social networks,in which only one relationship exists.In fact,social network is actually a multirelationship network,for example,there are user trust,similar interests and other relationships among users of social network.These explicit or implicit relationships determine the interaction characteristics between users together,which have different impact on the recommendation results.Therefore,the social network introduced into the recommendation system should be a multi-relational complex network with heterogeneous nodes,while the traditional complex network model cannot describe these complex relationships well,nor can it describe the interaction between heterogeneous nodes.Based on that,the article constructs a multi-relationship social network based on the multi-subnet complex network model which can describe the multi-relationship between different types of individuals,and divides the social structure of the network by information dissemination method.When predicting users’ rating,the similarity of users’rating is calculated according to the degree of users’ preference,and then the k-nearest neighbor set is selected for the recommended users in the community by combining the revised similarity of users’ rating and the similarity of user social interaction,thus greatly reducing the computational complexity of the recommendation algorithm.

    3 Construction of multi-relationship social network based on multi-subnet complex network model

    In the recommendation system,its elements can be divided into two sets:user andcommodity.Suppose there are users and commodities,the user set can be expressed as,and the commodity setcan be expressed as.Based on that,the user's rating matrixfor commodities is composed ofelements,which is represented by,whererepresents the userrating for the product,and the user-item rating matrix is shown in Fig.1.

    Figure1:User-item rating matrix

    In reality,the recommendation process is usually accompanied by the integration of social information,such as the user's behavior of inquiring about friends’ shopping opinions,the sharing of shopping information between friends and so on.Actually,if there is a trust relationship between users,the similarity of their interest preferences will be significantly higher than that of strangers [Guo,Li and Zhao (2016)],that is,if one makes a high rating on a commodity,then another user is likely to show a high rating on the commodity.

    The trust relationship in social networks can be represented by the trust-rating matrix,which is expressed as,whererepresents the userrating on user,it reflects the degree of trust between users.The trust-rating matrix is shown in Fig.2.

    Figure2:Trust-rating Matrix

    In order to effectively integrate the social feature,the user-item rating network and the user trust relationship network are combined into multi-relational social network based on the multi-subnet complex network model.

    3.1 Construction of user-item rating network and user trust relationship network

    The multi-subnet complex networkmodel is a quaternion satisfying the followingconditions:,where denotes the set of nodes in the network,is the number of nodes in the set;is the set of edges between nodes;and,denotes the set of interaction relations among nodes,is one of them,n is the total number of interaction relations,and if,the network is a multi-relationship network;denotes that the mapping of the edge setis projected by a functionto find a unique corresponding mapping in,which represents the type of relationship on edge.

    Single-relationship complex network is a special case of multi-subnet complex network.By using multi-subnet complex network model,user-item rating network R_G and user trust relationship network T_G are constructedbasedon user-item rating matrix and trust-rating matrix respectively,and thresholdand are set.If the user rating ofcommodityis greater than or equal to the threshold ,it is considered that the commodity satisfies the user’s preference..Similarly,if the usertrust rating of another useris greater than or equal to the threshold,it is considered that there is a trust relationship between users.In the user-item rating network R_G,each nodesymbolizes a user or a commodity,andindicates the preference relationship between users and commodities.Similarly,in the user trust relationshipnetwork T_G,each node represents a user,andrepresents the trust relationship between users.

    The network is constructed according to the user-item rating matrix and trust-ratingmatrix given in Fig.1 and Fig.2,where the thresholdand are both set to 2,then the network topology is constructed as shown in Fig.3 and Fig.4.

    Figure3:User-item rating network

    Figure4:User trust relationship network

    3.2 Construction of multi-relationship complex networks

    Multi-subnet complex network can compose two or more complex networks by subnet loading operation.In the process of loading,the network is mapped to vector space,and the relationship between nodes is mapped to multidimensional vector in vector space.Therefore,the process of subnet loading can be transformed into the base transformation of vector space,and the change of the relationship between the related nodes is accomplished by the correlation operation of the space vector.Due to the inheritance of the nature of each subnet,the dimension of the relationship vector space of the constructed multi-relational complex network increases correspondingly.

    The user-item rating network and user trust relationship network are loaded into a multirelational complex network by subnet loading operation.The loading process of subnet is as follows:Suppose there are vector composite networksand,where,,and the set of relation is,is the base network.Loadingtounder load mappingand relation strength mapping,which generates a new composite network,where,is called the set of loading relations,represents the external edge,andrepresents the boundary node.

    Compared with the traditional complex network,the topological properties of the composite network after loading have changed differently.Under the circumstances,the edges associated with the nodes contain all the relationships of the subnet.Therefore,the mixing weight and mixing degree of the nodes in the composite network are defined as follows:Mixing weight of node:if,the ratio coefficient of relationship strength expressed as,and the mixing weightof nodeis expressed as:

    The user-item rating network R_G of Fig.3 and the user trust relationship network T_G of Fig.4 are loaded into a composite recommendation network,where the relationship strength parametersandof relation and are set to 1,the edge weight set of subnet R_G is,the edge weight set of subnet T_G is,and the composite network of sub-network is constructed as shown in Fig.5:

    Figure5:Composited recommendation network

    4 Community structure detection in multi-relationship social networks

    4.1 The process of information propagation on complex networks

    Initially,all nodes in the network do not carry information.At the beginning of the first propagation,a node is set as the source node and given its non-empty information quantity,and the set of neighbor nodes of the node is set as .At this time,the information in the network is distributed as,source nodes transmit information to nodes inby connecting edges.After receiving information,the neighbor nodes will continue to transmit information to their neighbor nodes of the set.However,the node as the receiver does not receive the information from the source node completely,but receives it with a certain probability.In the second propagation,represents the set of nodes carrying information,and the nodes in the setcan either transmit information to neighboring nodes or receive information from their neighboring nodes.Because the receiving node cannot receive all the foreign information,when the network scale is large,it only relies on a single dissemination,and some nodes far from the source node receive less information,which cannot adequately express the impact of the source node on these nodes.Accordingly,the distribution of information carried by all nodes in the network can represent the influence of source nodes on the information dissemination of thewhole network only after the process iterates fortimes.Thus,the similarity between nodes can be calculated according to the distribution vectors of information of the network when each node acts as the source node.

    The elements in the adjacent mixing weight matrix are the weighted summation of the strength of the relationships between nodeand,wheredenotes the proportional coefficient of the strength of relationbetween nodeand,anddenotes the strength of relation.

    On the basis of adjacency matrix,the mixing weight of nodecan also be expressed as:

    In turn,the nodes of the network are selected as the source nodes and given 1 unitofinformation,and the information quantity of the remaining nodes is 0.Taking node as an example,assuming that nodeis the source node,its carrying information quantity is 1,and the vector of nodeis initialized to.The first process is that the node transmits information to its neighbors,which can be expressed asby mathematical formula,and the element in vectoris the information quantity carried by each node after the first dissemination;in the second process,the nodes have receiving information transmits the information to its neighboring node,and if the neighboring node has information,it can also receive the information from its neighboring node,thus the process can be expressed as,and the element in vectoris the information quantity carried by each node after the seconddissemination;Similarly,the information after each dissemination canbe expressed as;Before the nextinformation dissemination,assume that,whichmeans that the informationquantity of the source node is always 1 at the beginning ofprocess.In this way,the information distribution matrix which represent the influence of nodes can be obtained by using the information distribution vector of each node aftertimes dissemination.

    Further,the vector of first dissemination is obtained as follows:

    4.2 Community detection algorithm for multi-relationship social networks based on information propagation

    After the nodes of network are represented by vectors,their data structure is suitable to be processed by clustering algorithm,and the similarity between nodes is calculated according to the information dissemination,thus the mature data mining clustering algorithm is applied to the community detection for multi-relational social networks.

    In previous studies,CSDM algorithm [15] was proposed based on classical K-means clustering algorithm to discover the community structure of multi-relational social networks.The biggest disadvantage of K-means algorithm is that it is sensitive to the selection of initial clustering centers,since the initial value of the cluster center will directly affect the final clustering result.

    Generally,the importance of nodes in complex networks is called centrality,which is the basic index of describing the importance of nodes.Degree centrality considers that the influence of nodes is related to the number of neighbor nodes,which means that the more neighbor nodes,the greater the influence of the node in thenetwork.Assuming that the path length of nodeandwith respect to relationis ,and the centrality of nodewith respect to relationis defined as:

    Where the m?olecular part represents the minimum sum of the path lengths of therelationship with the remaining nodes,and the denominator part represents the sum of the path lengths of the nodeand other nodes with respect to the relationship.Specifically,the closer the value of centralityis to 1,the smaller the sum of the path length of the node and the remaining nodes is,the stronger the centrality will be;conversely,the closer the value of centrality is to 0,the weaker the centrality will be.

    Further,the CSDM algorithm chooses the nodes with larger centrality as the initial clustering centers,and then divides the multi-relational social networks into communities according to classicalK-means clustering algorithm.

    5 Collaborative filtering recommendation algorithm based on community structure detection

    Through information dissemination,it solves the problem that the social network composed of users of recommendation system is difficult to calculate similarity due to multi-relationship,and the information quantity carried by each member after dissemination reflects the similarity degree of different nodes in the network.By dividing these users with high similarity into a community,members of the community will share information such as rating preferences and interest preferences.On the one hand,the social preference vector can be obtained from the trust relationship of users contained in the community structure,which represents the user's social preference and can be measured by social similarity;on the other hand,users' interest preferences can be obtained from the rating information contained in the community structure,and the similarity of interest preferences can be determined by extracting interest preference vectors.Finally,the social similarity and interest preference similarity are synthesized in a certain proportion,and a new comprehensive similarity index,recommendation degree,is obtained,which is used to rank and recommend commodities.

    5.1 User social similarity and rating similarity

    Assuming that there is a social relationship between userand user,and the interest betweenandis closer,thus their preference similarity is higher than that of strangers [Wang and Ma (2016)].The kind of social relationship with promoting recommendation effect can be expressed as trust relationship between users.In this way,users in the same community are mapped to a new user trust network,and they are mapped to vectors by information dissemination method,then Pearson similarity is used to measure the social similarity between usersand,the formula is as follows:

    In fact,users’ common rating items and rating preference also reflect users’ interests to a certain extent.In addition to Pearson correlation to measure the user’s rating preference,the user preference degree [11] is defined,and the more consistent the rating preference is,the closer the user's interest preference is,the formula is as follows:

    where the threshold value ofis consistent with the threshold value when constructing user-item rating network.In Section 3.1,the user’s rating below the thresholdindicates that the user does not have a preference relationship for the commodity.While,in this case,it is used to indicate that the user’s rating is positive or negative.If the rating isgreater than the threshold ,it meansthat the rating is positive;and represent theset of commodities scored by user and respectively.When user preference is integrated into the process of calculating the rating similarity,the user’s rating of commodities in the community constitutes a new rating vector,is the Pearson correlation between the rating vector ofand the rating vector of.Thus,user rating similarity based on rating preference is obtained:

    In order to improve the reliability of user similarity,the following formulas are used to integrate user social similarity and user rating similarity:

    5.2 Prediction of rating

    The recommendation system mainly relies on the prediction of the items that are not scored by the recommended users,and finally the top-n with the highest rating will be recommended.It is not necessary to use all users in the community when predicting.Usually,sets of nearest neighbors with the highest similarity of current users are selected.In the process of selecting the nearest neighbor users,if the value ofis too large,additional noise will be broughtto the prediction when the neighbors with lowsimilarity are added;if the value of is too small,it will lead to deviation due to insufficient nearest neighbors used to predict.The prediction rating of userabout commodity is calculated by the following formula:

    6 Experiment and result analysis

    In the process of experiment,the effectiveness and recommendation quality of the proposed algorithm are proved by comparing with different classical recommendation algorithms,and the influence of parameter setting on the recommendation results is analyzed as well.

    6.1 Experimental environment and dataset

    In order to analyze big dataset,a Hadoop cluster platform composed of 5 nodes is built,and the specific configuration of each node machine is Intel (R) Core (TM) i7-7700 CPU@3.60 GHz,8 GB memory,64-bit operating system.

    Epinions dataset [Lauw,Lim and Wang (2012)] commonly used in recommendation system is used in the experiment.The dataset contains users’ rating information and social information.There are 40163 users and 139738 commodities in the dataset,and the rating records reach 664824 items.The user's rating of commodities is expressed by 1 to 5.If there is trust relationship between users,it is expressed by 1;if there is no trust relationship,it is marked as 0;and other statistical information in the dataset is shown in Tab.1:In order to use the algorithm proposed for rating prediction,it is necessary to determine the relationship coefficients and strength between rating relationship and trust relationship.For this reason,users with number 2 and number 10004 are chosen as seed nodes.After eliminating some obvious outliers,150 sample data were selected as the sample set of linear regression analysis data for calculating the proportional coefficients of the tworelationships.The linear regression model was constructed for ratingrelationship and trust relationship in multi-relational social networks as follows:

    Table1:The statistics of Epinions dataset

    The average value of the parameter estimation is obtained as follows:.

    Therefore,when dividing the community of the multi-relationship social network constructed in the experiment,assumed that the ratio coefficient of the two relationships is,the strength of the relationship is 1,the parameter of the K-means clustering algorithm is,and the community partition is shown in Fig.6.

    Figure6:Result of community detection

    6.2 Analysis and comparison of experimental results

    Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) are selected as evaluation indicators to evaluate the accuracy of recommendation results,which evaluate the accuracy of recommendation algorithm by comparing the errors between the real rating and the predicted rating.The smaller the value,the better the recommended effect is.The specific definitions are as follows:

    Mean Absolute Error (MAE):

    Root Mean Squared Error (RMSE):

    The selection of the nearest neighbor numberswill affect the experimental results,and too small or too large of the ?nearest neighbor set will increase the uncertainty ofprediction.In the experiment,of the proposed algorithm are set to be 10 and 30,respectively.As shown in Tab.2,it is clear that MAE and RMSE ofat 30 are significantly lower than 10.Accordingly,the nearest neighbor numbersis set in the following comparison experiments of kinds of recommendation algorithms.In the experiment,the proposed algorithm (CDRec) is compared with three commonly used social recommendation algorithms TDSRec [Guo,Wang and Hou (2018)],Social MF [Forsati,Mahdavi and Shamsfard (2014)] and TDRec [Park,Kim and Oh (2016)],and the results are shown in Fig.7.

    Table2:Influence of different neighbor numbers on algorithm

    Figure7:The results comparison of four recommendation algorithms

    From Fig.7,it is can be seen that the proposed algorithm is similar to other social recommendation algorithms on MAE and slightly better than other algorithms on RMSE,which means that the proposed algorithm can predict ratings of the recommended items more accurately.

    In view of the datasets vary in scale,the execution comparison of four social recommendation algorithms is shown in Fig.8.

    Figure8:The execution comparison of four recommendation algorithms

    As can be seen from Fig.8,with increase of the scale of dataset,the execution of the proposed algorithm is significantly less than that of the other three social recommendation algorithms.

    In conclusion,the proposed algorithm can improve the recommendation efficiency on the basis of guaranteeing the accuracy of recommendation when dealing with big sample datasets.

    7 Conclusion

    For recommendation system in big data,the efficiency of recommendation al-gorithm is very important.In this paper,a collaborative filtering algorithm based on multi-subnet complex network model is proposed to construct multi-relationship complex network.By using community structure detection,similar users can be partitioned into a community,which greatly reduces the computa-tional complexity of recommendation algorithm.Experiments on real data sets show that the proposed algorithm can improve the efficiency of recommendation algorithm on the basis of guaranteeing the accuracy of recommendation.

    Acknowledgement:This work is supported by the Humanity and Social Science Youth foundation of Ministry of Education of China under Grant No.15YJC860001.This research is also supported by Shandong Provincial Natural Science Foundation,China under Grant No.ZR2017MG011 and Shandong Social Science Research Project under Grant Nos.17CHLJ16 and 18BJYJ04.This work is also supported by the Grant of Fujian Middle-aged and Young Teachers’ Educational and Scientific Research Projects in 2018,Grant No.JZ180469.

    国产不卡一卡二| 国产成人精品久久二区二区免费| 高清av免费在线| 欧洲精品卡2卡3卡4卡5卡区| 伦理电影免费视频| 亚洲精品在线美女| 国产视频一区二区在线看| 国产深夜福利视频在线观看| 成年动漫av网址| 国产三级黄色录像| 电影成人av| 国产精华一区二区三区| 亚洲欧美激情综合另类| 黄色丝袜av网址大全| 在线观看舔阴道视频| 精品国产超薄肉色丝袜足j| 久久国产精品男人的天堂亚洲| 在线天堂中文资源库| 欧美激情高清一区二区三区| 国产一区二区三区在线臀色熟女 | 男女午夜视频在线观看| 99热只有精品国产| 成在线人永久免费视频| 露出奶头的视频| 免费高清在线观看日韩| 中文字幕另类日韩欧美亚洲嫩草| 久久国产精品大桥未久av| 久久久久精品人妻al黑| 亚洲精品一二三| 一本一本久久a久久精品综合妖精| 视频区欧美日本亚洲| 久久精品国产亚洲av高清一级| 亚洲七黄色美女视频| 色精品久久人妻99蜜桃| 国产精品成人在线| aaaaa片日本免费| 精品午夜福利视频在线观看一区| 久久ye,这里只有精品| 精品高清国产在线一区| 国产精品一区二区精品视频观看| 国产99白浆流出| tube8黄色片| 精品人妻在线不人妻| av有码第一页| 亚洲一区二区三区欧美精品| 日韩有码中文字幕| 波多野结衣一区麻豆| 国产欧美日韩一区二区精品| 在线观看日韩欧美| 国产成人av教育| av网站在线播放免费| 国产人伦9x9x在线观看| 在线观看免费高清a一片| 国产精品国产高清国产av | 黄频高清免费视频| 色尼玛亚洲综合影院| 国产成人欧美在线观看 | 国产一区在线观看成人免费| 他把我摸到了高潮在线观看| 久久中文字幕人妻熟女| 丝袜人妻中文字幕| 精品一品国产午夜福利视频| av欧美777| 国产欧美日韩精品亚洲av| 亚洲av第一区精品v没综合| 最新的欧美精品一区二区| 一进一出抽搐动态| √禁漫天堂资源中文www| 少妇猛男粗大的猛烈进出视频| 欧美精品亚洲一区二区| 天天躁狠狠躁夜夜躁狠狠躁| 亚洲五月天丁香| 午夜久久久在线观看| 欧美不卡视频在线免费观看 | 他把我摸到了高潮在线观看| 精品熟女少妇八av免费久了| netflix在线观看网站| 女人高潮潮喷娇喘18禁视频| 亚洲一码二码三码区别大吗| 欧美大码av| 午夜两性在线视频| 欧美日本中文国产一区发布| 首页视频小说图片口味搜索| 久热爱精品视频在线9| 精品国产美女av久久久久小说| 中国美女看黄片| 国产高清videossex| 侵犯人妻中文字幕一二三四区| 久久精品国产亚洲av香蕉五月 | 欧美亚洲日本最大视频资源| 老司机影院毛片| 亚洲免费av在线视频| 国产免费男女视频| 国产不卡av网站在线观看| 麻豆av在线久日| 精品免费久久久久久久清纯 | 精品久久久久久,| 露出奶头的视频| 国产精品亚洲av一区麻豆| 午夜福利影视在线免费观看| 国产男靠女视频免费网站| 深夜精品福利| 午夜福利免费观看在线| 国产人伦9x9x在线观看| 啦啦啦在线免费观看视频4| 精品福利观看| 纯流量卡能插随身wifi吗| videos熟女内射| 成人av一区二区三区在线看| 国产精品国产av在线观看| 亚洲中文av在线| 最近最新免费中文字幕在线| 国产单亲对白刺激| 丁香六月欧美| 在线观看www视频免费| 一a级毛片在线观看| 久久人妻福利社区极品人妻图片| 大型av网站在线播放| 国产成人一区二区三区免费视频网站| 亚洲欧美日韩高清在线视频| 久久香蕉精品热| www日本在线高清视频| 激情视频va一区二区三区| 欧美在线黄色| 国产欧美日韩精品亚洲av| 亚洲五月天丁香| 成年女人毛片免费观看观看9 | 日韩免费高清中文字幕av| 女人被躁到高潮嗷嗷叫费观| 热99久久久久精品小说推荐| 免费观看人在逋| 69精品国产乱码久久久| 国产高清videossex| 又黄又爽又免费观看的视频| 午夜影院日韩av| 亚洲精品粉嫩美女一区| 1024视频免费在线观看| 99久久精品国产亚洲精品| 黄片播放在线免费| 岛国在线观看网站| 久久久久久久国产电影| 老汉色av国产亚洲站长工具| 青草久久国产| 伦理电影免费视频| 人人澡人人妻人| 久久热在线av| 欧美成人午夜精品| 国产精品国产av在线观看| 久久中文字幕人妻熟女| 亚洲人成伊人成综合网2020| 淫妇啪啪啪对白视频| av不卡在线播放| 国产男靠女视频免费网站| 国产免费现黄频在线看| 一区二区日韩欧美中文字幕| 两性午夜刺激爽爽歪歪视频在线观看 | 亚洲色图 男人天堂 中文字幕| 极品人妻少妇av视频| 三级毛片av免费| 啦啦啦视频在线资源免费观看| 黄网站色视频无遮挡免费观看| 亚洲少妇的诱惑av| 在线国产一区二区在线| 成人永久免费在线观看视频| 亚洲专区国产一区二区| 亚洲三区欧美一区| 国产精品久久久久久精品古装| 日韩欧美一区二区三区在线观看 | a在线观看视频网站| 日本vs欧美在线观看视频| 精品少妇久久久久久888优播| 777久久人妻少妇嫩草av网站| 国产日韩欧美亚洲二区| 十八禁网站免费在线| 亚洲七黄色美女视频| 日本撒尿小便嘘嘘汇集6| 午夜福利影视在线免费观看| 亚洲 国产 在线| 久久精品aⅴ一区二区三区四区| 国内毛片毛片毛片毛片毛片| 女警被强在线播放| 免费一级毛片在线播放高清视频 | 亚洲人成电影免费在线| 成人国语在线视频| 中文字幕最新亚洲高清| 亚洲精品国产区一区二| 国产一卡二卡三卡精品| 少妇裸体淫交视频免费看高清 | 国产1区2区3区精品| 99国产极品粉嫩在线观看| 少妇裸体淫交视频免费看高清 | 在线免费观看的www视频| 高清毛片免费观看视频网站 | 成人手机av| 日韩欧美在线二视频 | 欧美日韩视频精品一区| 人人澡人人妻人| 99久久人妻综合| 成人亚洲精品一区在线观看| 欧美人与性动交α欧美精品济南到| 国产精品一区二区免费欧美| 少妇粗大呻吟视频| 露出奶头的视频| 精品欧美一区二区三区在线| 老司机在亚洲福利影院| 亚洲成人国产一区在线观看| 制服诱惑二区| 国产日韩一区二区三区精品不卡| 777久久人妻少妇嫩草av网站| 久久精品aⅴ一区二区三区四区| 香蕉国产在线看| 久久久久精品国产欧美久久久| 色婷婷久久久亚洲欧美| 国产无遮挡羞羞视频在线观看| 久久久水蜜桃国产精品网| 色精品久久人妻99蜜桃| 午夜成年电影在线免费观看| 亚洲午夜精品一区,二区,三区| 精品人妻1区二区| 精品免费久久久久久久清纯 | 亚洲欧美激情在线| 高清视频免费观看一区二区| 中文字幕人妻熟女乱码| 好看av亚洲va欧美ⅴa在| 成人国产一区最新在线观看| 国产又色又爽无遮挡免费看| 欧美激情极品国产一区二区三区| 午夜精品久久久久久毛片777| 欧美大码av| 美女福利国产在线| 在线观看www视频免费| 啦啦啦 在线观看视频| 免费不卡黄色视频| 又黄又爽又免费观看的视频| 久久久国产欧美日韩av| 黄频高清免费视频| 老司机午夜福利在线观看视频| 久久精品国产综合久久久| 国产高清视频在线播放一区| 亚洲国产看品久久| 天堂√8在线中文| 一区二区三区精品91| 黑人猛操日本美女一级片| 欧美精品av麻豆av| 这个男人来自地球电影免费观看| 国产精品一区二区免费欧美| 少妇粗大呻吟视频| 国产一区二区三区在线臀色熟女 | 久久精品91无色码中文字幕| 中出人妻视频一区二区| 欧美日韩精品网址| 久久久水蜜桃国产精品网| 女人被狂操c到高潮| 又黄又爽又免费观看的视频| 99国产精品免费福利视频| 欧美日韩视频精品一区| 91精品国产国语对白视频| 男女午夜视频在线观看| 欧美在线一区亚洲| 免费久久久久久久精品成人欧美视频| 国精品久久久久久国模美| 一区二区三区激情视频| 1024香蕉在线观看| 亚洲熟女精品中文字幕| 亚洲色图av天堂| 久久精品人人爽人人爽视色| 黄色怎么调成土黄色| 国产又色又爽无遮挡免费看| 久久久久久亚洲精品国产蜜桃av| 男人舔女人的私密视频| 一区在线观看完整版| 下体分泌物呈黄色| 黄频高清免费视频| 亚洲午夜理论影院| 啦啦啦免费观看视频1| 在线观看免费午夜福利视频| 热re99久久精品国产66热6| 亚洲av第一区精品v没综合| 啦啦啦在线免费观看视频4| 久热这里只有精品99| 日本wwww免费看| 亚洲五月色婷婷综合| 黄色视频,在线免费观看| 男女高潮啪啪啪动态图| 国产在线观看jvid| 欧美人与性动交α欧美精品济南到| 成年版毛片免费区| 日本a在线网址| 成人精品一区二区免费| 精品人妻熟女毛片av久久网站| 色尼玛亚洲综合影院| 亚洲美女黄片视频| x7x7x7水蜜桃| 午夜精品国产一区二区电影| 最近最新免费中文字幕在线| 亚洲一区中文字幕在线| 叶爱在线成人免费视频播放| 99久久99久久久精品蜜桃| 91精品三级在线观看| 亚洲avbb在线观看| 波多野结衣av一区二区av| 国产熟女午夜一区二区三区| 久久香蕉精品热| 老司机影院毛片| 女人被狂操c到高潮| 人妻丰满熟妇av一区二区三区 | 成人永久免费在线观看视频| 国产蜜桃级精品一区二区三区 | 亚洲精品一卡2卡三卡4卡5卡| 午夜福利乱码中文字幕| 欧美精品人与动牲交sv欧美| 成人18禁高潮啪啪吃奶动态图| 亚洲欧美激情在线| 亚洲成av片中文字幕在线观看| 99久久国产精品久久久| 成年人午夜在线观看视频| 午夜免费观看网址| 亚洲一区高清亚洲精品| 成人国产一区最新在线观看| 在线观看免费午夜福利视频| 每晚都被弄得嗷嗷叫到高潮| 欧美日韩视频精品一区| 妹子高潮喷水视频| 后天国语完整版免费观看| 黑人巨大精品欧美一区二区mp4| 亚洲七黄色美女视频| 王馨瑶露胸无遮挡在线观看| 亚洲熟妇熟女久久| 欧美精品av麻豆av| 亚洲精品一二三| 法律面前人人平等表现在哪些方面| 国产在线一区二区三区精| 一区二区三区精品91| 九色亚洲精品在线播放| 高清黄色对白视频在线免费看| 美国免费a级毛片| 久久香蕉国产精品| 看片在线看免费视频| 日日爽夜夜爽网站| 国产一卡二卡三卡精品| 丰满迷人的少妇在线观看| 欧美黑人精品巨大| 色尼玛亚洲综合影院| av欧美777| 大码成人一级视频| 亚洲成人免费av在线播放| 中文字幕人妻熟女乱码| 91字幕亚洲| 男人的好看免费观看在线视频 | 亚洲av成人不卡在线观看播放网| bbb黄色大片| 久久婷婷成人综合色麻豆| 成人手机av| 国产视频一区二区在线看| 日韩人妻精品一区2区三区| 不卡av一区二区三区| 叶爱在线成人免费视频播放| 国产成+人综合+亚洲专区| 一个人免费在线观看的高清视频| 咕卡用的链子| 免费观看人在逋| 精品久久久精品久久久| 男女之事视频高清在线观看| 亚洲专区字幕在线| 免费女性裸体啪啪无遮挡网站| 精品久久久精品久久久| 极品教师在线免费播放| 两性午夜刺激爽爽歪歪视频在线观看 | 国产又爽黄色视频| 欧美一级毛片孕妇| 亚洲国产看品久久| 男女高潮啪啪啪动态图| 亚洲成人国产一区在线观看| 国产精品一区二区免费欧美| 十八禁网站免费在线| 精品国产乱子伦一区二区三区| 久久精品aⅴ一区二区三区四区| 制服诱惑二区| 少妇 在线观看| 久久精品成人免费网站| 搡老熟女国产l中国老女人| 国产一卡二卡三卡精品| 亚洲色图 男人天堂 中文字幕| 女人爽到高潮嗷嗷叫在线视频| 飞空精品影院首页| 91av网站免费观看| 成人影院久久| 嫩草影视91久久| 国产精品 国内视频| 99精品在免费线老司机午夜| 在线观看一区二区三区激情| 两性夫妻黄色片| 欧美性长视频在线观看| 美女高潮到喷水免费观看| 伦理电影免费视频| 女人高潮潮喷娇喘18禁视频| 91国产中文字幕| 欧美性长视频在线观看| 欧美日本中文国产一区发布| 成人影院久久| 久热这里只有精品99| 欧美国产精品一级二级三级| 午夜免费观看网址| 免费观看精品视频网站| 午夜福利免费观看在线| 两性午夜刺激爽爽歪歪视频在线观看 | 一区福利在线观看| 黑人猛操日本美女一级片| av视频免费观看在线观看| av一本久久久久| 亚洲国产看品久久| 国产片内射在线| 在线免费观看的www视频| 亚洲在线自拍视频| 国产亚洲欧美在线一区二区| 韩国av一区二区三区四区| 极品教师在线免费播放| 老司机亚洲免费影院| 99国产精品一区二区蜜桃av | 男人操女人黄网站| 老熟妇仑乱视频hdxx| 一进一出抽搐动态| 免费黄频网站在线观看国产| 国产男女内射视频| 黄片小视频在线播放| xxxhd国产人妻xxx| 免费高清在线观看日韩| 久久久久国产精品人妻aⅴ院 | 老司机靠b影院| 别揉我奶头~嗯~啊~动态视频| 成人国产一区最新在线观看| 可以免费在线观看a视频的电影网站| 丁香欧美五月| 久久草成人影院| 精品亚洲成a人片在线观看| 国产单亲对白刺激| 久久精品aⅴ一区二区三区四区| www.熟女人妻精品国产| 国产精品国产av在线观看| 久久国产精品人妻蜜桃| 亚洲中文av在线| 啦啦啦 在线观看视频| 欧美精品一区二区免费开放| 一二三四在线观看免费中文在| 最新的欧美精品一区二区| 国产精品一区二区精品视频观看| 亚洲人成电影观看| 999久久久精品免费观看国产| 国产精品久久久久久人妻精品电影| 精品国产国语对白av| 午夜亚洲福利在线播放| 在线十欧美十亚洲十日本专区| 欧美日本中文国产一区发布| 精品少妇久久久久久888优播| 午夜福利免费观看在线| 老司机深夜福利视频在线观看| 精品福利永久在线观看| 日韩中文字幕欧美一区二区| a级毛片在线看网站| 99国产精品一区二区三区| 欧美性长视频在线观看| 一个人免费在线观看的高清视频| 国产激情欧美一区二区| 他把我摸到了高潮在线观看| 老司机午夜福利在线观看视频| www.自偷自拍.com| 国产成人av教育| 欧美亚洲日本最大视频资源| 最新的欧美精品一区二区| 亚洲五月色婷婷综合| 亚洲五月天丁香| 国产精品久久久久成人av| 国产成人精品在线电影| 色综合婷婷激情| 国产精品一区二区在线观看99| 久久国产亚洲av麻豆专区| 美女午夜性视频免费| 可以免费在线观看a视频的电影网站| www.自偷自拍.com| 欧美色视频一区免费| 精品少妇一区二区三区视频日本电影| 国产精品久久久久久人妻精品电影| 岛国在线观看网站| 成在线人永久免费视频| 亚洲av成人一区二区三| videosex国产| 一级毛片精品| 岛国毛片在线播放| 久久人人爽av亚洲精品天堂| 欧美黑人精品巨大| 欧美中文综合在线视频| 亚洲美女黄片视频| 亚洲成a人片在线一区二区| 色94色欧美一区二区| 男女高潮啪啪啪动态图| 黄色女人牲交| 最新美女视频免费是黄的| 日韩欧美一区二区三区在线观看 | 一级毛片女人18水好多| 欧美黑人欧美精品刺激| 亚洲情色 制服丝袜| 美女高潮喷水抽搐中文字幕| 成人18禁高潮啪啪吃奶动态图| 欧美激情 高清一区二区三区| 亚洲在线自拍视频| 日本精品一区二区三区蜜桃| 91老司机精品| 免费在线观看完整版高清| 精品第一国产精品| 一进一出好大好爽视频| 久久人妻av系列| 深夜精品福利| 国精品久久久久久国模美| 女人久久www免费人成看片| 国精品久久久久久国模美| 久久久国产精品麻豆| www.精华液| 国产又色又爽无遮挡免费看| 99热网站在线观看| 欧美日韩黄片免| 交换朋友夫妻互换小说| 又黄又粗又硬又大视频| 亚洲中文av在线| 亚洲色图综合在线观看| 99热只有精品国产| 久久热在线av| a级片在线免费高清观看视频| svipshipincom国产片| 久久精品国产亚洲av香蕉五月 | 午夜福利欧美成人| 色综合欧美亚洲国产小说| 欧美激情 高清一区二区三区| 男男h啪啪无遮挡| 国产精品久久久久久人妻精品电影| 黄色 视频免费看| 十分钟在线观看高清视频www| 久久久精品免费免费高清| 窝窝影院91人妻| 亚洲av日韩在线播放| 精品国产亚洲在线| 首页视频小说图片口味搜索| 国产精品免费一区二区三区在线 | 国产精品久久久人人做人人爽| 一级片'在线观看视频| 老熟女久久久| 欧美激情极品国产一区二区三区| 黑人猛操日本美女一级片| 亚洲,欧美精品.| 欧美久久黑人一区二区| 亚洲第一青青草原| 精品国内亚洲2022精品成人 | 日韩视频一区二区在线观看| 国精品久久久久久国模美| av有码第一页| 9191精品国产免费久久| 成在线人永久免费视频| 日韩免费高清中文字幕av| 免费观看人在逋| 欧美激情久久久久久爽电影 | 大型黄色视频在线免费观看| 日本一区二区免费在线视频| 亚洲第一av免费看| 脱女人内裤的视频| 夜夜爽天天搞| 老熟女久久久| 搡老乐熟女国产| 久久人妻福利社区极品人妻图片| 国产亚洲av高清不卡| 操美女的视频在线观看| 欧美精品高潮呻吟av久久| 亚洲精品国产一区二区精华液| 精品国产一区二区三区四区第35| √禁漫天堂资源中文www| 80岁老熟妇乱子伦牲交| 夫妻午夜视频| 久久精品人人爽人人爽视色| 韩国精品一区二区三区| 免费在线观看亚洲国产| 欧美精品高潮呻吟av久久| 欧美成人免费av一区二区三区 | 亚洲av欧美aⅴ国产| 黑人巨大精品欧美一区二区mp4| 18禁裸乳无遮挡免费网站照片 | 老司机亚洲免费影院| 亚洲黑人精品在线| 亚洲欧美激情在线| 别揉我奶头~嗯~啊~动态视频| www.自偷自拍.com| 欧美丝袜亚洲另类 | 女人被狂操c到高潮| 99精品欧美一区二区三区四区| 老司机亚洲免费影院| 99精品久久久久人妻精品| 在线观看免费视频网站a站| 美女福利国产在线| 日韩大码丰满熟妇| 免费黄频网站在线观看国产| 我的亚洲天堂| 亚洲av第一区精品v没综合| 国产成人欧美| 亚洲综合色网址| 19禁男女啪啪无遮挡网站| 9热在线视频观看99| 国产成人av教育| 亚洲视频免费观看视频| 美国免费a级毛片| 老鸭窝网址在线观看| 国产一区二区三区视频了| 亚洲成a人片在线一区二区| 一本一本久久a久久精品综合妖精| 国产精品乱码一区二三区的特点 | av网站免费在线观看视频| 99国产极品粉嫩在线观看| 久久人人爽av亚洲精品天堂| av网站免费在线观看视频|