400-6710-390
客户服务热线
English 简体中文
网站首页 ⊙ 语言 ⊙ 学术论文
学术论文
academic paper
中译英
英译中​
微信作为一种社交网络（SNS），与微博、人人不同之处在于它是基于现实的人际关系而创造出的虚拟网络社会。因此父母在微信中的参与度要远高于微博、人人等交友网站。以微信为代表的社交平台的发展已经不单单是横向的、以兴趣为集合点的陌生人交友平台，而是基于熟人间的社交关系网络。在这种模式中，个人的兴趣、观点，甚至文化水平、人格都得到鲜明的彰显。亲子双方通过刷新新鲜事，了解对方在干什么、对什么有兴趣，甚至能察觉到对方的情感、心理状态。

As a social networking service (SNS), Wechat is different from microblogs and Renren in that it's a virtual network society created out of realistic interpersonal relationships, which means parents are involved in Wechat to a much larger extent than dating websites like microblogs and Renren. SNS platforms represented by Wechat are growing both horizontally as gathering places for strangers with common interests and vertically as social relation circles for acquaintances. This communication mode allows clear manifestation of individuals' interests, opinions, or even educational level and personalities. Thus in a parent-child relation, both sides have better chances to learn more about each other such as their current activity or interest by updating What's New, and they may even perceive emotional or psychological conditions of the other side. 

Many domain specific characteristics of the data are used for fraud detection. For example, it is well known that large absolute values of the transaction amounts may correspond to anomalies. The most common technique is to build user profiles on short segments of transaction sequences.  Typically, the ordering among a short segment of the transactions is immaterial. If desired, a single transaction of the user can also be used. Either a single transaction or a short sequence of transactions can be converted into a feature vector, which is compared to the user’s profile. The key is to design a similarity function,  which can  encode the wide diversity of attribute types, the collective profile within a short segment, and domain-specific knowledge (eg. higher values of transactions or sudden bursts of high-value transactions are more likely to be fraudulent). 

在欺诈检测中，会使用很多与具体领域相关的数据特征。例如，众所周知：如果交易额的绝对值很大，就可能意味着存在异常情况。最常见的方法是使用短交易段和交易序列来创建用户资料。通常情况下，短交易段的排序是不重要的。如有必要，还可以利用用户的统一交易信息。统一交易或短交易序列都可以转换为一个特征矢量，并与用户的资料进行比较。其重点是设计一个类似函数，该函数能够对多种属性、一个短交易段内的整合资料、以及与具体领域相关的知识进行编码（比如在出现较高交易值或突然出现高价值交易的情况下，存在欺诈的概率更高）。