文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > A connectionist approach for evaluating the complexity of interaction in the World Wide Web
侵权投诉

A connectionist approach for evaluating the complexity of interaction in the World Wide Web

The increasing success of Internet is a great opportunity for business activities but also for interpersonal communications. Several researchers studied the interesting field of social interaction [ROS, HEY] through the net. One of the big difficulties to

A connectionist approach for evaluating the complexity of interaction in the

World Wide Web

The case of News groups

luigi.lancieri@cnet.francetelecom.fr

France Telecom – Centre national d’études des télécommunications (CNET)

The increasing success of Internet is a great opportunity for business activities but also for interpersonalcommunications. Several researchers studied the interesting field of social interaction [ROS, HEY] through the net.One of the big difficulties to have a good view of human interaction is to define suitable tools and metrics torepresent these characteristics. The goal of this paper is to going ahead in this field and more specifically to evaluatethe complexity of the knowledge interaction on collaborative groups.

One of the first difficulties to achieve such studies is to find data correlated with the different level of interaction. Itis easy to find in HTTP server's or caches log files data that inform on the activities of users: for example retrievingthe top 10 more frequented site or extracting the thematic or behavioral profile of users[LAN99]. What is verydifficult however is to have a view, in addition of what users get from the network, what they “give” to the network.These 2 actions (get and give) are the basis of the interaction. The evolution of the network is directly driven byhuman knowledge interaction. As in direct human interaction people are more or less easily influenced by theinformation they receive (consciously or not). In a group of people some are more or less leader on giving to thegroup. The knowledge of the group is a complex result of all this kind of interactions. Even the quantitativeinteraction (global flow of data) in the global network found its roots in the local human's interactions.

As we said before, it is difficult to study this kind of phenomena because of a lack of information related to theseinteractions. It is true in the field of sociology but also regarding the network. Regarding the evaluation of what usersgives to the network, one of the most interesting service to study is the Usenet news groups. The study ofcollaborative working group like Usenet news can also be interesting in smaller electronic working groups as incompany’s Intranet.

In this context, we present a method to evaluate different forms of influence within and between discussion groups.This method involves a basic linear associative neural network which is an interesting tool to learn and show longterm cross influence between groups.

Method

We measure, first, the activities of each participant of the group. For that, we compute in a period of time thequantity of information (messages) each subscribers sent to a given news group. The statistic study of the resultingfrequency distribution allows to extract, in the same way as the principle of Zipf law [BRE, MAN, ALM], thecharacteristics of the interaction on the group (see below).

The main hypothesis made in this study is that a person that gives information to the group is supposed to receivealso information. This is not always true because one can send a mail to a group one time and then never consult thegroup again. However, as in Zipf law this kind of situation is automatically under considered, this limits the risk oferrors. At the contrary, people that are active in a group are by force "listening" the group. The main reason of this isthat, as in real groups, people talking without first listening what others have to says are rapidly excluded from thegroup. The computation of the interaction characterization of the group is very easy, as the following plots show.Zipf state that series of data following his law should be a straight line after log(x)/ log(y) transformation.

y=1

Palog(y)= alog(p)

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email