3.3. Neighborhood formation
The main goal of the neighborhood formation is to find for each customer x a set of
n customers that are the most similar to him. The similarity is given by the similarity
function sim. The neighbors are formed by applying proximity measures, such as thePearson
correlation (Sarwar et al. 2000, Shardanand et al. 1995), or cosine similarity (Good et al.
1999, Sarwar et al. 2000) or mean squared differences (Shardanand et al. 1995), between two
opinions or profiles of the customers. Given experiments done in other systems (Herlocker
et al. 2000, Shardanand et al. 1995), OSGS uses the mean squared differences proximity
measure. We apply it to (a) the history of the customers and to (b) their demographics and
preference profile in order to decide if customerx is a neighbor of customery . This is done
by applying a similarity function for each field of the profiles (see Lin 2002 for a detailed
description). Different similarity functions are used for each field in order to normalize
the fields’ values to the range [0, 1]. As stated above, the similarities of all the fields are
combined using the mean squared differences proximity measure in order to obtain one
weight for the profile (Shardanand et al. 1995).
For example, in GMSIM the similarity function for the gender, age, quality, price, domain
expertise, and warranty fields is given by the following formula:
(Sarwar et al. 2001)). OSGS constructs two neighborhoods for each customer, as described above. The idea behind this is that when a customer is searching for a specific product, it might be useful to assist him in finding the desired product by using information derived from customers who have similar tastes, or similar characteristics.
3.4. Weight functions
In order to prune the search tree effectively, each node in the search tree is assigned a weight
that estimates the customer’s interest in the category or product associated with the node.
In this section we present a description of the four weight functions that are used in our
algorithms. The weight functions are based on the customer’s preferences, the keywords
provided by the customer (in case of a keyword search), the customer’s history and his
neighbors (neighbors-by-history-profile and neighbors-by-demographics-and-preferenceprofile).
Table 1 summarizes all the weight functions and their notations.
是原文中几个段落,我翻不来,高手帮我翻译下。最好告诉我哪里有翻译原文,好的话追加100分
3.3 周边邻里数据库的形成
周边邻里数据库形成的主要目标是为每个客户找到一套对他们来说相似的环境,而相似性是通过相似的功能性给出的。邻居是由申请接近措施,如the Pearson 相关(迪斯沙尔瓦等人。 2000年, Shardanand等人。 1995年) ,或余弦相似性(善等人。 1999年,迪斯沙尔瓦等。 2000 )或平均平方差异( Shardanand等。 1995年) ,在两种意见或配置文件的客户。鉴于实验做其他系统( Herlocker等。 2000年, Shardanand等。 1995年) ,OSGS使用均方分出相似措施。我们运用它来比较(A)历史上的客户和( B )它们的人口和
偏好的个人资料,以便决定是否客户X是近邻客户Y。 这样做是采用相似功能的每个领域的概况(见林2002年详细描述)。不同的相似性功能用于各个领域,以规范领域的价值观的范围[ 0 , 1 ] 。正如上文所述,相似的所有领域相结合,利用平均平方差异接近措施,以便获得一个体重为配置文件( Shardanand等。 1995年) 。
例如,在GMSIM的相似功能的性别,年龄,质量,价格,专业知识,和保修等领域给予下列公式:(Sarwar et al. 2001)).
OSGS构造两个社区为每个客户,如上所述。背后的想法是,当客户正在寻找特定的产品,它可能是有用的,协助他找到想要的产品使用信息来自客户谁也有类似的口味,或类似的特征。
3.4 。权函数
为了修剪有效的搜索树,每个节点在搜索树的重量分配估计客户的兴趣类别或产品相关的节点。在本节中,我们现在的描述四个权函数中使用我们的算法。重量职能是根据客户的喜好,关键字 由客户提供(如果是关键字搜索) ,客户的历史和他的邻居(邻居通过历史资料和邻居按人口统计和preferenceprofile )。表1总结了所有的重职能及其符号。
哥们,里面有的东西翻译不好,或没翻译出来是因为你给的单词貌似错了,我大概看了下,它讲的是个网络自选择和分配比较人物数据的程序。
A B X Y 代表的虚拟人物,假设如你 我 他,并不是单词。
翻译来自:四川外语学院国际商学院08级。
参考资料:sa