Mahout中基于用户喜好的推荐系统
我们如何根据用户喜好或浏览历史在 Apache Mahout 上运行推荐系统?简而言之,在基于内容的网站上,95% 的流量来自未登录的用户,他们将通过搜索引擎访问。我们可以通过使用 IP 来使它们独一无二。 Apache Mahout 上是否有可以找出用户相似浏览行为并推荐相关内容的地方?
How we can run a recommendation system on Apache Mahout based on user liking or browsing history? In short on a content based websites 95% traffic by non logged in users and they will come via search engine. They only way we can unique them by using IP. Is there anyway on Apache Mahout where we can find out the similar browsing behavior of users and recommend relevant content?
如果你对这篇内容有疑问,欢迎到本站社区发帖提问 参与讨论,获取更多帮助,或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱,如果其他用户或者作者回复了您的评论,将不能在第一时间通知您!
发布评论
评论(1)
一个简单但可能非常有效的起点是使用 IP 地址作为用户 ID(解释为长),并将喜欢或浏览的页面用作项目。我什至会忘记分配评级。
然后在 Mahout 中使用 GenericBooleanPrefItemBasedRecommender ,并在适合您的任何
DataModel
之上加上合适的相似性度量,例如LogLikelihoodSimilarity
,您就可以了。A simple but probably pretty effective starting point would be to use the IP address as a user ID (construed as a long), and use pages liked or browsed as items. I would start by even forgetting about assigning ratings.
Then use
GenericBooleanPrefItemBasedRecommender
in Mahout plus a suitable similarity metric likeLogLikelihoodSimilarity
on top of whateverDataModel
suits you, and you're pretty much there.