A universally accepted definition of information filtering is, unfortunately, still lacking. So here is my personal definition, which I have used to build the Information Filtering Resources web page. Generally, the goal of an information filtering system is to sort through large volumes of dynamically generated information and present to the user those which are likely to satisfy his or her information requirement.
In order to sharpen this definition, a distinction should be drawn between information collection and information filtering. In some domains (e.g. USENET News) the collection effort is minimal because the information comes to you. In other domains (e.g. the World Wide Web) the collection effort can be considerable because no mechanism exists to draw new information to the attention of a filtering system. The point to be made here, though, is that information collection is an interesting area in its own right, but I do not propose to include it in my definition of information filtering. In my view, the information filtering problem begins only after you have gained access to the new information.
数据挖掘工具
Information filtering has been applied to a several domains using a variety of technical approaches. The original methods were manual alerting services that brought new information to the attention of users of research and special libraries. At the time this was referred to as Selective Dissemination of Information (SDI), a name which fell from favor about the time the Strategic Defense Initiative (SDI) was introduced in the United States :-) A few modern systems have adopted this remarkably descriptive name for the filtering process, however, and the interest in information filtering that has resulted from the present research thrusts in digital libraries arises at least in part from this tradition.
With the growth if the internet and other networked information, research in automatic filtering of networked information has exploded in recent years. Becuase of their low cost, large volume, and ease of recognizing new information, the most popular domains for research systems have been USENET News and electronic mail. The recent explosive growth of the World Wide Web has made this an interesting domain which has attracted some good research, although the information collection problem appears to make this a more difficult domain in which to conduct basic research on information filtering techniques. Another domain which has attracted considerable research interest is the annual Text REtrieval Conference (TREC) in which a standard text collection is used and a carefully controlled evaluation methodology is enforced. In TREC the information filtering task is refered to as "routing," adding somewhat to the confusion of terminology in this field. In fact, TREC recently adopted a special interest "filtering" track which adopts a different evaluation methodology but which conforms to the definition of filtering presented above. Commercial systems which filter newswire articles and other specialized information sources are becoming available as well. Filtering techniques will likely be applied to other domains such as images, sound and video in the future. 数据挖掘论坛
The distinction between information filtering and the more established field of information retrieval has proven to be the source of some confusion as well. Information retrieval broadly deals with the selection of information, and many of the features of information retrieval system design (e.g. representation, similarity measures or boolean selection, document space visualization) are present in information filtering systems as well. If one considers information retrieval from a very general "information selection" viewpoint, information filtering is simply a special case in which the information space is very dynamic. If, on the other hand, your personal definition of information retrieval involves selection of relatively static information in response to relatively dynamic queries, then information filtering is best viewed as the dual problem to information retrieval. Regardless of which viewpoint you take, though, it is clear that researchers in information filtering will likely benefit from familiarity with the legacy of research in various aspects of information retrieval. For practical reasons I have not attempted to compile a comprehensive listing of network-accessible resources on information retrieval, however, so the interested researcher should refer to the Related Web Pages section of the Information Filtering Resources web page for some starting points on information Retrieval.
