Data Mining for Information Retrieval and Processing in Web 2.0 Social Networking

在这个研究计划中，数据挖掘和模糊逻辑将被用于为Web2.0社交网站Facebook、Twitter、YouTube和LinkedIn构建一组新颖的推荐系统。对于FaceBook和YouTube，推荐系统将采用模糊逻辑，这将能够提供基于这些网站中存在的基于感知的数据的正确解释的信息。因此，由此产生的推荐系统可以作为检索系统的附加部分向用户提供建议。当用户在Facebook或YouTube上说，“我通常喝蒂姆·霍顿的咖啡”或“我喜欢这个视频”时，“通常”和“喜欢”的含义因用户而异。即使用户使用“通常”、“大多数”和“经常”等不精确的词来表示他们的兴趣，模糊检索系统也可以做出相关的推荐。例如，可以专门为Tim Horton ' s的Facebook粉丝建立一个模糊检索系统，并可以根据对用户兴趣的分析来确定他们最喜欢的口味，为Tim Horton ' s的咖啡饮用者推荐一种特殊的饼干。随着域名在Web2.0网站中变得更加具体（例如在LinkedIn中），单词之间的模糊关系可以被定义并用于指导发送特定搜索查询的用户。例如，Linkedin特定网络中的用户阅读特定文档，这些文档可用于构建知识库，作为该网络用户的推荐系统。因此，当用户在计算机工程师网络中搜索工作时，输入“设计”一词时，推荐系统只会显示“芯片设计”或“集成电路设计”等相关词，而当属于计算机科学家网络的求职者输入“设计”时，推荐系统可能会显示“算法设计”或“软件设计”等相关词。类似地，Twitter的推荐系统可以分析tweet并识别具有相同兴趣的用户。Twitter拥有比其他所有社交网站更多的公开信息；因此，针对Twitter提出的另一项研究活动是将其用作事件检测平台。通过使用各种数据挖掘技术，本研究活动旨在开发一个框架，从tweet的大型数据集中有效地提取新兴事件（关于世界上真实发生的有用信息）。本文的最后一部分是利用多智能体系统开发数据收集和评估工具。这些多智能体系统将被用作数据爬虫来主动更新这些推荐系统的知识库。此外，与其他基于本体或语义的搜索引擎相比，多智能体系统还可以用于验证所提出的信息检索（IR）和推荐系统。在社交媒体网站中使用多智能体系统来评估IR和推荐系统是一个新颖的想法，除了常见的准确性指标（即精度和召回率）之外，它还将提供两个额外的性能指标（即分布式处理有效性和可扩展性）。在上述Web2.0网站的信息处理研究活动中使用模糊逻辑是另一个在本研究计划提出之前尚未开发和实施的新想法。

In this proposed research program, data mining and fuzzy logic will be used to build a group of novel recommendation systems for the Web2.0 social networking sites Facebook, Twitter, YouTube, and LinkedIn. For FaceBook and YouTube, fuzzy logic will be employed in the recommendation systems, which will be capable of providing information that is based on the correct interpretation of perception-based data existing in these sites. Therefore, the resulting recommendation systems can provide advice to the user as an additional part of a retrieval system. When in Facebook or YouTube a user says, “I usually drink Tim Horton’s coffee,” or “I like this video”, the meaning of “usually” and “like” varies among users. A fuzzy retrieval system can make relevant recommendation even when the users deploy imprecise words such as “usually,” “most,” and “often” to show their interests. For example, a fuzzy retrieval system can be built specifically for Facebook fans of Tim Horton’s, and can recommend a special cookie for the Tim Horton’s coffee drinker based upon an analysis of users’ interests to determine their favourite tastes. As domains become more specific in Web2.0 sites (for example in LinkedIn), fuzzy relations between words can be defined and used to direct users who are sending specific search queries. For example, users in a particular network of Linkedin read specific documents that can be used for building a knowledge base that acts as the recommendation system for the users of that network. Therefore, whenever a user who is searching for a job in computer engineers’ network types the word “design,” only the related words such as “chip design” or “IC design” should be shown by the recommendation system, whereas when a job seeker who belongs to the computer scientists’ network types “design,” the suggested related words might be “algorithm design” or “software design”. Similarly for Twitter, the recommendation systems can analyze the tweets and identify users with the same interests. Twitter has more public information than all other social networking sites; therefore, another research activity proposed for Twitter is using it as a platform for event detection. By using various data mining techniques, this research activity aims to develop a framework that effectively extracts emerging events (useful information about real happenings in the world) from the large data set of tweets. The last part of this proposal is developing data collection and evaluation tools by using multi-agent systems. These multi-agent systems will be used as data crawlers to actively update the knowledge bases of such recommendation systems. Moreover, the multi-agent systems can also be used for validation of the proposed Information Retrieval (IR) and recommendation systems in comparison with other ontology or semantic-based search engines. Use of the multi-agent systems for evaluation purpose of IR and recommendation systems in social media sites is a novel idea that will provide two more performance metrics (i.e., distributed processing effectiveness and scalability) in addition to the common accuracy metrics (i.e., precision and recall) . Using fuzzy logic in the above research activities for information processing of the Web2.0 sites is another new idea that has not been developed and implemented prior to this proposed research program.