华海的生活， 工作， 学习.

 Tag:学习     
甘永成1， 祝智庭2
（1．加拿大多伦多大学安大略教育学院，多伦多 M5S
1V6；2．华东师范大学教育信息网络中心，上海 200062）

内容提要：本文侧重从系统的整体性、智能的整体性和动态性、学习模式、知识管理这四个维度，来构建虚拟学习社区知识建构和集体智慧发展的学习构架。根据建构主义-、情境学习理论，在Internet上建立虚拟学习社区，就在于学习社区可以应用学习者的个体智慧，发挥集体智慧的优势，共同解决学习中遇到的难题，通过个体间-长期的互动、合作和知识建构，达到集体智慧的凝聚和升华。
关键词:
集体智慧，知识建构，虚拟学习社区，多元智能，e-Learning，知识管理，学习型组织

一、引言
集体智慧是集体在创造、创新和发明上共同合作的一种能力。当这个社会越来越依赖知识时，这种集体能力就变得越来越重要。集体智慧已经成为知识社会中竞争、创造和-发展的决定因素

虚拟学习社区是由具有共同兴趣及学习目的人们组成的学习团体在Internet上构建的虚拟学习环境。他们利用多种网络通讯工具，通过相互的交流、互动、讨论和-协作，以及通过协作学习等多种学习方式，共享彼此的观点、思想、资源、知识、学习经验和集体智慧，从而促进知识建构和个体智慧的发展，达到学习的目的和促进自身-学习能力的发展。

图1  支持集体智慧的四层架构
我们可以从信息通讯技术、e-Learning、知识管理、学习社区这四层架构来说明集体智慧发展的技术支持（见图1）。（1）信息通讯技术（ICT）。ICT-和Internet网络的迅猛发展为信息、知识的广泛传播提供了可靠的技术保障和底层通讯协议架构。"每一个新思想都被并入了下一个思想，从而产生一种对人类本-质和我们所生活的宇宙的积累性的新认识，这种新的智慧模式则由全世界新发展出来的网络所分享
。"（2）e-Learning。朝着数字化、网络化、多媒化和智能化的方向迈进的信息化教育，正引起一场学习方式的重大变革。 e-Learning使用网络技-术来设计、传输、选择、管理和扩展学习，将成为学习的主要方式之一和终身学习的主要平台。（3）知识管理。知识管理作为获取、存储、传播、应用知识的一种管理方-式，通过知识共享、运用集体智慧提高应变和创新能力。知识管理在知识时代具有举足轻重的作用。（4）学习社区。虚拟学习社区为学习者享有他们所需的知识提供一个-开放的学习平台，并能够对这些知识获取、分类、存储、共享。学习者在这样的学习环境中交流信息、探讨问题、提出新观念、拓展问题思路、交流学习心得、相互争论、-达成共识，进行知识建构，不仅学习到了知识，而且获得了学习能力，提高了多种智能，促进了团队合作精神，提升了集体智慧。
我们将虚拟学习社区中的集体智慧理解为：在学习过程中，学习小组或集体协力加强整体性与相互联系，以便加深我们对事物的理解，使个体间的智慧达到进一步的凝聚，-进而达到更高层次的整体性和密切联系，形成共同创造的能力。本文拟从整体性、智能、学习和知识管理这四个维度，来构建虚拟学习社区集体智慧发展的学习构架（见图-2）。
二、原理分析及讨论
（一）整体性维度
1．共性和差异性
整体性是系统最普遍和最本质的属性，共性和差异性是整体性的两个重要属性。共性是系统要素共同的、普遍的属性。这种共同、普遍的属性使各个系统要素相互联结和贯-通，形成统一的有机整体。系统要素的差异性是指它的个体性、独特性，以及系统要素与要素之间的差异性。

图2  集体智慧发展的学习框架
小组、组织、社区乃至社会是一个有机的整体。为了分析组织内成员的性质，两个最重要的视角是共性和差异性。因为只有共同的兴趣和目的，人们才能组织起来形成一个-整体，而每个成员在个性、学习风格、能力、智力水平等方面又都存在很大的差异性。正如霍华德·加德纳所言："多元智能观点的核心--不管是理论上和实践上--都-在于认真地看待人的个性差异
。"
首都师范大学王陆教授等学者，专门对学习者个性因素和学习类型与远程教学效果进行了研究，并从个性因素的五个方面（有恒性、实验性、创造能力、在新环境中的成长-能力和性别）分别进行了对比试验。他们的结论显示，学习者的个性因素在很大程度上影响网络教学效果，应该根据学习者个性因素提供不同的网络教学模式
。
2．系统论视角
系统论主张从整体性、关联性、动态性、层次结构性和自组织性等观点出发，研究系统与系统、系统与组成部分、以及系统与环境之间的普遍联系。虚拟学习社区是按照一-定的学习目的组织起来的，具有相应的组成要素、结构和特点。下面从四个方面分析其主要特点（见图3）。

图3  系统论视角
（1）层次结构性/多变性。学习者的成长环境、人生经历、文化背景不尽相同，在个性特征、学习风格、学习方法上也各有特点，知识水平、认知能力有高有低，因此，-整个学习社区的学习个体具有差异性和多变性。这就好像一块未被磁化的磁铁，其磁极子是"各向异性"的。
（2）关联性/协同性。学习者之间是相互关联、相互影响的。如果不加以适当的控制，虚拟学习社区的学习环境就会遭到破坏。重要的一点就是要利用学习者在学习目的 -、学习兴趣、学习承诺方面的共性，使之产生协同作用，即产生"正相关"作用而不是"负相关"作用，使各种因素朝着学习知识、增进学习绩效、提高认知水平、提升集-体智慧的方向发展。协同性就像磁化后的磁极子产生"各向同性"的作用。
（3）交互性/动态性。合作学习是学习者之间互教互学、彼此之间交流信息的过程，也是互爱互助、情感交流、心理沟通的过程。合作中的学习活动的任务分担与成果共-享，相互交流与相互评价，使学习者能体验到一种被他人接受、信任和认同的情感。通过相互启发、激励，发展认知能力，会极大地促进个人的合作能力与团队精神的形成-。
（4）系统效应/整体效应。系统的整体功能大于内部各元素孤立功能之和。虚拟学习社区里的智慧凝聚现象正是"系统效应"的一种体现。学习者既相互独立，又相互依-赖、相互制约，有机地组成了一个学习的集体。当然，学习社区并不一定产生"正的系统效应"，也可能产生"负的系统效应"。要取得成功，就要采取一定的方法和手段-，促进 "系统效应"向正的方向发展，努力促进虚拟学习社区集体智慧的显现和结晶。
（二）智能维度
从个体智能到集体智慧是一个连续统。从智能整体性和动态性两个维度考察，它包括多元智能、个体共性智能、合作智能和集体智能四个上升的层次，以及自我组织（发散-）、相互连接（收敛）、智慧结晶（凝聚）和共同创造（创新）四种状态。
1．智慧连续统
从个体到集体、智能到智慧发展过程来看，集体智慧可分为四个维度（见图4）。（1）多元智能。霍华德·加德纳把个体所具有的智能分为语言、数理逻辑、视觉空间、-肢体运动、音乐、人际沟通、自我认识、自然观察者智能。这是一种应用我们个体智能谱系的能力。每个个体都具有由潜在的或活跃的多元智能组成的品质，显现为不同程-度的个人影像。（2）合作智能。在学习和工作过程中，与他人和集体进行有效的合作，最大限度地发挥团队的能力。合作智能是我们与所处的环境和环境的关系中产生协-同的能力。（3）集体智能。小组、团队、组织、社区、以致整个社会学习，解决问题，计划未来，理解和适应内部环境和外部世界的能力。（4）集体智慧。一个系统（-个人、小组、组织、社区、社会）有效地集成和利用参与者不同才智的能力，有效地学习、工作、解决问题的能力，使之像一个有机运转的整体。集体智慧是各种智能的凝-聚，可形成深远的洞察力和远见卓识。

图4  集体智慧连续统
从个体智能到集体智慧，这是一个由个体、小组/团队、组织、社区到社会，由低级到高级、由弱小到强大的发展过程，也是一个动态螺旋向上的发展过程。
2．智能整体性视角
集体智慧是一个系统现象。从个体智能到集体智能是一个连续统。我们可以从多元智能、个体共性智能、合作智能和集体智能这几个上升的层次来分析虚拟学习社区的智能-形态（见图5）。

图5  智能整体性视角
（1）多元智能。每个学习者在某种程度上都具备多元智能，只是表现的形式和发挥的程度不同；他们又都具有各自的优势智能和弱势智能、个性特征和学习风格。个体对-同一个事物、同一个问题的认知角度不同，观念不同，呈现多样化的趋势，这正是个体产生创造性的心理特征依据所在。
（2）个体共性智能。个体的每种智能具有相对独立性，但各种相对独立的智能是以不同的方式和程度有机地组合在一起的。从每个个体的共性来看，有两个共性部分：（-1）智能起点大致相同。虽然个体之间的智能呈现高低不同的水平，但组成一个学习社区的学习者在综合智能水平上是大体相当的。（2）认知结果大致相似。通过学习者-个体的努力和集体间的合作，在学习单元结束之后，学习者的智能都得到了相应的提高，达到了某种学科大纲所要求的目标，对学习的知识达到了大致共同的认知。
（3）合作智能。学习者遇到的每一个难题也是一个机会，它需要彼此合作来认识这种难题和提高其智能。合作的一个挑战是创造性地运用个体的差异性，因为个体之间的-差异性对学习者既有正面的积极作用，也有负面的消极作用。合作智能就是要运用各自的差异性创造合力，达到共同解决问题的目的。
（4）集体智能。集体智能表达了两种智能的联结，一是肯定了个体智能的独立性；二是肯定了个体智能在观察复杂情境时的交互性。由于存在"针对某一场景的不止一种-角度"的"复眼式观察"，因而允许我们从许多角度综合一种强烈的视觉感知，使我们把镶嵌某一对象的环境连接起来，从而使意义充满张力。于是，复眼式观察或思维展-示了比任何个体观察更宽阔和更逼真的视野
。集体智能是分布式个体智能的有效组合，是个体智能的合作和凝聚，从而使整体智能得以加强和提高。
3．智能动态性视角
集体智慧的螺旋上升的周期可以分为四个阶段：发散、收敛、凝聚和创新。整体中个体智能之间的相互作用可分为四种状态：自我组织、相互连接、智慧结晶、共同创造（-见图6）。

图6  智能动态性视角
（1）自我组织--发散。每个学习者都是一个独立的学习个体，都有自己的学习计划、学习方法和学习风格。当遇到学习难题时，他可以在虚拟学习社区里发表自己的问-题，向同伴请求帮助。他们每个人看问题的方法、视角、思路、层次和观点都可能不一样，可能用不同的方法对问题进行求解，从而产生多种结果和方案，这是一个思维发-散的过程。
（2）相互连接/依赖--收敛。学习者可以对同一问题，发表不同的观点，开展热烈的讨论，或反驳某一观点，指出其中的不足或错误；或赞同某一观点，提出改进意见-。这是学习者之间相互影响、相互联系和交互的过程。指导者在其中起着重要的作用。第一，提出具有启发性的意见和思路，积极引导学习者更深入的思考；第二，引导讨-论的气氛和议程；第三，引导讨论问题的方向。使问题的解答向收敛的方向发展，达成共识。这也就是问题求解的逐步收敛的过程。
(3 )
智慧结晶--凝聚。学习者对问题多次讨论、争论和挑战的过程，是学习者调动自己的知识和智能、积极主动思考的过程；也是学习者之间相互交流、沟通、反馈和反思的-过程；同时也是学习者理解能力逐步形成、认知水平逐步提高并最终形成共识的过程。这里的共识有两层意思：第一，对良构问题，学习者最终达成一致的求解方案（可能-不止一种解答）；第二，对劣构问题，常常没有唯一的答案。这里的共识是一种趋同，通过对问题逐步求同存异的过程，达到对问题的核心的相近理解和把握
。这是一种对事物的理解和洞悉能力的"共同"提升。
（4）共同创造--创新。学习者不仅要学习知识，更应该培养创新意识、创新思维和创新能力。学习者不能仅仅满足寻求问题的答案，还要获取更高层次的思维能力，即 -在已有的问题基础上，进行发散性和联想性思考，提出新问题、新观念。创造性地利用学习者个体的差异性，进行"头脑风暴"式的思考，这是学习社区共同创造的一大特-征。
（三）学习维度
从传统的学校教育到e-Learning也是一个连续统，实际上大部分的学习是混和式学习（Blended
Learning）。网络教育的学习模式主要有两种：一是个别化自主学习，另一种是协作学习。虚拟学习社区正是实施这两种学习方式的最佳途径（见图7）。

图7  学习模式维度
1．个别化自主学习
个别化学习是指根据学生的不同特征而因材施教，通过网络提供学习环境、学习方式、学习内容的多重选择，为每个学生提供最佳的教学支持，让学生自由选择学习内容，-制定学习计划，安排学习时间、地点，从而自我获取知识，更新知识
。它打破了传统的学习群体的结构，学习者作为独特个体，拥有了比传统教育中更多的个性化色彩和更多的个别化学习机会，能最大限度的满足不同个体的需求，使自我规-划、自我调整的独立学习变得更容易。
2．协作性学习
协作学习是指利用计算机网络建立协作学习的环境，通过小组或团队的形式组织学生学习，使教师与学生、学生与学生在讨论、协作与交流的基础上进行协作学习
。它有利于培养学生形成良好的学习态度、协作精神和人际关系，培养认知领域的某些高层次技能。
协作学习的理论来源主要有两个：（1）心理学家维果茨基的社会学习理论。根据维果茨基的观点，学生在协作交互活动中获得的能力，能够被内化而变成他们的独立发展-成果。（2）建构主义学习理论。建构主义认为，知识不是通过教师直接传授得到的，而是学习者在一定的情境下，借助其他人（教师和学习伙伴）的帮助，利用必要的学-习资源，通过意义建构的方式获得的。"情境"、"协作"、"会话"和"意义建构"是学习环境中的四大要素。协作学习把同学和教师当成一种学习资源和环境，通过协-作和互动获取知识，是建构主义学习理论的一种体现。要使协作性学习取得成功，使学生学习效果最大化，必须考虑积极的相互依赖、促进性的互动、个体职责、社交技能-、小组自加工这五大基本要素
。
（四）知识管理维度
1．知识管理宏观视角
学习的目的就是要获取知识，进而将知识转变成人生的智慧。我们可以从个人知识管理、学习型组织、组织记忆和社区文化四个方面来考察（见图8）。
（1）个人知识管理。个人的学习和工作，对知识的收集、整理、存储、应用与创新，应从知识管理的高度，结合个人的智能开发、专业发展、人生规划，不断增长自己的 -知识和智慧。教育的目的不仅是要获取知识，更重要的是促进学习能力的提高，个人智慧的增长。正如1980年诺贝尔化学奖得主保罗·博格在谈到教育对他的影响时所-说："教育对一生最大的贡献是帮助你发展好奇心和培养你寻找有创造性答案的直觉。随着时间的流逝，我们了解的许多事实都会被遗忘，但我们发现问题和解决问题的能-力却永远不会消逝
。"这就是知识与智慧之间关系的最好说明与解读。
（2）学习型组织/社区。学习型组织是一个"不断创新、进步的组织，在其中，大家得以不断突破自己的能力上限，创造真心向往的结果，培养全新、前瞻而开阔的思考-方式，全力实现共同的抱负，以及不断一起学习如何共同学习
。"
彼得·圣吉关于学习型组织的"五项修炼"，为学习社区/实践社区提供了强大的理论基础和方法论指导。要将学习社区各阶段呈现的知识联系起来，把智慧凝聚起来，需-要指导者、教师进行系统思考，运用知识管理的手段，促成知识的发现、传播、存储、应用、共享和创新，并逐步培养学习者的学习能力、思维能力和创造力。

图8  知识管理的宏观视角
（3）组织/社区文化。虚拟社区之所以能将成员培养成稳定的、忠实的用户，其最大特点就是由共同命运感、归属感及自我价值的实现构成的文化维系力。如果不注意社-区文化的培育，学习社区同样会失败。比如，有的学习者不愿意将自己的知识与大家共享、不愿意将自己的知识进行总结后提交到知识库中去，还有一些人甚至不习惯到知-识管理系统中去获取知识等等。所以，必须十分重视变革管理，促进学习社区文化创新和学习者的观念转变。
（4）组织/社区记忆。组织记忆是指在组织中建立知识库，以储存组织所累积的方法知识及其他知识资产，并促使这些知识资产增进知识密集的工作历程之效能与效率
。组织/社区记忆是建立学习社区的重要阶段，需要将不同来源的知识分类、整理、提炼并加以存储，将分散知识提升为组织/社区记忆。
2．知识管理微观视角
知识管理主要是通过知识的分享而达到价值倍增效应的，而虚拟社区显然是实现知识共享的最直接方法
。知识管理的典型应用就是实践社区或知识社区
。研究虚拟学习社区里的知识转化，是以个体知识与集体知识的转化为对象的（见图9）。
（1）外化。学习者针对大家共同有兴趣的课题，分享自己特殊的经验、感受和观点，让参与讨论的成员将其个人的隐性知识表达出来，促使成员在学习社区的互动中，产 -生创新的观念。在个人知识"外化"过程中，隐性知识会通过隐喻、模拟、图表或观念的方式表达出来，经过模式化后就形成"观念性知识"。
（2）综合。学习者经过多次交流和互动后，其知识是零碎的、散乱的。如果不加以归纳、整理、综合，也就达不到知识共享、知识转化的目的。综合的过程就是把已获得 -的信息和知识进行排序、增减、分类、综合，存入知识数据库系统，并可以通过多种途径检索，随时随地查找所需的信息。同时，这个过程能够产生新的、更加系统化的知-识。

图9  知识管理的微观视角
（3）内化。内化的过程实质上是一个学习的过程。当通过社会化、外化、综合获得的集体知识，被内化成个人的隐性知识，并形成一种共享的心智模式和技术诀窍的时候-，它们才会变成有价值的知识。学习者个人是通过内化过程不断积累和丰富自己的知识的。
（4）社会化。社会化是一个共同分享各人的经验，转而创造新的隐性知识（增值），如共享心智模式、技能和诀窍的过程。
上述四种知识转化模式是前后连续、螺旋上升的。一方面，学习共同体通过充分的对话和共同的探索，将公共知识转化为个人知识；另一方面，将个人问题转化为公共问题-，并借助公共知识解决个人困境，这构成一种加速个人学习和创新的良性循环。
三、总结
从系统整体性、智能整体性和动态性、学习模式、知识管理这四个维度，来构建虚拟学习社区知识建构和集体智慧发展的学习构架，是必要和可行的途径：（1）整体性维-度。共性和差异性则是整体性的两个重要属性。共性是系统要素共同的、普遍的属性，差异性是系统要素的个体性、独特性与要素之间的差异性。从系统论视角（层次结构-性/多变性、关联性/协同性、交互性/动态性、系统/整体效应）研究虚拟学习社区，就是要使其诸要素产生"正的系统效应"，努力促进其集体智慧的显现和结晶。（-2）智能维度。从个体智能到集体智能是一个连续统。我们可以从多元智能、个体共性智能、合作智能和集体智能这几个上升的层次来分析虚拟学习社区的智能形态。而智-能动态性则显示集体智慧的螺旋上升周期可以分为四个阶段：即发散、收敛、凝聚和创新，相对于整体中个体智能的相互作用便显示为四种状态：即自我组织--发散、相-互连接--收敛、智慧结晶--凝聚、共同创造--创新。（3）学习维度。网络教育的学习模式主要有两种：一是个别化自主学习，另一种是协作学习。虚拟学习社区正-是实施这两种学习方式的最佳途径。（4）知识管理维度。本文从知识管理的宏观视角（个人知识管理、学习型组织、组织记忆和社区文化）考察了虚拟学习社区中的学习-技术，从知识管理的微观视角（外化、综合、内化、社会化）考察了虚拟学习社区中的知识转化过程。

A Learning Framework for Knowledge Building and Collective Wisdom
Advancement in Virtual Learning Communities

Yongcheng Gan1, Zhiting Zhu2

The goal of this paper is to provide an overview of technologies that can be applied to knowledge management and to assess their actual or potential contribution to the basic processes of knowledge creation and sharing within organizations. The aim is to identify trends and new developments that seem to be significant and to relate them to technology research in the field, rather than to provide a comprehensive review of available products.

Knowledge management (see, for example, Davenport and Prusak¹) is the name given to the set of systematic and disciplined actions that an organization can take to obtain the greatest value from the knowledge available to it. “Knowledge” in this context includes both the experience and understanding of the people in the organization and the information artifacts, such as documents and reports, available within the organization and in the world outside. Effective knowledge management typically requires an appropriate combination of organizational, social, and managerial initiatives along with, in many cases, deployment of appropriate technology. It is the technology and its applicability that is the focus of this paper.

To structure the discussion of technologies, it is helpful to classify the technologies by reference to the notions of tacit and explicit knowledge introduced by Polanyi in the 1950s^2,3 and used by Nonaka^4,5 to formulate a theory of organizational learning that focuses on the conversion of knowledge between tacit and explicit forms. Tacit knowledge is what the knower knows, which is derived from experience and embodies beliefs and values. Tacit knowledge is actionable knowledge, and therefore the most valuable. Furthermore, tacit knowledge is the most important basis for the generation of new knowledge, that is, according to Nonaka: “the key to knowledge creation lies in the mobilization and conversion of tacit knowledge.”⁵ Explicit knowledge is represented by some artifact, such as a document or a video, which has typically been created with the goal of communicating with another person. Both forms of knowledge are important for organizational effectiveness.⁶

These ideas lead us to focus on the processes by which knowledge is transformed between its tacit and explicit forms, as shown in Figure 1.⁵ Organizational learning takes place as individuals participate in these processes, since by doing so their knowledge is shared, articulated, and made available to others. Creation of new knowledge takes place through the processes of combination and internalization. As shown in Figure 1, the processes by which knowledge is transformed within and between forms usable by people are

• Socialization (tacit to tacit): Socialization includes the shared formation and communication of tacit knowledge between people, e.g., in meetings. Knowledge sharing is often done without ever producing explicit knowledge and, to be most effective, should take place between people who have a common culture and can work together effectively (see Davenport and Prusak,¹ p. 96). Thus tacit knowledge sharing is connected to ideas of communities and collaboration. A typical activity in which tacit knowledge sharing can take place is a team meeting during which experiences are described and discussed.
• Externalization (tacit to explicit): By its nature, tacit knowledge is difficult to convert into explicit knowledge. Through conceptualization, elicitation, and ultimately articulation, typically in collaboration with others, some proportion of a person's tacit knowledge may be captured in explicit form. Typical activities in which the conversion takes place are in dialog among team members, in responding to questions, or through the elicitation of stories.
• Combination: (explicit to explicit): Explicit knowledge can be shared in meetings, via documents, e-mails, etc., or through education and training. The use of technology to manage and search collections of explicit knowledge is well established. However, there is a further opportunity to foster knowledge creation, namely to enrich the collected information in some way, such as by reconfiguring it, so that it is more usable. An example is to use text classification to assign documents automatically to a subject schema. A typical activity here might be to put a document into a shared database.
• Internalization (explicit to tacit): In order to act on information, individuals have to understand and internalize it, which involves creating their own tacit knowledge. By reading documents, they can to some extent re-experience what others previously learned. By reading documents from many sources, they have the opportunity to create new knowledge by combining their existing tacit knowledge with the knowledge of others. However, this process is becoming more challenging because individuals have to deal with ever-larger amounts of information. A typical activity would be to read and study documents from a number of different databases.

Figure 1

These processes do not occur in isolation, but work together in different combinations in typical business situations. For example, knowledge creation results from interaction of persons and tacit and explicit knowledge. Through interaction with others, tacit knowledge is externalized and shared.⁷ Although individuals, such as employees, for example, experience each of these processes from a knowledge management and therefore an organizational perspective, the greatest value occurs from their combination since, as already noted, new knowledge is thereby created, disseminated, and internalized by other employees who can therefore act on it and thus form new experiences and tacit knowledge that can in turn be shared with others and so on.⁷ Since all the processes of Figure 1 are important, it seems likely that knowledge management solutions should support all of them, although we must recognize that the balance between them in a particular organization will depend on the knowledge management strategy used.⁸

Table 1 shows some examples of technologies that may be applied to facilitate the knowledge conversion processes of Figure 1. These technologies and others are discussed in this paper. The individual technologies are not in themselves knowledge management solutions. Instead, when brought to market they are typically embedded in a smaller number of solutions packages, each of which is designed to be adaptable to solve a range of business problems. Examples are portals, collaboration software, and distance learning software. Each of these can and does include several different technologies.

The approach to the technology of knowledge management in this paper emphasizes human knowledge. Sometimes in computer science “knowledge management” is interpreted to mean the acquisition and use of knowledge by computers, but that is not the meaning used here. In any case, automatic extraction of deep knowledge (i.e., in a form that captures the majority of the meaning) from documents is an elusive goal. Today the level of automatic extraction is deemed to be rather shallow because only a subset of the meaning, sometimes a very limited one, can be captured, ranging from recognition of entities such as proper names or noun phrases to automatic extraction of ontological relations of various kinds (e.g., References 9 and 10), and there is no system that can reason (in the sense of deducing something new from what it already knows) over the extracted knowledge in a way that even approaches the capabilities of a human. As an example of the current state of the art in applications for extracting knowledge automatically, Figure 2 shows a system¹¹ for analyzing reports of appellate court decisions to find the precedents they may affect. Court opinions are analyzed to find language that refers to other cases that the opinion may modify or invalidate. The candidate cases are retrieved from a database of law reports and are presented to an analyst for final judgment. The results are used to enrich the database with appropriate cross-references. Here the approach is that a template defines the fragment of knowledge to be sought, and the system tries to fill it by extracting information from the text. However, the candidate pieces of extracted knowledge must still be presented to a human for review and final decision, so that the value of the system is in increasing the productivity of the human analysts. For the foreseeable future, knowledge management in business will be about human knowledge in its various forms.

Figure 2

The use of technology in knowledge management is not new, and considerable experience has been built up by the early pioneers. Even before the availability of solutions such as Lotus Notes**¹² on which many contemporary knowledge management solutions are based, companies were deploying intranets, such as EPRINET,¹³ based on early generations of networking and computer technology that improved access to knowledge “on line.” Collaboration and knowledge sharing solutions also arose from the development of on-line conferencing and forums¹⁴ using mainframe computer technology. Today, of course, intranets and the Internet are ubiquitous, and we are rapidly approaching the situation where all the written information needed by a person to do his or her job is available on line. However, that is not to say that it can be used effectively with the tools currently available.

It is important to note that knowledge management problems can typically not be solved by the deployment of a technology solution alone. The greatest difficulty in knowledge management identified by the respondents in a survey¹⁵ was “changing people's behavior,” and the current biggest impediment to knowledge transfer was “culture.” Overcoming technological limitations was much less important. The role of technology is often to overcome barriers of time or space that otherwise would be the limiting factors. For example, a research organization divided among several laboratories in different countries needs a system that scientists with common interests can use to exchange information with each other without traveling, whereas a document management system can ensure that valuable explicit knowledge is preserved so that it can be consulted in the future. Two caveats must be stated at this point. First is the point made by Ackerman¹⁶ that in many respects the state of the art is such that many of the social aspects of work important in knowledge management cannot currently be addressed by technology. Ackerman refers to this situation as a “social technical gap.” Second, the coupling between behavior and technology is two-way: the introduction of technology may influence the way individuals work. People can and do adapt their way of working to take advantage of new tools as they become available, and this adaptation can produce new and more effective communication within teams (e.g., the effect of introducing solutions based on Lotus Notes on process teams in a paper mill described by Robinson et al.¹⁷ or the adaptations made by people in a customer support organization studied by Orlikowski¹⁸ after Notes was introduced).

Other surveys of technology for knowledge management can be found in the book, Working Knowledge by Davenport and Prusak,¹ and in a paper by Jackson.¹⁹ Prospects for using artificial intelligence (AI) techniques in knowledge management have been discussed recently by Smith and Farquhar.²⁰

In the following sections of this paper the technologies that support the processes of Figure 1 are described in more detail and illustrated with examples drawn largely from current research projects.

Tacit to tacit

The most typical way in which tacit knowledge is built and shared is in face-to-face meetings and shared experiences, often informal, in which information technology (IT) plays a minimal role. However, an increasing proportion of meetings and other interpersonal interactions use on-line tools known as groupware. These tools are used either to supplement conventional meetings, or in some cases to replace them. To what extent can these tools facilitate formulation and transfer of tacit knowledge?

Groupware. Groupware is a fairly broad category of application software that helps individuals to work together in groups or teams. Groupware can to some extent support all four of the facets of knowledge transformation. To examine the role of groupware in socialization we focus on two important aspects: shared experiences and trust.

Shared experiences are an important basis for the formation and sharing of tacit knowledge. Groupware provides a synthetic environment, often called a virtual space, within which participants can share certain kinds of experience; for example, they can conduct meetings, listen to presentations, have discussions, and share documents relevant to some task. Indeed, if a geographically dispersed team never meets face to face, the importance of shared experiences in virtual spaces is proportionally enhanced. An example of current groupware is Lotus Notes,¹² which facilitates the sharing of documents and discussions and allows various applications for sharing information and conducting asynchronous discussions to be built. Groupware might be thought to mainly facilitate the combination process, i.e., sharing of explicit knowledge. However, the selection and discussion of the explicit knowledge to some degree constitutes a shared experience.

A richer kind of shared experience can be provided by applications that support real-time on-line meetings—a more recent category of groupware. On-line meetings can include video and text-based conferencing, as well as synchronous communication and chat. Text-based chat is believed to be capable of supporting a group of people in knowledge sharing in a conversational mode.²¹ Commercial products of this type include Lotus Sametime** and Microsoft NetMeeting**. These products integrate both instant messaging and on-line meeting capabilities. Instant messaging is found to have properties between those of the personal meeting and the telephone: it is less intrusive than interrupting a person with a question but more effective than the telephone in broadcasting a query to a group and leaving it to be answered later.

In work on the Babble system,²² chat was evaluated by at least some users as being “… much more like conversation,” which is promising for the kind of dialog in which tacit knowledge might be formed and made explicit. However, not all on-line meeting systems have the properties of face-to-face meetings. For example, the videoconferencing system studied by Fish et al.²³ was judged by its users to be more like a video telephone than like a face-to-face meeting. Currently, rather than replacing face-to-face meetings, many on-line meetings are found to complement existing collaboration systems and the well-established phone conference and are therefore probably more suited to the exchange of explicit rather than tacit knowledge. On-line meetings extend phone conferences by allowing application screens to be viewed by the participants or by providing a shared whiteboard. An extension is for part of the meeting to take place in virtual reality with the participants represented by avatars.²⁴ One research direction is to integrate on-line meetings with classic groupware-like applications that support document sharing and asynchronous discussion. An example is the IBM-Boeing TeamSpace project,²⁵ which helps to manage both the artifacts of a project and the processes followed by the team. On-line meetings are recorded as artifacts and can be replayed within TeamSpace, thus allowing even individuals who were not present in the original meeting to share some aspects of the experience.

Some of the limitations of groupware for tacit knowledge formation and sharing have been highlighted by recent work on the closely related issue of the degree of trust established among the participants.²⁶ It was found that videoconferencing (at high resolution—not Internet video) was almost as good as face-to-face meetings, whereas audio conferencing was less effective and text chat least so. These results suggest that a new generation of videoconferencing might be helpful in the socialization process, at least in so far as it facilitates the building of trust. But even current groupware products have features that are found to be helpful in this regard. In particular, access control, which is a feature of most commercial products, enables access to the discussions to be restricted to the team members if appropriate, which has been shown²² to encourage frankness and build trust.

Another approach to tacit knowledge sharing is for a system to find persons with common interests, who are candidates to join a community. In Foner's Yenta System,²⁷ the similarity of the documents used by people allowed the system to infer that their interests were similar. Location of other people with similar interests is a function that can be added to personalization systems, the goal of which is to route incoming information to individuals interested in it. There are obvious privacy problems to overcome.

Expertise location. Suppose one's goal is not to find someone with common interests but to get advice from an expert who is willing to share his or her knowledge. Expertise location systems have the goal of suggesting the names of persons who have knowledge in a particular area. In their simplest form, such systems are search engines for individuals, but they are only as good as the evidence that they use to infer expertise. Some possible sources of such evidence are shown in Table 2.

The problem with using an explicit profile is that persons may not be motivated to keep it up to date, since to them it is just another form to fill in. Thus it is preferable to gather information automatically, if possible, from existing sources. For example, a person's resume or a list of the project teams that he or she has worked on may exist in a company database. Another automatic approach is to infer expertise from the contents of documents with which a person's name is associated. For example, authorship (creation or editing) of a document presumably indicates some familiarity with the subjects it discusses, whereas activities such as reading indicate some interest in the subject matter. Two approaches to using document evidence for expertise location suggest themselves: either the documents can be classified according to some schema, thus classifying their authors; or when a user submits a query to the expertise location system, it searches the documents, transforms the query to a list of authors (suitably weighted), and returns the list as the result of the expertise search.

The current state of the art is to use the first three sources of evidence listed in Table 2: explicit profiles, evidence mined from existing databases, and evidence inferred from association of persons and documents. For example, the Lotus Discovery Server** product contains a facility whereby an individual's expertise is determined using these techniques,²⁸ while it and the Tacit Knowledge Systems KnowledgeMail** product²⁹ analyze the e-mail a person writes to form a profile of his or her expertise. Given the properties of on-line discussions, discussed below, it is reasonable to suppose that a fourth source of evidence could be the content of the questions answered by a person in such a system, with the added advantage that such a person is already willing to be helpful. This example is a simple case of the social interaction dimension in expertise location which, as found in empirical studies (e.g., Reference 30), is an important factor but is not yet reflected in available applications, perhaps because of the difficulty of capturing aspects such as the expert's communication skills, in order to rate how useful he or she is likely to be.

Tacit to explicit

According to Nonaka, the conversion of tacit to explicit knowledge (externalization) involves forming a shared mental model, then articulating through dialog. Collaboration systems and other groupware (for example, specialized brainstorming applications³¹) can support this kind of interaction to some extent.

On-line discussion databases are another potential tool to capture tacit knowledge and to apply it to immediate problems. We have already noted that team members may share knowledge in groupware applications. To be most effective for externalization, the discussion should be such as to allow the formulation and sharing of metaphors and analogies, which probably requires a fairly informal and even freewheeling style. This style is more likely to be found in chat and other real-time interactions within teams.

Newsgroups and similar forums are open to all, unlike typical team discussions, and share some of the same characteristics in that questions can be posed and answered, but differ in that the participants are typically strangers. Nevertheless, it is found that many people who participate in newsgroups are willing to offer advice and assistance, presumably driven by a mixture of motivations including altruism, a wish to be seen as an expert, and the thanks and positive feedback contributed by the people they have helped.

Within organizations, few of the problems experienced on Internet newsgroups are found, such as flaming, personal abuse, and irrelevant postings. IBM's experience in this regard is described by Foulger.¹⁴ Figure 3 shows a typical exchange in an internal company forum, rendered here using a standard newsgroup browsing application. It illustrates how open discussion groups are used to contribute knowledge in response to a request for help. Note both the speed of response and the fact that the answerer has made other contributions previously. The archive of the forum becomes a repository of useful knowledge. Clearly the question answerer in this case has made a number of contributions and could be considered to be an expert. Although the exchange is superficially one of purely explicit knowledge, the expert must first make a judgment as to the nature of the problem and then as to the most likely solution, both of which bring his or her tacit knowledge into play. Once the knowledge is made explicit, persons with similar problems can find the solution by consulting the archive. A quantitative study³² of this phenomenon in the IBM system showed that the great majority of interchanges were of this question-and- answer pattern, and that even though a large fraction of questions were answered by just a few persons, an equal proportion were answered by persons who only answered one or two questions. Thus the conferencing facility enabled knowledge to be elicited from the broad community as well as from a few experts.

Figure 3

Explicit to explicit

There can be little doubt that the phase of knowledge transformation best supported by IT is combination, because it deals with explicit knowledge. We can distinguish the challenges of knowledge management from those of information management by bearing in mind that in knowledge management the conversion of explicit knowledge from and to tacit knowledge is always involved. This leads us to emphasize new factors as challenges that technology may be able to address.

Capturing knowledge. Once tacit knowledge has been conceptualized and articulated, thus converting it to explicit knowledge, capturing it in a persistent form as a report, an e-mail, a presentation, or a Web page makes it available to the rest of the organization. Technology already contributes to knowledge capture through the ubiquitous use of word processing, which generates electronic documents that are easy to share via the Web, e-mail, or a document management system. Capturing explicit knowledge in this way makes it available to a wider audience, and “improving knowledge capture” is a goal of many knowledge management projects. One issue in improving knowledge capture is that individuals may not be motivated to use the available tools to capture their knowledge. Technology may help by improving their motivation or by reducing the barriers to generating shareable electronic documents.

One way to motivate people to capture knowledge is to reward them for doing so. If rewards are to be linked to quality rather than quantity, some way to measure the quality of the output is needed. Quality in the abstract is extremely difficult to assess, since it depends on the potential use to which the document is to be put. For example, a document that explains basic concepts clearly would be useful for a novice but useless to someone who is already an expert. If we focus on usefulness as a measure of quality, and if we substitute “use” for “usefulness,” then we have something that IT systems can measure. In fact, portal infrastructures that mediate access to documents can easily accumulate metrics of document use, and hence can estimate usefulness and quality. The next generation of products will include such features.²⁸

Another measure of quality is the number of times a document has been cited, as in the scholarly literature, or the number of times it has been hyperlinked to, as on the Internet. A citation or hyperlink is evidence that the author of the citing or linking document thought that the target document is valuable. The most valuable or authoritative documents can be detected in Internet applications by analyzing the links between Web pages, thus measuring the cumulative effects of numerous value judgments (e.g., see References 33 and 34). The numeric quality estimate that can be derived is useful in information retrieval, where it can be used to boost the position of high-quality documents in the search results list. This method has been applied to citation analysis in scientific papers by the ResearchIndex search engine^35,36 and to Web search by the Google search engine.³⁷

Citation analysis of this kind detects quality assessments made in the course of authoring documents. Quality judgments by experts are another way to capture their knowledge. There are, of course, many deployed solutions in which documents undergo a quality review through a refereeing process, often facilitated by a workflow application. In this case, the quality judgment acts as a gate, and documents judged to be of low quality are not distributed. However, technology also makes it feasible to record judgments as annotations of existing documents.³⁸ Here, the association of an annotation with a document is recorded in some infrastructure, such as a special annotation server that the user's browser accesses to find annotations of the Web page being viewed. Numeric data stored in databases can also be annotated³⁹ to record various interpretations, judgments, or cautions. Annotations may also support collaboration around documents,⁴⁰ although, as in other applications where the underlying documents may be altered, the annotation system needs to be robust in the face of changes.

Although the most common way to capture knowledge by far is to write a document, technology has made the use of other forms of media feasible. Digital audio and video recordings are now easily made, and an expert may find that speaking to a camera or microphone is easier or more convenient than writing, particularly if the video is of a presentation that has to be made in the ordinary course of business, or if the audio recording can be made in an otherwise unproductive free moment. It is also now relatively easy to distribute audio and video over networks. However, nontext digital media have the disadvantage of being more difficult to search and to browse than text documents and, hence, are less usable as materials in a repository of knowledge. Browsing of video has been improved by summarization techniques that automatically produce a gallery of extracted still images, each of which represents a significant passage in the video.⁴¹ If the video is of someone giving a presentation, images of the speaker alone will not convey as much as a summary that includes images of any visual aids, such as slides or charts, that accompany the narrative. Several systems that key a recording of a presentation to the slides have been described.^42-44

Although video searching systems have been built that use image searching⁴⁵ of extracted frames,^46,47 they are hampered by the difficulty of composing a semantically meaningful image query. A more fruitful approach to searching is to extract text from the multimedia object, if possible. Although in some cases the video may contain text (on images of text slides), in most cases the challenge is to convert speech to text.

Speech recognition. Improvements in the accuracy of automatic speech recognition (ASR) hold out the promise of usable speaker-independent recognition with unconstrained vocabulary in the foreseeable future. Figure 4 shows progress with time in a number of standardized speech recognition tasks. Word error rates were reported in the Speech Recognition Workshop conferences of the National Institute of Standards and Technology. The accuracy varies with the difficulty of the task. The resource management task involves reading speech with a 1000-word vocabulary. Broadcast news uses recordings with an approximately 20K word vocabulary, whereas the CallHome and switchboard are telephone (lower speech quality) recognition tasks with unconstrained vocabulary. In all cases the accuracy shows steady improvement with time.

Figure 4

Accuracy for speech recorded under controlled conditions is already acceptable, but the error rate for poor quality recordings (for example, from the telephone) is still high enough to cause problems for applications unless the vocabulary is constrained. However, the trends depicted in Figure 4 show that future improvements can reasonably be expected and will lead to new ways to capture knowledge.

Although perfect or near-perfect transcription produces a text transcript that can be browsed like any other piece of text, ways to make an imperfect transcript usable as a browsing aid are being investigated.^48,49 In this work even an imperfect transcript supports browsing because certain words and phrases, which are judged to be significant and for which the estimated accuracy of ASR is high, are highlighted. Such techniques can be used to make the replay of audio more usable even where the transcript as a whole is unreadable because of the density of errors. The highlights can be used to find the passage of interest.

Search. The most important technology for the manipulation of explicit knowledge helps people with the most basic task of all: finding it. Since the trend in most organizations is for essentially all documents to become available in electronic form on line, the challenge of on-line access has been transformed into the challenge of finding the materials relevant for some task. Furthermore, the total amount of potentially relevant information, including what is on the Internet and company intranets and what is available from commercial on-line publishers, continues to grow rapidly. Thus text search, which only 10 years ago was a tool primarily used by librarians to search bibliographic databases, has become an everyday application used by almost everyone. Not surprisingly, the new uses of text search have motivated new work on the technology.

Another driving factor in the use of on-line explicit knowledge is the diversity of sources from which it is available. It is not uncommon for users to have to look in several databases or Web sites for potentially relevant information. Since there is little standardization, users have to cope with different user interfaces, different search language conventions, and different result list presentations. Portals—described in another paper in this issue⁵⁰—are a popular approach to reducing the complexity of the user's task. The key aspect that allows a portal to do this is that it maintains its own meta-data about the information to which it gives access. In the current state of the art, the meta-data may be quite simple, consisting of a list of sources and a search index formed from the content of the sources. Even this simple function provides great value because it relieves the user of the need to visit all the sources to find out whether they contain relevant information. The user is therefore made more productive, and the quality of his or her work is improved. Most portal systems use a single search index, which requires that the documents in the domain of interest have to be retrieved by “spidering” or “crawling” at indexing time. The alternative, using distributed search as in, for example, the Harvest project,⁵¹ has not proved to be popular for knowledge management applications, perhaps because advances in hardware have made it cheaper to build a central index. Recent developments in peer-to-peer applications, such as Gnutella⁵² and the collaboration application Groove,⁵³ have promoted a new interest in distributed search, which may lead to new advances.

The index that is built by a text search engine consists of a list of the words that occur in the indexed documents, along with a data structure (the inverted file) that allows the documents in which the words occurred to be determined efficiently at search time.⁵⁴ Users can therefore use query words that they expect to occur in the documents. The problem is that not all the documents will use the same words to refer to the same concept and, therefore, not all the documents that discuss the concept will be retrieved. In a world of information overload this situation is not usually a problem, but for applications where it is important to have high recall, an alternative approach can be used in which documents are assigned meta-data that describe the concepts they discuss in a controlled vocabulary. This is a classical approach used in bibliographic databases. However, where searches are being done by untrained end users rather than librarians, the evidence is that searching with natural language gives better results than does searching with a controlled vocabulary.⁵⁵

The most common problem in a search is that a query retrieves many documents that are irrelevant to the user's needs, known as the problem of search precision (a measure of accuracy). Precision is of paramount importance in a world of “info-glut.” However, results from TREC (Text REtrieval Conference)⁵⁶ indicate that the accuracy of natural language search engine technology has reached a plateau in recent years. What are the prospects of improvements to the search function that will benefit knowledge management systems? Two areas of potential improvement can be identified: increased knowledge of the user and of the context of his or her information need, and improved knowledge of the domain being searched.

The notion that increased knowledge of the user can be beneficial comes from the realization that in almost all search systems today the only information about the user's information need that is available to the system is the query. The most common query submitted to Web-based search services is two words, and the average query length is only about 2.3 words.⁵⁷ Obviously, this amount is not much information. A challenging research area is to gather better information about the context of a search and to build search engines that can use this information to good advantage.

The goal of gathering and using more information about the domain being searched is one that is well-established, but progress so far has been limited. It is common to use a thesaurus—a kind of simple domain model—as an adjunct to a search, although this is more common in systems designed for specialists. Expansion of a query with synonyms is known to improve the recall in a text search, but expansion is only effective in well-defined domains where the ambiguity of words, and the validity of term relationships, is not an issue. To improve precision in broad-domain searching by reducing the ambiguity of ordinary words using thesauri or other structures such as ontologies has been a goal of much research, with many negative results (e.g., Reference 58). Recently, however, some encouraging findings have been obtained.⁵⁴ Using WordNet⁵⁹ (a large manually built thesaurus that is widely available), combined with automatically built data structures encoding co-occurrence and head-modifier relations, Mandala et al.⁶⁰ showed significant improvements in average precision, a measure of accuracy, as shown in Figure 5. The results were obtained using TREC data, from queries derived from the search topics using the title field, the title and description fields, or all the fields in the topic. Woods et al.⁶¹ also reported improvements by using a different approach to encoding knowledge of the domain, in this case a semantic network that integrated syntactic, semantic, and morphological relationships.

Figure 5

Taxonomies and document classification. Knowledge of a domain can also be encoded as a “knowledge map,” or “taxonomy,” i.e., a hierarchically organized set of categories. The relationships within the hierarchy can be of different kinds, depending on the application, and a typical taxonomy includes several different kinds of relations. The value of a taxonomy is twofold. First, it allows a user to navigate to documents of interest without doing a search (in practice, a combination of the two strategies is often used if it is available). Second, a knowledge map allows documents to be put in a context, which helps users to assess their applicability to the task in hand. The most familiar example of a taxonomy is Yahoo!,⁶² but there are many examples of specialized taxonomies used at other sites and in company intranet applications.

Manually assigning documents to the categories in a taxonomy requires significant effort and cost, but in recent years automatic document classification has advanced to the point where the accuracy of the best-performing algorithms exceeds 85 percent (F₁ measure) on good quality data.⁶³ This degree of accuracy is adequate for many applications and is in fact comparable to what can be achieved by manual classifiers in a well-organized operation,⁶⁴ although the accuracy of automatic classification over different types of data varies quite widely.⁶⁵ An attractive feature of the current generation of automatic classifiers is their inclusion of machine-learning algorithms that train themselves from example data, whereas the previous generation required construction of a complex description of the category in the form, for example, of an elaborate query. Selecting documents as training examples is a simpler task.

Automatic classification, although simple in concept, is capable of surprisingly refined distinctions, given enough training data. For example, it has been known for some time (see the brief review in Kukich⁶⁶) that automatic essay marking systems can assign grades to student essays with an accuracy and consistency only slightly worse than human graders, and recently it has been shown that a document classifier can perform well in this application.⁶⁷ Table 3 shows the results of comparing two human graders and an automatic classifier. The automatic classifier performed very nearly as well as the human graders, both in accuracy and consistency, even though the test essays were on unconstrained subjects.

Despite the power of automatic classification, there are many challenges in implementing solutions using taxonomies. The first challenge is the design of the taxonomy, which has to be comprehensible to users (so that they can use it for navigation with no or minimal training) and has to cover the domain of interest in enough detail to be useful. There are a number of strategies for building a taxonomy,⁶⁸ including the use of document clustering to propose candidate subcategories. However, human input is probably required to ensure that the taxonomy reflects business needs (e.g., it emphasizes some aspect that may be significant but is not a strong theme in the documents). Thus, clustering can be seen as an adjunct to human effort. One usability challenge is to ensure that the user of a taxonomy editor can understand the clusters that are proposed, using automatically generated labels. The labels typically contain words or phrases that are chosen to represent the documents in the cluster; recently a technique for using extracted sentences has been proposed.^69,70

Taxonomies have proved to be a popular way in which to build a domain model to help users to search and navigate, so much so that the trend seems to be for each group of users of any size to have their own taxonomy. This popularity is understandable because as on-line tools become central to individuals' work, they naturally want to see the information displayed within a schema that reflects their own priorities and worldview, and that uses the terminology that they use. This trend is likely to lead to a proliferation of taxonomies in knowledge management applications. It follows that there will be an increasing focus on the need to map from one taxonomy to another so as to bridge between the schemas used by different groups within an organization.

Portals and meta-data. As already mentioned, portals provide a convenient location for the storage of meta-data about documents in their domain, and two examples of such meta-data, search indexes and a knowledge map or taxonomy, have been discussed. In the future, increasing use of natural language processing (NLP) in portals is likely to generate new kinds of meta-data. The general trend is for more structured information—meta-data—to be automatically generated as part of the indexing service of the portal. It is efficient to generate these meta-data when the document has been retrieved for text indexing. The value of the meta-data is in encapsulating information about the document that can be used to build selected views of the information space, such as a list of the documents in a given subject category, or mentioning a geographic location, through a database lookup in response to a user click. This makes exploration of the information easier and more rewarding, in effect providing the user with a new experience based on the exploration on which new tacit knowledge can be built as part of the internalization process to be discussed later.

Summarization. Document summaries are examples of meta-data of this kind. The value of a summary is that it allows users to avoid reading a document if it is not relevant to their current tasks. Figure 6 shows results from Tombros and Sanderson⁷¹ who showed that users performing a simple information-seeking task had to read many fewer full documents when they used a system that provided summaries than when the system provided document titles alone. Automatic generation of summaries is an active area of research. Commercially available summarizers use the sentence-selection method, originated by Luhn in 1958,⁷² in which an indicative summary is constructed from what are judged to be the most salient sentences in a document. However, the summary may be incoherent, e.g., if the selected sentences contain anaphors. Construction of more coherent summaries, implying the use of natural language generation, currently requires that the subject domain of the documents be severely restricted, as for example, to basketball games.⁷³ Summarization of long documents containing several topics is improved by topic segmentation⁷⁴ and can be further condensed for presentation on handheld devices,⁷⁵ whereas summarization of multiple documents, either about the same event⁷⁶ or in an unconstrained set of domains,⁷⁰ is another challenge being addressed by current research. For other recent work see References 77 through 79.

Figure 6

Explicit to tacit

Technology to help users form new tacit knowledge, for example, by better appreciating and understanding explicit knowledge, is a challenge of particular importance in knowledge management, since acquisition of tacit knowledge is a necessary precursor to taking constructive action. A knowledge management system should, in addition to information retrieval, facilitate the understanding and use of information. For example, the system might, through document analysis and classification, generate meta-data to support rapid browsing and exploration of the available information. It seems likely that the future trend will be for information infrastructures to perform more of this kind of processing in order to facilitate different modes of use of information (e.g., search, exploration, finding associations) and thus to make the information more valuable by making it easier to form new tacit knowledge from it. Other processing of explicit knowledge, already described, can support understanding. For example, putting a document in the context of a subject category or of a step in a business process, by using document categorization, can help a user to understand the applicability or potential value of its information. Discovery of relationships between and among documents and concepts helps users to learn by exploring an information space.

A quite different set of technologies applies to the formation of tacit knowledge through learning, especially in the domain of on-line education or distance learning. Within organizations, on-line learning has the advantage of being able to be accomplished without travel and at times that are compatible with other work. A wide variety of tools and applications support distance learning.⁸⁰ The needs of the corporate training market, emphasizing self-directed learning rather than instructor-led learning, have led to a focus on interactive courseware based on the Web or on downloaded applications. In the future, modules of self-directed training will be found in portals, along with other materials.

Information overload is a trend that motivates the adoption of new technology to assist in the comprehension of explicit knowledge. The large amounts of (often redundant) information available in modern organizations, and the need to integrate information from many sources in order to make better decisions, cause difficulties for knowledge workers and others.⁸¹ Both of these trends result directly from the large amounts of on-line information available to knowledge workers in modern organizations. Information overload occurs when the quality of decisions is reduced because the decision maker spends time reviewing more information than is needed, instead of reflecting and making the decision. Various approaches to mitigating information overload are feasible. The redundancy and repetition in the information can be reduced by eliminating duplicate or overlapping messages (related to the Topic Detection and Tracking track at TREC⁸²). An agent can filter or prioritize the messages, or compound views can make it easier to review the incoming information. Finally, visualization techniques can be applied in an attempt to help the user understand the available information more easily.

Different visualizations of a large collection of documents have been used with the goal of making subject-based browsing and navigation easier. These methods include text-based category trees, exemplified by the current Yahoo! user interface. Several graphical visualizations have also been described. Themescape⁸³ uses (among other things) a shaded topographic map as a metaphor to represent the different subject themes (by location), their relatedness (by distance), and the proportional representation of the theme in the collection (by height), whereas VisualNet⁸⁴ uses a different map metaphor for showing subject categories. Another approach is represented by the “Cat-a-Cone” system⁸⁵ that allows visualization of documents in a large taxonomy or ontology. In this system the model is three-dimensional and is rendered using forced perspective. Search is used to select a subset of the available documents for visualization.

Other visualization experiments have attempted to provide a user with some insight into which query terms occur in the documents in a results list, as was done in Hearst's TileBars⁸⁶ and the application described by Veerasamy and Belkin.⁸⁷ However, the evaluation described in the latter paper showed that the advantage of the visualization in the test task was small at best. A later study,⁸⁸ which compared text, two-dimensional, and pseudo three-dimensional interfaces for information retrieval, found that the richer interfaces provided no advantage in the search tasks that were studied. This result may explain why graphical visualization has not been widely adopted in search applications, whereas text-based interfaces are ubiquitous.

Perhaps a more promising application of visualization is to help a user grasp relationships, such as those between concepts in a set of documents as in the Lexical Navigation system described by Cooper and Byrd⁸⁹ or the relationships expressed as hyperlinks between documents.⁹⁰ This use is more promising because of the difficulty of rendering relationships textually. Furthermore, figuring out the relationships within a set of documents is a task that requires a lot of processing, and computer assistance is of great value.

Conclusion

This paper has surveyed a number of technologies that can be applied to build knowledge management solutions and has attempted to assess their actual or potential contributions to the processes underlying organizational knowledge creation using the Nonaka model. The essence of this model is to divide the knowledge creation processes into four categories: socialization (tacit knowledge formation and communication), externalization (formation of explicit knowledge from tacit knowledge), combination (use of explicit knowledge), and internalization (formation of new tacit knowledge from explicit knowledge). The value of this model in the present context is that it focuses attention on tacit knowledge (which is featured in three of the four processes) and thus on people and their use of technology.

Because all four of the processes in the Nonaka model are important in knowledge management, which aims to foster organizational knowledge creation, we might seek to support all of them with technology. Although early generations of knowledge management solutions (solutions typically integrate several technologies) focused on explicit knowledge in the form of documents and databases, there is a trend to expand the scope of the solutions somewhat to integrate technologies that can, to some extent, foster the use of tacit knowledge. Among these technologies now being applied in some knowledge management solutions are those for electronic meetings, for text-based chat, for collaboration (both synchronous and asynchronous), for amassing judgments about quality, and for so-called expertise location. These technologies are in addition to those for handling documents, such as search and classification, which are already well-established yet are still developing.

Despite these trends, there are still significant shortfalls in the ability of technology to support the use of tacit knowledge—for which face-to-face meetings are still the touchstone of effectiveness. As Ackerman has pointed out, this lack of ability is not just because the designers of the applications do not appreciate how important the human dimension is (although that is true in some cases). We simply do not understand well enough how to accommodate this dimension in computer-supported cooperative work. Many of the factors that mediate effective face-to-face human-human interactions are not well understood, nor do we have good models for how they might be substituted for or synthesized in human-computer interactions. We can expect gradual progress in this direction, perhaps aided by improvements in the general fidelity with which people's faces, expressions, and gestures are rendered in (for example) high-bandwidth videoconferencing, but there can be no assurance of an immediate breakthrough because of the complexity of the problem and the current shortfall in the basic understanding of its elements.

However, the survey in this paper has highlighted many factors that provide grounds for some optimism when we consider how technology can help in knowledge management. Technology can assist teams, who in today's world may meet only occasionally or even never, to share experiences on line in order to be able to build and share tacit knowledge, and more generally to work effectively together, even if the efficiency is less than in face-to-face meetings. From the perspective of tacit knowledge formation and sharing, the relative informality of text-based chat is probably superior to more structured discussions, which may, however, be effective for sharing explicit knowledge. The importance of limiting access to team members has been highlighted by recent work. The chat archive, and other recordings of on-line meetings, have the added advantage of being able to help in the socialization of people who miss parts of the original interaction. It is also encouraging that recent work by Olson and Olson and their collaborators has shown that studio-quality video is helpful in some tasks related to knowledge management, such as collaboration (in some cases) and trust building.²⁶

Another encouraging use of technology is to help persons who need to share knowledge to find each other. Expertise location systems are in their infancy in industrial practice but hold out the promise of being able to identify individuals with the right knowledge. Even without actually identifying a person, unrestricted forums and bulletin boards have been shown to be effective in eliciting assistance both from experts and from the broader community. It seems likely that appropriate integration of this approach with chat on the one hand and expertise location on the other will result in more effective access to and communication of the knowledge in an organization.

Another way to tap the knowledge of experts is through capturing their judgments, expressed as annotation, hyperlinks, citations, and other interactions with documents. Portal infrastructures, which mediate and can collect metrics on the interaction of people and documents, are ideal for amassing this kind of information. Currently, portal products are just becoming capable of accumulating meta-data of this kind. Another trend is for their meta-data to become richer and to support a broader range of tasks. In particular, the meta-data can support the formation of new tacit knowledge from the explicit knowledge indexed by the portal, for example, by situating documents within a new conceptual framework represented by a knowledge map. It is becoming cheaper to use several different frameworks for this purpose, and thus to match them better to the needs of different groups of users, because the accuracy of automatic text classification is improving and, for some classes of content such as news stories, is already as good as the accuracy of human indexers.

Technology will clearly become more helpful in dealing with information overload. Techniques such as summarization can reduce the load of persons attempting to find the right documents to use in some task. There is some promise, as yet unfulfilled, that intelligent agents may in the future help persons to prioritize the messages they receive. And the meta-data stored by portals can be used to draw visualizations of large amounts of information, although, contrary to intuition, graphical visualizations seem not to be better than their text-based equivalents, at least for information retrieval tasks.

Finally, it should be emphasized again that this paper has dealt with human knowledge, not with the formation or use of expert systems or similar knowledge-based systems that aim to replace human reasoning with machine intelligence. The current capability of machine intelligence is such that, for the great majority of business applications, human knowledge will continue to be a valuable resource for the foreseeable future, and technology to help to leverage it will be increasingly valuable and capable.

**Trademark or registered trademark of Lotus Development Corporation, Microsoft Corporation, or Tacit Knowledge Systems.

Cited references

Accepted for publication June 15, 2001.

一 生平简介
亚伯拉罕·马斯洛(Abraham Harold Maslow, 1908-1970) 出生于纽约市布鲁克林区。美国社会心理学家、人格理论家和比较心理学家，人本主义心理学的主要发起者和理论家,心理学第三势力的领导人。1926年入康乃 尔大学，三年后转至威斯康辛大学攻读心理学，在著名心理学家哈洛的指导下，1934年获得博士学位。之后，留校任教。1935年在哥伦比亚大学任桑代克学 习心理研究工作助理。1937年任纽约布鲁克林学院副教授。1951年被聘为布兰戴斯大学心理学教授兼系主任。1969年离任，成为加利福尼亚劳格林慈善 基金会第一任常驻评议员。第二次世界大战后转到布兰代斯大学任心理学教授兼系主任，开始对健康人格或自我实现者的心理特征进行研究。曾任美国人格与社会心 理学会主席和美国心理学会主席（1967），是<<人本主义心理学>>和<<超个人心理学>>两个杂志 的首任编辑。
主要著作有：《动机与人格》（1954）、《存在心理学探索》（1962）、<<宗教、价值观和高峰体验>>（1964）、《科学心理学》（1967）、《人性能达的境界》（1970）等。
二 人格自我实现论
（一）需要层次
按马斯洛的理论，个体成长发展的内在力量是动机。而动机是由多种不同性质的需要所组成，各种需要之间，有先后顺序与高低层次之分；每一层次的需要与满足，将决定个体人格发展的境界或程度。
1 生理需要 (physiological need)
生存所必须的基本生理需要，如对食物，水和睡眠和性的需要。
2 安全需要 (safety need)
包括一个安全和可预测的环境，它相对地可以免除生理和心理的焦虑。
3 爱与归属的需要 (love and belongingness need)
包括被别人接纳、爱护、关注、鼓励、支持等，如结交朋友，追求爱情，参加团体等。
4 尊重需要 (esteem need)
包括尊重别人和自尊重两个方面。
5 自我实现需要（self-actualization need）
包括实现自身潜能。
在心理学上，需要层次论是解释人格的重要理论，也是解释动机的重要理论。
（二）自我实现
自我实现是马斯洛人格理论的核心。他认为可以将其“定义为不断实现潜能、智能和天资，定义为完成天职或称之为天数、命运或禀性，定义为更充分的认 识、承认了人的内在天性，定义为在个人内部不断趋向统一、整合或协同动作的过程”。也就是说，个体之所以存在，之所以有生命意义，就是为了自我实现。马斯 洛对自己的学生进行抽样调查，并对历史上和当时仍然健在的著名人物，如斯宾诺莎、贝多芬、歌德、爱因斯坦、林肯、杰弗逊、罗斯福等人进行个案研究，概括出 了自我实现的人所共同具有的人格特征。
1、对现实更有效的洞察力和更适意的关系
2、对自我、他人和自然的接受
3、行为的自然流露
4、以问题为中心
5、超然的独立性：离辟独居的需要
6、自主性：对文化与环境的独立性；意志；积极的行动者
7、体验的时时常新
8、社会感情
9、自我实现者的人际关系
10、民主的性格结构
11、区分手段与目的、善与恶
12、富有哲理的、善意的幽默感
13、创造力
14、对文化适应的对抗
（三）高峰体验
高峰体验是自我实现的短暂时刻，只有在生活中经常产生高峰体验，才能顺利地达到自我实现。
马斯洛在阐述高峰体验时认为：“这种体验是瞬间产生的，压倒一切的敬畏情绪，也可能是转瞬即逝的极度强烈的幸福感，或甚至是欣喜若狂、如痴如醉、 欢乐至极的感觉。”许多人都声称自己在这种体验中仿佛窥见了终极的真理、人生的意义和世界的奥秘。人们好像是经过长期的艰苦努力和紧张奋斗而达到了自己的 目的地。
“这些美好的瞬间来自爱情，和异性的结合，来自审美感觉，来自创造冲动和创造激情，来自意义重大的领悟和发现，来自女性的自然分娩和对孩子的慈爱，来自与大自然的交融……”
这种高峰体验可能发生于父母子女的天伦情感之中，也可能在事业获得成就或为正义而献身的时刻，也许在饱览自然、浪迹山水的那种“天人合一”的刹那。
三 教育心理学思想
马斯洛的思想以人性本善为前提。强调教育的功能，教育的目的——人的目的，人本主义的目的，追根究底就是人的自我实现，是人所能达到的最高度的发 展，即帮助人达到他能够达到的最佳状态。在马斯洛看来，人具有一种与生俱来的潜能，发挥人的潜能，超越自我是人的最基本要求。环境具有促使潜能得以实现的 作用。然而，并非所有的环境条件都有助于潜能的实现，只有在一种和睦的气氛下，在一种真诚、信任和理解的关系中，潜能才能像得到了充足阳光和水分的植物一 样蓬勃而出。为了使儿童健康成长，应当充分信任他们和信赖成长的自然过程，即不过多干扰，不揠苗助长或强迫其完成预期设计，不以专制的方式，而是以道家的 方式让他们自然成长和帮助他们成长。
马斯洛把完善的人性教育作为人本教育的基本内容。通常人在低级需要获得满足后即追求高级需要的满足。自我实现追求的内容是实现人的“内在价 值”，包括真理、美、新颖、独特、公正、严密、简洁、善、效率、爱、诚实、单纯、改善、秩序、文雅、成长、清洁、宁静、和平等。如果能实现这些价值，便可 以达到人生最大的幸福和快乐。马斯洛关于教育原则的思想大致可概括为五个方面：
1 自我同一性原则
教育应该使学生寻找内在的同一性，减少或消除学生内心的矛盾和精神上的分裂。并且帮助学生认识到自我与非我的统一，即个人与社会和自然的统一。
2 启发性原则
此原则主要是为了激发和培养学生的创造性。不仅要通过知识教育来培养学生的理性控制、逻辑思维等，更重要的是通过情感交流、优美人格的形成、自我的充分自由的体现等活动来激发学生的非智力因素。
3 美育原则
重视音乐、舞蹈、美术等艺术教育。
4 超越性原则
对自我的超越，即超越自私，超越自我中心，从而达到忘我的境界。此外，还有对文化的超越，能够以某种超脱的和客观的态度对自己出身于其中的文化进行审查，就是要培养一种具有批判精神的人。
5 价值原则
教育应使学生获得价值感，应该挖掘、激发学生的内在价值，使受教育者获得生存的意义。
四 以人为本的管理
马斯洛很早即关注人的因素。当美国60年代的管理大师杜拉克、麦格莱高等都将注意力集中于工业化的工作场地时，马斯洛最早认识到人本管理的重要 性，他说：“工业领域也许能够当作心理动力学研究高级人类发展研究和人类思想生态学研究的新实验室。” 他认为，任何组织的管理问题，都可以用一种新方法来加以解决，建立起某种环境条件，使个人目标与该组织的目标结合起来。即，无论何种管理都应以人为本。
首先，他尊重个人，强调自我实现，主张以最简单的方式，对人类劳动、生活和谋生的方式进行合适的管理，认为合适的管理是一种理想化的或革命性的技巧。
其次，他认为，推进健康管理和协同管理，要修改在大型组织里一直在起作用的那种顺从性的行为，让人们在一个公司里保持自己的个性，使工作不成为一种工作，而成为自己喜欢的娱乐。
第三，他盛赞协同，主张一个追求自身利益的人，同时也自动地帮助了别人；无私奉献帮助他人的人，同时也得到自我需要的好处。
最后，他极力主张不要将权力赋予给“为得到权力而追求权力的人”。
马斯洛的观点很具有颠覆性、穿透性和准确的预见性。几十年过去了，其有关要求自我实现的员工、培养客户忠诚、树立领导风范以及把不确定性作为一种创造力源泉的主张等等，描绘了我们今天数字化时代的图景，显得非常深刻。
五 评价
人本主义思想的局限性主要体现为：
1 理论体系不够严谨，缺乏对基本观点的明确目标和充分论证，一些概念也描述得很模糊。
2 过分强调自我实现和自我选择，认为这是一种与生俱来的自然倾向，忽视社会环境和后天教育对人成长的影响和制约。
3 人格问题研究方法有其积极意义，但作为一种方法论体系存在一些不可忽视的缺陷。排除整体分析和经验描述，单纯以自然科学的实验和分析方法不足以说明人的精神生活相互联系和因果关系。
尽管人本主义心理学有其不足之处，但它探讨了人的本性和价值，试图提供心理学的证明，不仅扩大了心理学的领域，丰富了人的精神生活的研究，并且加强了实证科学和规范科学的联系，也促进了心理学向高级发展。
马斯洛的人本主义心理学为我们开创了认识人生，改善人生的新天地，它研究的问题与社会生活紧密相联，提出引人深思的社会问题，虽然不够尽善尽美，但这是积极的，对社会的个体，民族乃至人类整体的生活提高都是有益的

   作者：孙定

知识管理解决方案的核心内容是制定知识管理策略。知识管理策略要解决观念问题，要突破信息时代形成的思维定式，更新知识，使观念向知识时代校正。知识管理 策略还要解决机构的文化问题，使机构具有知识时代所要求的组织学习能力并建立知识共享机制。接下来是选择适当的产品，开发知识管理项目。因此，知识管理市 场具有咨询服务需求与技术产品需求共生的特点。前者解决知识更新、观念更新、策略制定、文化改造、调整机制等问题；后者解决具体实现的问题。正是基于知识 管理的这种特点，重要的知识管理供应商都同时提供咨询服务和技术产品。

　　知识管理产品与服务的另一特点是种类繁多，每个供应商都有自已的一套说法，这些说法互不相同，甚至差别巨大。这是由两个原因造成的：首先，目前 无论在学术上还是在实际应用中，知识管理都处于非常早期的阶段，其定义有数百种，学术上也有很多不同的观点，供应商当然是各取所需；其次，供应商都是从自 已原先的领域进入知识管理领域，拥有不同的技术和产品，而知识管理本身与其说是一种新技术不如说是一种新观念，大量现有产品与技术都与知识管理相关，供应 商所做的只是根据知识管理的需求，重新定位现有的产品。

　　这里着重讨论一些重要的、具有不同特点的知识管理产品与服务。

　　Lotus:以专取胜

　　虽说Lotus与IBM本是一家，知识管理论调也一样，但各自有各自的知识管理产品，所以还是要分开说。

　　在所有知识管理解决方案厂商中，Lotus给人印象最为深刻。知识管理所必需的文档管理和群件技术在1998年前后已经是Lotus的主打产 品。而Lotus Notes本身是一个可完成多种应用的平台，虽然不是浏览器界面，但在原理上已经很接近企业门户，这些都是Lotus进入知识管理市场的先天优势。这两年 知识管理的兴起，对Lotus来说实在是一个天赐良机。Lotus在知识管理上狠下一番功夫，拼命赌一把也就在情理之中了。

　　Lotus、IBM研究中心、IBM知识管理研究所共同对Lotus专业服务以及IBM全球知识管理服务机构在全球的2万个客户的知识管理实践 进行了调查，以Lotus现有技术为基本出发点，制定出独特的理论框架，并确立了知识管理产品策略。第一个产品K-Station企业门户和其配套产品 Discovery Server已经完成。

　　Lotus认为，仅仅将知识管理局限在从海量信息中提取有用资料是不够的，还要找到具有专业知识的人，这些人还要交流、互动、进行创造性的工 作。于是，Lotus将数据、资料及处理过程定义为“事物（Thing）”、将建立在网上的虚拟工作环境定义成“场所（Place）”、将员工、客户、专 家、合作伙伴等定义成“人（People）”，而在人、场所、事务之间建立有机关联才是理想的知识管理环境。　

　　其中，K-Station已经具有知识管理系统必备的知识管理功能，Discovery服务器则是对前者的增强。

　　在K-Station中，每个人都有自已的场所——个人场所（Personal Place）。个人场所为担任不同角色的人员提供定制的日常工作环境。在个人场所中可进行电子邮件处理、管理日程、讨论、获取订阅资料、编辑文档等操作。 沟通场所（Community Place）为由相关人员组成的小组提供了共享与共同工作的环境。所有个人文档都被加上了基于场所的标签，并按场所将文档进行分类归档。这种机制为文档的 共享和检索提供了方便。在场所中可以看到何人正在线上，并列出共享场所的清单，在线上的人可以相互进行即时的消息沟通。目前，K-Station必须在 Domino环境下运行，因此系统中至少要有一个Domino服务器。

　　微软：追求通俗

　　微软一方面将现有产品基本上都贴了知识管理的标签，一方面也在开发新一代知识管理产品。微软的新一代知识管理产品正在进行第三版β测试，其产品 代号为“Tahoe（太湖）”。与Lotus不同，微软没在知识管理理论上标新立异，在这一点上，微软比Lotus“通俗”得多。

　　按照微软的说法，Tahoe是集文档管理、文档索引/检索和协同工作于一身的企业门户。Tahoe的文档管理包括版本控制、文档的作者与密码属 性管理、文档发布控制、签发控制等功能。在文件索引方面，Tahoe可以进行全文检索，也可以对网站、文件系统、Exchange服务器、Lotus服务 器等多种信息源进行检索。

除此之外，在Tahoe系统中还可以采用人工方法对文档进行分类处理，在处理过程中，Tahoe的分类助理可以学习人工分类规则，当样本达到一定数量，分类助理就可以自动进行分类。

　　Tahoe由文档服务器、索引服务器和检索服务器组成。这些服务器既可以安装在一台机器上，也可以分装在三台机器上。使用时，既可以以WWW方式进入Tahoe，也可以通过MS Office中的Tahoe插件进入，还可以直接从Windows文件系统进入。

　　微软的策略是只提供知识管理系统平台，而针对各行各业的知识管理解决方案由经微软认证的合作伙伴提供。

　　Lotus与微软在知识管理领域各有所长:Lotus像一个老谋深算的智者，微软更像一个门槛精的小伙。Lotus先建立一个严格的体系，然后 再一步步推进;微软则不太在意体系，缺什么就做什么。Lotus的难点在于它所建立的体系是否能被各界人士认同，而微软的问题是体系不严密。

　　IBM:挖掘文本

　　在文本挖掘软件中，IBM的Text Miner很有代表性，其主要功能是特征抽取、文档聚集、文档分类和检索。

　　Text Miner的特征抽取器能从文档中抽取人名、组织名和地名以及由多个字组成的复合词。此外，特征抽取器还能抽取表达数字的词汇，例如，“钱”、“百分 比”、“时间”等。抽取完特征以后，有相似特征的文档就被自动聚集成一个集合。利用这一功能，知识管理系统可以从大量文档中找到相关文档。Text Miner还可以对文档进行自动分类。

　　Autonomy:去除冗余

　　在中国，知道Autonomy公司及其技术的人不多。但实际上，Autonomy及其CEO迈可·林奇（Mike Lynch）在知识管理界的知名度很高。迈可·林奇1991年毕业于剑桥大学，主修神经网络。他受模式识别所用的概率算法的启发，创立了 Neurodynamics公司，以概率论中的贝叶斯公式和香农的信息论作为其