2007年--計算機外文翻譯--社交網(wǎng)絡(luò)數(shù)據(jù)的整合與推論（節(jié)選）

1、<p>　　2450單詞，12500英文字符,3940漢字</p><p>　　出處：Mika P, Jain R, Sheth A. Modeling and Aggregating Social Network Data[J]. Social Networks & the Semantic Web, 2007.P109-113</p><p>　　Aggregatin

2、g and reasoning with social network data</p><p>　　Supposing that we have started out with some data sets in traditional formats (rela- tional databases, Excel sheets, XML files etc.) our first step is to con

3、vert them into an RDF-based syntax, which allows to store the data in an ontology store and ma- nipulate it with ontology-based tools. In this process we need to assign identifiers to resources (an issue that we deal wit

4、h in Section 5.4.1) and re-represent our data in terms of a shared ontology such as FOAF.</p><p>　　In case our data sets come from external sources it is often more natural to pre- serve their original schem

5、a. For example, in case of converting data from a relational database or Excel sheet it is natural to preserve the schema of the database or spread- sheet as represented by the table definitions or table headings. We can

6、 then apply ontology mapping to unify our data on the schema level by mapping classes (types) and properties from different schemas to a shared ontology such as FOAF. In ef</p><p>　　The task of aggregation,

7、however, is not complete yet: we need to find identical</p><p>　　resources across the data sets. This is a two step process. First, it requires capturing the domain-specific knowledge of when to consider two

8、 instances to be the same. As we will see the FOAF ontology itself also prescribes ways to infer the equal- ity of two instances, for example based on their email address. However, beyond these properties it is likely th

9、at we need to introduce some domain-specific criteria based on domain-specific properties. In order to do this, we need to consider the ge</p><p>　　Once we determined the rules or procedures that determine e

10、quality in our do-</p><p>　　main, we need to carry out the actual instance unification or smushing (see Sec- tion 5.4.4). Unlike much of the related work on instance unification, we consider smushing as a re

11、asoning task, where we iteratively execute the rules or procedures that determine equality until no more equivalent instances can be found. The advan- tage of this approach (compared to a one-step computation of a simila

12、rity measure) is that we can take into account the learned equalities in subsequent rounds of rea- so</p><p>　　We will discuss the advantages and disadvantages of using rule-based reasoners</p><p

13、>　　and Description Logic reasoners for this task and the trade-off between forward- and backward chaining reasoning. We will also outline the approach in case we need to combine procedural and rule based reasoning.<

14、;/p><p>　　Representing identity</p><p>　　One of the main advantages of RDF over other representation formalisms such as UML is the possibility to uniquely identify resources (instances, classes, as

15、 well as properties). The primary mechanism for this is the assignment of URIs to resources. Every resource, except blank nodes is identified by a URI.</p><p>　　However, in practice it is often the case that

16、 there are a number of candidates for identifiers. For example, in the publishing world a number of identifier schemes are in use. Standard schemes such as ISBN and ISSN numbers for books and periodicals as well as DOIs

17、(Digital Object Identifiers) are widely adopted, while each publisher and online repository also maintains its own identifier scheme. Further, publications that are accessible online can be represented by their URL. All

18、of these ident</p><p>　　Multiple identifiers can be represented in RDF in two separate ways. First, one can introduce a separate resource and use the identifiers as URIs for these resources. Once separate re

19、sources are introduced for the same object, the equality of these re- sources can be expressed using the owl:sameAs property (see the following section). The other alternative is to chose one of the identifiers and use i

20、t as a URI.</p><p>　　Note that in all cases resource identifiers need to conform to the URI specifica- tion and good practice. Many modern identifier schemes such as DOIs have been de- signed to conform to t

21、he URI specification. Other identifier schemes can be recoded as URIs with the new20 info: prefix (protocol), which is regulated in [dSHNW06]. It is also a common practice to create URIs within a web domain owned or cont

22、rolled by the creator of the metadata description. For example, if the registered domain name </p><p>　　the volume and the page number. If there turns out to be a data entry mistake in one of the values, the

23、 identifier becomes obsolete and is potentially ambiguous.</p><p>　　The use of http:// URIs has been somewhat complicated by the long outstanding issue of whether such URIs could be used for only web resourc

24、es or also for ab- stract concepts (which has been the practice), and if yes, what should web servers respond to HTTP requests related to such URIs.22 The resolution of this issue is that the response code of the web ser

25、ver should indicate whether the URI denotes a web resource (HTML page, image about a person, publication etc.) or some abstract con- cept (su</p><p>　　A practical consequence is that for abstract concepts on

26、e should not choose URIs that are recognized by a server, e.g. the location of an existing HTML page, as this would be a case of URI clash. While the decision also opens the road to look up metadata about abstracts conce

27、pts using the HTTP protocol there are currently very few web servers configured to support this.</p><p>　　On the notion of equality</p><p>　　RDF and OWL (starting with OWL Lite) allow us to iden

28、tify resources and to repre- sent their (in)equality using the owl:sameAs and owl:differentFrom properties. How- ever, these are only the basic building blocks for defining equality. In particular, the meaning of equalit

29、y with respect to the objects of the domain depends on the domain itself and the goals of the modelling. For example, it is part of the domain knowl- edge what it takes for two persons or publications to be considered th

30、e same,</p><p>　　Nevertheless, it is interesting to take a short de-tour to the domain of philoso- phy, where a number of attempts have been made over the centuries to capture the general characteristics of

31、the notions of identity and equality (indiscernibility). The most well-known formulation of equality was given by Wilhelm Gottfried Leibniz in his Discourse on Metaphysics. The Identity of Indiscernibles or Leibniz-law c

32、an be loosely phrased as the principle of all things being identical unless we are able to</p><p>　　?P : P (x) ? P (y) → x = y(5.1)</p><p>　　?P : P (x) ? P (y) ← x = y(5.2)</p><p>

33、;　　The reflexive, symmetric and transitive properties of equality follow from these definitions. Notice that both formulas are second-degree due to the quantification on properties. This quantification is also interestin

34、g because it provides the Leibniz-law different interpretations in open and closed worlds. Namely, in an open world the number of properties is unknown and thus the Leibniz-law is not useful in practice: we can never con

35、clude that two resources are equal since we can never be certai</p><p>　　In practice, a closed world assumption can be equally undesirable as an open</p><p>　　one. In most cases we have almost c

36、omplete information about our resources, but still we may not want two resources to be considered identical just because of a lack of information. For example, if we have two resources and we only know that they are the

37、same gender, we may not want to assume they are identical (which would be the consequence in a closed world).</p><p>　　Philosophical ontologists have also argued against the Leibniz-law in the original</p

38、><p>　　form because it is stronger than our natural notion of equality. Consider for exam- ple a perfectly symmetrical space with two perfect spheres at some distance d to each other. Our natural intuition woul

39、d consider the two spheres indistinguishable. However, they can be distinguished as the first sphere is distance d from the second sphere, which is not true for the second sphere (it is zero distance to itself). The so-

40、lution to this problem is to limit the kind of properties to be considered, i</p><p>　　The same strategy can be followed in closed worlds to introduce weaker notions</p><p>　　of equality at will

41、. For example, one might specify the set of properties to be checked and exclude such properties as foaf :based near, which gives the geo-location of the individual or transient properties such as foaf :gender, which may

42、 change during the lifetime of an individual.</p><p>　　Lastly, let’s consider the relationship of the Leibniz-law to OWL and the seman-</p><p>　　tics of the owl:sameAs relationship. First, we sh

43、ould note that the properties we are interested in are the statements that are made about a resource. We could always distinguish two resources for example by their URIs. However, we do not want to consider the URI a pro

44、perty of the resource, since this would be too strong of a notion of equality. (Resources could never be equal, only bNodes.)</p><p>　　The semantics of OWL is built on an open world assumption, which means t

45、hat the Leibniz-law cannot be used to infer identity, not even if we reduce the property space even further. However, we can still infer the equality of instances by necessity (see the following Section).</p><

46、p>　　On the other hand, the semantics of owl:sameAs conforms to Formula 5.2. Namely, owl:sameAs restricts the interpretation of the theory to those models where the two symbols denote the same object and thus they must

47、 be indiscernible in the sense that they are interchangeable in statements:</p><p>　　(s1, owl:sameAs, s2) ∧ (s1, p, o) → (s2, p, o)</p><p>　　(p1, owl:sameAs, p2) ∧ (s, p1, o) → (s, p2, o)</p&

48、gt;<p>　　(o1, owl:sameAs, o2) ∧ (s, p, o1) → (s, p, o2)(5.3)</p><p>　　The reflexive, symmetric and transitive properties of sameAs also follow:</p><p>　　?s : (s, owl:sameAs, s)</p>

49、<p>　　(s1, owl:sameAs, s2) → (s2, owl:sameAs, s1)</p><p>　　(s1, owl:sameAs, s2) ∧ (s2, owl:sameAs, s3) → (s1, owl:sameAs, s3) (5.4)</p><p>　　Note that it is not inconsistent to have reso

50、urces that are owl:sameAs but have different stated properties, e.g. Formula 5.5 is not an inconsistent ontology. The ex- planation lies again in the open world assumption: we can assume that the missing statements (that

51、 s1 has the foaf :name Paul and s2 has the foaf :name John) exist somewhere. In a closed world this ontology would be inconsistent.</p><p>　　(s1, owl:sameAs, s2) (s1, foaf :name, ”John”)</p><p>

52、　　(s2, foaf :name, ”Paul”)(5.5)</p><p>　　Determining equality</p><p>　　In our case, we are interested in capturing knowledge about the identity of resources that can lead to conclude the (in)eq

53、uality of resources.</p><p>　　In OWL there are a limited set of constructs that can lead to (in)equality state-</p><p>　　ments. Functional and inverse functional properties (IFPs) and maximum ca

54、rdinality restrictions in general can lead to conclude that two symbols must denote the same re- source when otherwise the cardinality restriction could not be fulfilled. For example, the foaf :mbox property denoting the

55、 email address of a person is inverse-functional as a mailbox can only belong to a single person. As another example, consider a hypothetical ex:hasParent property, which has a maximum cardinality of two. If</p>&

56、lt;p>　　5.4 社交網(wǎng)絡(luò)數(shù)據(jù)的整合與推論</p><p>　　假設(shè)在傳統(tǒng)平臺（相關(guān)數(shù)據(jù)庫，電子表格，XML文件等）上，我們已經(jīng)邁出了將其轉(zhuǎn)化為RDF基點句構(gòu)的第一步。在這種句構(gòu)中，我們可以將數(shù)據(jù)存儲到本體庫中并用本庫工具對其操作。在此進程中，我們?yōu)橘Y源分發(fā)編號（詳見5.4.1）并以FOAF這類共享本體的形式重新呈現(xiàn)數(shù)據(jù)。</p><p>　　為了以防數(shù)據(jù)組來源于外部，我們通常

57、都會保留它們的源圖表。比如，為了避免轉(zhuǎn)換數(shù)據(jù)來源于相關(guān)數(shù)據(jù)庫或者電子表格，數(shù)據(jù)庫的圖表或者電子表格程序在用表定義或表格標題呈現(xiàn)時，我們通常都會將其保留一份。從不同的圖表或諸如FOAF這類共享本體的方向來構(gòu)劃等級和性質(zhì)，通過這種方式，我們可以將本體映射運用于圖表水平上的數(shù)據(jù)一體化。實際上，在本體映射中，我們可以把數(shù)據(jù)組當(dāng)做有單個共享圖表的類型進行處理。而且在語義網(wǎng)絡(luò)中，關(guān)于本體映射自動化理論的研究一直很熱門。由于涉及的本體數(shù)量及其規(guī)格對我

58、們典型案例中的自動化本體映射并非必要，因此也就不在此贅言。</p><p>　　但是整合的任務(wù)尚未完成，我們需通過數(shù)據(jù)組找出相同資源。統(tǒng)共需要兩步。第一步，確定兩個例子在何時為同樣的領(lǐng)域特異性信息。我們也會看到，為了推斷出兩個例子的對等性，F(xiàn)OAF本體也會規(guī)定一些方法（比如，以其郵件地址為基礎(chǔ)）。然而，除去這些性質(zhì)以外，我們似乎需要引進一些基于領(lǐng)域特異性質(zhì)的領(lǐng)域特異性標準。為了實現(xiàn)這一目標，我們就要考慮RDF/O

59、WL中對等的普遍意義（詳見5.4.2）。在實際操作中，只有很小一部分涉及到例子對等性的信息能夠在RDF或OWL中被捕捉到。比如，我們將臨界值應(yīng)用于相似性測驗中時，就常?？紤]到起決定作用的對等性。這種測驗是特定性質(zhì)相似性的重要結(jié)合。（臨界值由實驗或者機械學(xué)習(xí)得來）在這個以及其他涉及計算的實際案例中，我們需要一個程序性的信息呈現(xiàn)方式。</p><p>　　決定我們領(lǐng)域?qū)Φ刃缘囊?guī)則和程序一旦定下來，我們就要開始實施例子

60、一體化或者smushing（詳見5.4.4）。不同于例子一體化中的很多工作的是，我們將smushing看做是推論工作，我們在其中反復(fù)運行決定對等性的規(guī)則與程序，直到找不出更多同等例子為止。這種操作方法的優(yōu)勢（相較于相似性測量的單步計算而言）是，我們在一輪輪推論中可以將已知的對等性納入考慮范疇。</p><p>　　關(guān)于本任務(wù)使用以規(guī)則為基礎(chǔ)的推理和描述邏輯推理，正負鏈接推理之間的此消彼長，我們都會對其優(yōu)劣勢進行討

61、論。此外也為各種方法列出大綱以防需要將程序性推理和基于規(guī)則的推理結(jié)合起來。</p><p>　　5.4.1呈現(xiàn)一致性</p><p>　　RDF勝過其他類似UML的呈現(xiàn)形式的優(yōu)勢之一就在于它識別資源（實例、等級以及性質(zhì)）的獨特性。其基本方法是將各種URI錄入資源。除空白節(jié)點之外的每個資源URI都能夠識別別。</p><p>　　不過，在實際操作中往往有很多可用的識別

62、方法。比如，出版界在用的標識符就很多。像ISBN和ISSN這樣的標準圖書期刊標識符，以及各種DOL（數(shù)字化標識）也受眾很多，與此同時，每個出版商和在線知識庫也保留了其自己的標識符。不僅如此，可在線閱讀的出版物也可以用URL進行呈現(xiàn)。所有的標識符都是獨一無二的（一個標識符對應(yīng)一項資源），但絕大多數(shù)標識符都不是只此一種（即一項資源可受用于多種標識方法）URL即為這種情況（同一出版商也許受用于多種URL），甚至有些DOL也有這種情況，由于它們

63、注冊很集中，因此同一個出版商便接入了兩個或者更多標識符。</p><p>　　在RDF上呈現(xiàn)多種標識符的方法有兩種。第一，可以引入一份單獨資源并且使用URI這樣的標識符。一旦很多單獨資源是為同一對象引入，這些資源的對等性就可用owl same進行表達（詳見下一部分）。另一方法則是，選擇一個標識符并將其當(dāng)做URI來用。</p><p>　　而且在任何情況下，資源標識符都要與URL規(guī)格一致且可

64、操作。很多諸如DOL這樣的現(xiàn)代標識符都是與URL規(guī)格一致的。其他標識符可重新編碼成為有新信息的URL：前綴，也被規(guī)入[DSHNW06]。這在由元數(shù)據(jù)描述的創(chuàng)建者所有或者受其控制的網(wǎng)域中也是司空見慣。比如，如果某公共機構(gòu)的注冊域名是http://www.example.org, 那么一個有123標識的資源就是http://www.example.org/id/123. 這就滿足良好URL的準繩，尤其是滿足“良好URL應(yīng)該是唯一且穩(wěn)定的”。

65、第一條標準是良好的URL應(yīng)該是精準的或者至少應(yīng)該保證其他人不太可能用該URL操作不同的事項。這點之所以很重要，是因為擁有相同URL不同指定意義的資源很可能導(dǎo)致不一致（URL故障）。第二條標準也很重要，因為資源（在接入URL時）是無法重新命名的。一個URL一旦更改，唯一的解決辦法就是引入一個新資源并且保持其與舊資源的等價。不過，在像網(wǎng)絡(luò)這樣的大規(guī)模體系中，由于新的標識符無法進入較偏門的系統(tǒng)，故而此類系統(tǒng)中的資源很可能會繼續(xù)參用舊的標識符。

66、因此，在標識符里編碼不穩(wěn)定性質(zhì)（[PP06]就是如此）并不可取。在[</p><p>　　使用http：//URL已經(jīng)變成一個復(fù)雜的問題。因為長期以來，關(guān)于這類URL是只能夠運用于網(wǎng)絡(luò)資源還是也可以運用于抽象理念（已經(jīng)投入實踐）的問題一直很突出，如果說后者可以，網(wǎng)絡(luò)服務(wù)器該如何響應(yīng)與此類URL相關(guān)的HTTP請求。這一問題的解決方向是網(wǎng)絡(luò)服務(wù)器的響應(yīng)代碼應(yīng)該指明URL是否指示某個網(wǎng)絡(luò)資源（HTML頁面，人物圖片，出

67、版物等）或者是某些抽象理念（比如一個人，一本出版物等）。網(wǎng)絡(luò)服務(wù)器用來指示成功的響應(yīng)代碼只能夠被網(wǎng)絡(luò)資源使用。</p><p>　　對于抽象理念會有的一個實際的后果就是，我們不能選擇已被某個服務(wù)器（如已有HTML頁面的位置）識別的URL，因為此類操作容易造成URL運行失敗。雖然這一決定正好使得運用HTTP查找抽象理念相關(guān)元數(shù)據(jù)成為可能，目前受已配置網(wǎng)絡(luò)服務(wù)器支持的卻很少。</p><p>

68、　　5.4.2 對等的定義</p><p>　　在RDF和OWL（起于OWL Lite)中，我們可識別資源并用owl的類似和相異性質(zhì)呈現(xiàn)其對等性。不過，這些只是定義對等性的基本限制. 具體來說，每個對象的對等性由域名以及模型的目標決定。</p><p>　　比如，兩個人或者出版商如何才能算作是相同，就屬于域名的問題。（如：我們將何種特征納入考慮范圍）不僅如此，依據(jù)模型的等級，我們可以分辨一

69、個人或者一群人（依據(jù)其角色定位），并且把組內(nèi)個體看做對等的（角色對等）。</p><p>　　不過，稍微提一下域名的哲學(xué)性也無妨，數(shù)百年來，很多人曾嘗試在一致性和平等性之間找出其普遍特征。Wilhem Gottfeid Leibniz在其作品《論形而上學(xué)》中提出了最著名的平等性公式。除非我們能夠區(qū)分這兩者或者證明其毫無相似之處（任何兩個事物都不可能絕對相同），否則我們大體可以將模糊性或者Leibniz 定律的一致

70、性稱為萬物恒等原則。Leibniz定律如圖5.1所示。與Leibniz定律相反的是模糊的一致性如圖5.2所示。這兩種定律（均稱為Leibniz定律）是很多體系中對等性定義的基礎(chǔ)。</p><p>　　對等的反身性，對稱性和及物特點是由這些定義延展出來的。由于性質(zhì)的數(shù)量，兩個公式都屬二等。由于Leibniz定律在開放和封閉條件下的意義不同，數(shù)量也就非常有趣。也就是說，在開放條件下，性質(zhì)的數(shù)量是未知的，因此Leibn

71、iz定律并不適用于實際操作：由于我們無法確定是否存在其他能夠幫助我們區(qū)分他們的性質(zhì)的方法，因此我們也無法斷定說這兩者為對等。在封閉條件下，我們可以反復(fù)對比全部性質(zhì)來檢驗兩種資源是否對等；如果我們能夠根據(jù)某些性質(zhì)區(qū)分他們，則可以說他們是對等的。</p><p>　　但是在具體操作中，封閉條件的假設(shè)也可能會與開放條件一樣不太盡如人意。我們大致已經(jīng)獲得了資源的全部信息，但我們?nèi)圆幌Ｍ麅煞N資源在信息殘缺的情況下被認定為一

72、致。比如，如果我們有兩種資源，但只知道他們性別相同，在此情況下，我們也許就不能說他們是一致的（這就是封閉條件下的結(jié)果）。</p><p>　　本體哲學(xué)家也曾反對初始形式的Leibniz定律，因為這一定律較其他對等的天然理念而言更加強大。設(shè)想將兩個完美的球置于離彼此距離為d的位置上，以此構(gòu)成一個完美對稱的空間。人們直覺會認為這兩個球難以區(qū)別。但是，由于第一個球處在離第二個球距離為d的地方，而第二個球處于離第二個球距

73、離為0的地方，因此是可區(qū)分的。采用這種解決辦法是為了限制這種性質(zhì)的介入，尤其是為了排除諸如距離之間的不純粹的、外在的可能性。</p><p>　　在封閉條件下引入較弱的對等理念也可遵循同樣的策略。比如，有人也許會指出這組性質(zhì)是已經(jīng)檢驗過的，并且將像此類性質(zhì)當(dāng)做foaf:based near排除在外, 這類性質(zhì)會提供個人或諸如foaf：性別這樣的瞬時性質(zhì)的地理位置，而這些性質(zhì)在個體一生中也許會改變。</p&g

74、t;<p>　　最后，對Leibniz定律和OWL的關(guān)系與Owl:same As 關(guān)系的句構(gòu)進行探討。首先，我們要知道的是，我們所感興趣的性質(zhì)都是關(guān)于資源所作出的陳述。我們總是能夠分辨兩種資源（比如通過他們的URL）然而，由于以此作為對等的定義有些不妥，我們并不希望將URL當(dāng)做是資源的一種性質(zhì)。（除了Bnode, 別的資源絕不可能是對等的。）</p><p>　　OWL的句構(gòu)則是構(gòu)建于開放條件的設(shè)想

75、上的，這就意味著Leibniz定律不能用來推斷一致性，即使是我們進一步減少性質(zhì)空間也不可以。但我們依舊可以根據(jù)必要條件推斷出距離的對等性。（詳見下一部分）</p><p>　　另一方面，Owl:same As 的句構(gòu)是與表5.2相符的。也就是說。Owl: same As 限制了這一理論在那些模型中的解讀，在這些模型中，兩個標識代表同樣的事物，因此他們必須保證在替換表述時是沒有歧義的。</p><

76、;p>　　Same AS 的反身性、對稱性以及及物性如下：</p><p>　　要注意的是陳述特征不同的owl:same As資源并非不一致，公式5.5就不是不一致的本體。開放條件假設(shè)也可以解釋這種現(xiàn)象：我們可以假定丟失陳述（81 的foaf名為Paul，而82 的foaf名為John）存在。而在封閉條件下這一本體即為不一致。</p><p>　　5.4.3對等性的決定性作用<

77、/p><p>　　關(guān)于能引導(dǎo)得出對等性質(zhì)的資源，本研究旨在掌握與其一致性相關(guān)的信息。 在OWL中，能夠?qū)驅(qū)Φ刃躁愂龅慕Y(jié)構(gòu)很有限。函數(shù)與反函數(shù)性質(zhì)（IFP）和最大基數(shù)限制一般都能引出結(jié)論：兩個標志必須代表同一個資源，否則基數(shù)限制就不可用。比如，因為一個郵箱只能屬于一個人，所以代表某人郵箱地址的foaf:mbox 就是反函數(shù)。再舉個例子，假定有ex:has parent ，則其最大基數(shù)為2。如果我們說一個人有三位爸爸媽