CN104166670A - Information inquiry method based on semantic network - Google Patents

Abstract

The invention discloses an information inquiry method based on a semantic network. The information inquiry method is used for searching for ontology instances matched with semantic inquiry statements in an ontology library according to specific semantic inquiry statements and ranking the inquiry results. The method includes the following steps that first, ontology instance graphs are traversed; second, according to the weight and the number of semantic relation paths between the ontology instances and all inquiry keywords, the dependency between the current ontology instance and the current keyword is comprehensively calculated; third, according to the dependency between the ontology instances and all the inquiry keywords and the weight of all the inquiry keywords, the dependency between each ontology instance and an inquiry keyboard set is comprehensively calculated; fourth, the ontology instances are ranked according to the high-to-low sequence of the dependency. With the information inquiry method, the precision ratio and the recall ratio of semantic network information inquiry can be effectively increased.

Description

A kind of information query method based on semantic net

Technical field

The present invention relates to information network technique field, particularly relate to a kind of information query method based on semantic net.

Background technology

Traditional information query technique mainly uses the querying method based on key word, and the method mainly characterizes the key word of user's inquiry request and the information content of resource by handle and carries out strict mechanical matching and realize.Because the querying method of key word itself lacks the representation of knowledge and semantic processing ability, the coupling of only carrying out key word cannot reflect the semantic relation in user's inquiry request, thereby causes the precision ratio of Query Result on the low side.Semantic net is a kind of new network architecture being proposed by Tim Berners-Lee, and it can be the source document interpolation semantic information in network, thereby makes computing machine can understand semantic information document.Body is the gordian technique that realizes semantic net, and it is the important method of the representation of knowledge, can be with a kind of formal, accessible semanteme that represents to describe between concept of machine.Because body has good concept hierarchy and the support to reasoning from logic, thereby be widely used in information retrieval.

In recent years, there is the research of many inquiring technologies based on semantic net both at home and abroad.Liu etc. by build one based on keyword and between the concept relation graph of semantic relation, realized the identification to semantic relation in keyword query process, but this algorithm lacks the weight analysis to semantic relation and the sequence to Query Result.Castells etc. have proposed a kind of information retrieval framework based on body, this framework uses tf*idf algorithm to compose with weight each semantic relation, and Query Result is sorted according to vector space model, but this algorithm is thicker to the weight calculation granularity of semantic relation, and ignore the difference between semantic relation.Zhou etc. have proposed a kind of link sort method based on relational model, this algorithm has link structure main, external key relational model by structure, thereby realize sequence to Query Result, but this algorithm is not considered specificity between link and the problem such as covering scope and recognition capability of diversity and key word.Therefore, how to improve the validity of semantic net inquiry and accuracy and be still the Focal point and difficult point of information retrieval field research.

Body is the data model of describing semantic relation between concept and concept, it can describe by the relation between concept the semanteme of concept, body is made up of Schema and example thereof conventionally, it is indicated in an OWL-Lite subset that comprises RDF feature, object properties, data type attribute and backward attribute, conventionally has to give a definition:

Definition 1Schema S is defined as tlv triple <C, D, and P>, wherein C is class set, and D is data type collection, and P is property set.All classes, attribute and data type are all accurately represented by URI, and for any d ∈ C, r ∈ C ∪ D, has attribute p (d, r) ∈ P, and wherein d and r are called as respectively field and the scope of p.

Define 2 based on Schema S=<C, D, the instance graph of P> is defined as a digraph G=<V, E>, wherein V is example set, E is the set of relations between example in V.In sterogram, the example of a class of a resource representation.Make a set of [c] expression example c ∈ C ∪ D, for each v ∈ V, in the time of v.type=c, v ∈ [c].Make the set of [p (d, r)] expression attribute instance p (d, r) ∈ P, for each e (v _i, v _j) ∈ E, work as e=p, v _i∈ [d], v _jwhen ∈ [r], e (v _i, v _j) ∈ [p (d, r)], wherein v _iand v _jbe respectively the subject and object of e.

Defining 3 semantic path sp is Schema S=<C, D, a sequence of attributes p in P> ₁(d ₁, r ₁) p ₂(d ₂, r ₂) ... p _m(d _m, r _m), wherein p _i(d _i, r _i) ∈ P and r _iand d _i+1be identical class or there is identical parent.

Definition 4 is for semantic path sp=p ₁(d ₁, r ₁) p ₂(d ₂, r ₂) ... p _m(d _m, r _m), ip=e ₁(s ₁, o ₁) e ₂(s ₂, o ₂) ... e _m(s _m, o _m) be a semantic path examples of sp, work as e _i(s _i, o _i) ∈ [p _i(d _i, r _i)] and for all e _ithere is o _i=s _i+1time, s ₁, o _mit is respectively the source and destination of ip.

Define 5 users and inquire about Q and be defined as two tuple <T, K>, wherein T is class set, K is set of keywords.For a given Schema S=<C, D, P> and an instance graph G=<V based on S, E>, semantic search is searched Q=<T exactly, the answer set A of K>, wherein T ∈ C.For each resource a ∈ A, need in G, have at least one to be the semantic path examples of s from resource a to numerical value, wherein a ∈ [T] and numerical value s comprise key word k ∈ K.

Summary of the invention

For overcoming the problem of above existence, the present invention proposes following technical scheme:

Based on an information query method for semantic net, for searching the body class example matching with semantic query statement and Query Result is sorted at ontology library according to specific semantic query statement, it comprises the following steps:

S101, traversal instances of ontology figure, and return with semantic query statement in body types match, and the body class example being associated with the key word of the inquiry in semantic query statement.

S102, search respectively all semantic relation paths between current body class example and each key word of the inquiry for described each body class example, and the weight of computing semantic relation path respectively.

S103, according to the weight in semantic relation path between described body class example and each key word of the inquiry and the quantity in semantic relation path, the correlativity between COMPREHENSIVE CALCULATING current body class example and key word of the inquiry.

S104 is according to the correlativity between described body class example and each key word of the inquiry and the weight of each key word of the inquiry, the correlativity between COMPREHENSIVE CALCULATING each body class example and key word of the inquiry set.

S105, according to described correlativity order from big to small, described body class example is sorted.

Further, the weight of the computing semantic relation described in step 102 specifically comprises:

S2011, calculate in body frame figure the weight of attribute between each body class and between body class and data type.

S2012, weight according to the weight calculation body class of the each attribute in body frame figure to the semantic relation path between data type.

The weight in the semantic relation path in S2013, use body frame figure substitutes the weight of corresponding semantic relation path examples in instances of ontology figure.

Further, described step S2011 circular is:

w(p(d,r))＝α·I(p(d,r))+β·MI(p(d,r))

In formula, p (d, r) represents that body class d is to the attribute of body class or ontology data type r from body frame figure, I (p (d, r)) represent the quantity of information that attribute p (d, r) produces while generation, MI (p (d, r)) represent attribute p (d, r) the mutual information metric between d and r, α, β is respectively weight parameter, and 0≤α, β≤1.

Wherein, described I (p (d, r)) circular is:

$\left\{\begin{array}{c}I\left(p(d,r)\right)=-{\log }_2\mathrm{pr}\left(p(d,r)\right)\\ \mathrm{pr}\left(p(d,r)\right)=\frac{\left|\mathrm{sub}\left(p(d,r)\right)\right|}{\left|N\right|}\end{array}\right.$

(p (the d of pr in formula, r)) be attribute p (d, r) probability of occurrence, sub (p (d, r) be) that in instances of ontology figure, all examples from body class d are to the quantity of the attribute instance of the example of body class or data type r, N is the quantity of all class examples in instances of ontology figure.

Described MI (p (d, r)) circular is:

$\mathrm{MI}\left(p(d,r)\right)=\underset{o\&Element;r}{\mathrm{\&Sigma;}}\underset{s\&Element;d}{\mathrm{\&Sigma;}}\mathrm{pr}(s,o)\&CenterDot;{\log }_2\left(\frac{\mathrm{pr}(s,o)}{\mathrm{pr}\left(s\right)\mathrm{pr}\left(o\right)}\right)$

Pr (s in formula, o) be the probability of occurrence from the example s of body class d to the attribute instance of the example o of body class or data type r in instances of ontology figure, pr (s) is the probability of occurrence from the example s of body class d to the attribute instance of all examples of body class or data type r in instances of ontology figure, and pr (o) is the probability of occurrence from all examples of body class d to the attribute instance of the example o of body class or data type r in instances of ontology figure.

Further, described step S2012 circular is:

$w\left(\mathrm{sp}\right)=\left(\underset{p(d,r)\&Element;\mathrm{sp}}{\mathrm{\&Pi;}}w\left(p(d,r)\right)\right)\&CenterDot;{\mathrm{\&delta;}}^{\mathrm{length}\left(\mathrm{sp}\right)-1}$

In formula, sp represents that body class is to the semantic relation path of data type from body frame figure, w (p (d, r)) represent the weight of each attribute that this relation path comprises, δ is interval for (0,1) damped expoential, the number of attributes that length (sp) comprises for semantic path sp, p (d, r) represents that body class d is to the attribute of body class or ontology data type r from body frame figure.

Further, described step S2013 circular is:

w(ip)＝w(sp)

In formula ip represent with body frame figure in the corresponding instances of ontology figure of semantic relation path sp in semantic relation path examples.

Further, described step S103 circular is:

$R(a,k_i)=\underset{\mathrm{ip}\&Element;\mathrm{IP}(a,k_i)}{\mathrm{\&Sigma;}}(w\left(\mathrm{ip}\right)\&CenterDot;\mathrm{spec}\left(\mathrm{ip}\right))$

In formula, a represents body class example, k _irepresent searching keyword, ip represents semantic relation path examples, IP (a, k _i) represent to be key word k from class example a to data value _isemantic relation path examples set w (ip) represent the weight of semantic relation path ip, spec (ip) represents the specificity of path examples ip, computing formula is as follows:

$\mathrm{spec}\left(\mathrm{ip}\right)=\underset{i=1}{\overset{m}{\mathrm{\&Pi;}}}\frac{1}{\mathrm{degree}(s_i,e_i)}$

In formula, s _irepresent the body class example that path examples ip comprises, e _irepresent with body class example s _ifor the attribute instance of main body, degree (s _i, e _i) represent with body class example s _ifor all properties example e of main body _iquantity, m represents the quantity of the body class example that path examples ip comprises.

Further, described step S104 circular is:

$R(a,K)=1-{\left[\frac{{\mathrm{\&Sigma;}}_{1\&le;i\&le;\left|K\right|}{(D\left(k_i\right)\&CenterDot;(1-\mathrm{NR}(a,k_i)))}^p}{{\mathrm{\&Sigma;}}_{1\&le;i\&le;\left|K\right|}D{\left(k_i\right)}^p}\right]}^{\frac{1}{p}}$

In formula, a represents body class example, and K represents key word of the inquiry set, k _irepresent i key word in K, | K| represents to gather the number of elements in K, D (k _i) expression key word k _iweight, D (k _i) computing formula is as follows:

$D\left(k_i\right)=\log \frac{\left|\mathrm{DV}\right|}{\left|{\mathrm{DV}}_{k_i}\right|}$

In formula, | DV| represents the quantity of all data values in semantic instance graph, | DV _ki| represent all key word k that comprise in semantic instance graph _ithe quantity of data value;

NR (a, k _i) represent body class example a and key word of the inquiry k after equalization _ibetween correlativity, NR (a, k _i) computing formula is as follows:

$\mathrm{NR}(a,k_i)=\frac{R(a,k_i)}{\max \{R(a,k_1),R(a,k_2),\&CenterDot;\&CenterDot;\&CenterDot;,R(a,k_{\left|K\right|})\}}$

In formula, R (a, k _i) represent that body class example a is to key word of the inquiry k _icorrelativity, max{} represents to get maximal value; P is for regulating parameter, and p>0.

Brief description of the drawings

Fig. 1 is the body frame figure in the scientific research information field of Staffs in University

Fig. 2 be with Fig. 1 in the corresponding this subject instance graph of body frame figure

Fig. 3 is the process flow diagram of the information query method based on semantic net

Fig. 4 is the comparison diagram of the present invention and existing method test result.

Embodiment

Further illustrate technical scheme of the present invention below in conjunction with accompanying drawing and by embodiment.

Fig. 3 is the process flow diagram of the information query method based on semantic net described in the present embodiment, for searching the body class example matching with semantic query statement and Query Result is sorted at ontology library according to specific semantic query statement.As shown in Figure 3, the present embodiment said method comprising the steps of:

S101, traversal instances of ontology figure, and return with semantic query statement in body types match, and the body class example being associated with the key word of the inquiry in semantic query statement.

S102, search respectively all semantic relation paths between current body class example and each key word of the inquiry for described each body class example, and the weight of computing semantic relation path respectively.

Semantic relation path is made up of multiple attributes, and the weight in semantic relation path need to be added and be obtained by the weight of attribute that semantic relation path is comprised, comprises the following steps:

S2011 calculates in body frame figure the weight of attribute between each body class and between body class and data type, and computing formula is as follows:

w(p(d,r))＝α·I(p(d,r))+β·MI(p(d,r))

In formula, p (d, r) represents that body class d is to the attribute of body class or ontology data type r from body frame figure, I (p (d, r)) represent the quantity of information that attribute p (d, r) produces while generation, computing formula is as follows:

$\left\{\begin{array}{c}I\left(p(d,r)\right)=-{\log }_2\mathrm{pr}\left(p(d,r)\right)\\ \mathrm{pr}\left(p(d,r)\right)=\frac{\left|\mathrm{sub}\left(p(d,r)\right)\right|}{\left|N\right|}\end{array}\right.$

In formula, pr (p (d, r)) be attribute p (d, r) probability of occurrence, sub (p (d, r) be) that in instances of ontology figure, all examples from body class d are to the quantity of the attribute instance of the example of body class or data type r, N is the quantity of all class examples in instances of ontology figure.MI (p (d, r)) represents the mutual information metric of attribute p (d, r) between d and r, and computing formula is as follows:

$\mathrm{MI}\left(p(d,r)\right)=\underset{o\&Element;r}{\mathrm{\&Sigma;}}\underset{s\&Element;d}{\mathrm{\&Sigma;}}\mathrm{pr}(s,o)\&CenterDot;{\log }_2\left(\frac{\mathrm{pr}(s,o)}{\mathrm{pr}\left(s\right)\mathrm{pr}\left(o\right)}\right)$

Pr (s in formula, o) be the probability of occurrence from the example s of body class d to the attribute instance of the example o of body class or data type r in instances of ontology figure, pr (s) is the probability of occurrence from the example s of body class d to the attribute instance of all examples of body class or data type r in instances of ontology figure, and pr (o) is the probability of occurrence from all examples of body class d to the attribute instance of the example o of body class or data type r in instances of ontology figure.α, β is respectively weight parameter, and 0≤α, β≤1.

S2012, weight according to the weight calculation body class of the each attribute in body frame figure to the semantic relation path between data type, computing formula is as follows:

$w\left(\mathrm{sp}\right)=\left(\underset{p(d,r)\&Element;\mathrm{sp}}{\mathrm{\&Pi;}}w\left(p(d,r)\right)\right)\&CenterDot;{\mathrm{\&delta;}}^{\mathrm{length}\left(\mathrm{sp}\right)-1}$

In formula, sp represents that body class is to the semantic relation path of data type from body frame figure, w (p (d, r)) represent the weight of each attribute that this relation path comprises, δ is interval for (0,1) damped expoential, the number of attributes that length (sp) comprises for semantic path sp, p (d, r) represents that body class d is to the attribute of body class or ontology data type r from body frame figure.

The weight in the semantic relation path in S2013, use body frame figure substitutes the weight of corresponding semantic relation path examples in instances of ontology figure, and computing formula is as follows:

w(ip)＝w(sp)

In formula ip represent with body frame figure in the corresponding instances of ontology figure of semantic relation path sp in semantic relation path examples.

S103, according to the weight in semantic relation path between described body class example and each key word of the inquiry and the quantity in semantic relation path, the correlativity between COMPREHENSIVE CALCULATING current body class example and key word of the inquiry, computing formula is as follows:

$R(a,k_i)=\underset{\mathrm{ip}\&Element;\mathrm{IP}(a,k_i)}{\mathrm{\&Sigma;}}(w\left(\mathrm{ip}\right)\&CenterDot;\mathrm{spec}\left(\mathrm{ip}\right))$

In formula, a represents body class example, k _irepresent searching keyword, ip represents semantic relation path examples, IP (a, k _i) represent to be key word k from class example a to data value _isemantic relation path examples set w (ip) represent the weight of semantic relation path ip, spec (ip) represents the specificity of path examples ip, computing formula is as follows:

$\mathrm{spec}\left(\mathrm{ip}\right)=\underset{i=1}{\overset{m}{\mathrm{\&Pi;}}}\frac{1}{\mathrm{degree}(s_i,e_i)}$

In formula, s _irepresent i the body class example that path examples ip comprises, e _irepresent with body class example s _ifor the attribute instance of main body, degree (s _i, e _i) represent with body class example s _ifor all properties example e of main body _iquantity, m represents the quantity of the body class example that path examples ip comprises.

S104 is according to the correlativity between described body class example and each key word of the inquiry and the weight of each key word of the inquiry, the correlativity between COMPREHENSIVE CALCULATING each body class example and key word of the inquiry set, and computing formula is as follows:

$R(a,K)=1-{\left[\frac{{\mathrm{\&Sigma;}}_{1\&le;i\&le;\left|K\right|}{(D\left(k_i\right)\&CenterDot;(1-\mathrm{NR}(a,k_i)))}^p}{{\mathrm{\&Sigma;}}_{1\&le;i\&le;\left|K\right|}D{\left(k_i\right)}^p}\right]}^{\frac{1}{p}}$

In formula, a represents body class example, and K represents key word of the inquiry set, k _irepresent i key word in K, | K| represents to gather the number of elements in K, D (k _i) expression key word k _iweight, D (k _i) computing formula is as follows:

$D\left(k_i\right)=\log \frac{\left|\mathrm{DV}\right|}{\left|{\mathrm{DV}}_{k_i}\right|}$

In formula, | DV| represents the quantity of all data values in semantic instance graph, | DV _ki| represent all key word k that comprise in semantic instance graph _ithe quantity of data value.NR (a, k _i) represent body class example a and key word of the inquiry k after equalization _ibetween correlativity, NR (a, k _i) computing formula is as follows:

$\mathrm{NR}(a,k_i)=\frac{R(a,k_i)}{\max \{R(a,k_1),R(a,k_2),\&CenterDot;\&CenterDot;\&CenterDot;,R(a,k_{\left|K\right|})\}}$

In formula, R (a, k _i) represent that body class example a is to key word of the inquiry k _icorrelativity, max{} represents to get maximal value.P is for regulating parameter, and p>0.

S105, according to described correlativity order from big to small, described body class example is sorted.

Performance index

(1) method of testing

The scientific research information that in test, we choose the above academic title personnel of the associate professor of Qingdao Agricultural University is as data resource, and by semantic relation, these data resources being configured to body form, the ontology library after structure has 7984 class examples (comprising personnel, paper, works, problem etc.) and 31413 attribute instance (comprise deliver, write, preside over, participation etc.).For effective assessment algorithm herein, our 5 different query statements that have been inquiry experimental design, table 1 has shown expression formula, inquiry object and the related resource quantity of these query statements.

Table 1 semantic query statement

For the accurate evaluation inquiry accuracy of algorithm herein, in experiment, the algorithm of the proposition such as algorithm and Zhou of the proposition such as algorithm, Castells of the proposition such as the present invention and Liu is entered contrast by we.

(2) measurement index

For the Query Result of every query statement, we only evaluate front 10 results, and use the basic evaluation index in information inquiry field to evaluate:

Inquiry accuracy rate: the resource collection that wherein A is Query Result, RA is resource collection relevant in ontology library.

Inquiry recall rate: wherein, when | when RA|≤the 10, N=|RA|; When | when RA|>10, N=10.

F measure value: $F=\frac{2\mathrm{PR}}{P+R}.$

(3) Query Result

Hence one can see that, all obtain best inquiry effect for different query statement the present invention, this result is mainly owing to semantic relation weighting algorithm and sort algorithm in this paper querying method, wherein the application of semantic relation weighting algorithm contributes to inquiry to recessive information and the quantification of semantic relation weight, the sequence that sort algorithm can effectively improve related resource in Query Result simultaneously.By contrast, the algorithm of the propositions such as Liu is due to the sequence lacking Query Result, thereby it is the poorest to cause inquiring about effect; Although the algorithm of the propositions such as Castells has increased the sort algorithm of Query Result, due to semantic relation weight is not distinguished, thereby cause being inaccurate for the sequence of Query Result; The algorithm of the propositions such as Zhou, owing to failing that semantic path relation and keyword message are carried out to refinement, therefore cannot be inquired about for the recessive information in body, and the sequence of Query Result also exists certain error.

Claims (9)

1. the information query method based on semantic net, for searching the body class example matching with semantic query statement and Query Result is sorted at ontology library according to specific semantic query statement, its feature comprises the following steps:

S101, traversal instances of ontology figure, and return with semantic query statement in body types match, and the body class example being associated with the key word of the inquiry in semantic query statement.

S102, search respectively all semantic relation paths between current body class example and each key word of the inquiry for described each body class example, and the weight of computing semantic relation path respectively.

S103, according to the weight in semantic relation path between described body class example and each key word of the inquiry and the quantity in semantic relation path, the correlativity between COMPREHENSIVE CALCULATING current body class example and key word of the inquiry.

S104 is according to the correlativity between described body class example and each key word of the inquiry and the weight of each key word of the inquiry, the correlativity between COMPREHENSIVE CALCULATING each body class example and key word of the inquiry set.

S105, according to described correlativity order from big to small, described body class example is sorted.

2. method according to claim 1, is characterized in that, the weight of step 102 computing semantic relation specifically comprises:

S2011, calculate in body frame figure the weight of attribute between each body class and between body class and data type.

S2012, weight according to the weight calculation body class of the each attribute in body frame figure to the semantic relation path between data type.

S2013 uses the weight in the semantic relation path in body frame figure to substitute the weight of corresponding semantic relation path examples in instances of ontology figure.

3. method according to claim 2, is characterized in that, described step S2011 circular is:

w(p(d,r))＝α·I(p(d,r))+β·MI(p(d,r))

In formula, p (d, r) represents that body class d is to the attribute of body class or ontology data type r from body frame figure, I (p (d, r)) represent the quantity of information that attribute p (d, r) produces while generation, MI (p (d, r)) represent attribute p (d, r) the mutual information metric between d and r, α, β is respectively weight parameter, and 0≤α, β≤1.

4. method according to claim 3, is characterized in that, described I (p (d, r)) circular is:

$\left\{\begin{array}{c}I\left(p(d,r)\right)=-{\log }_2\mathrm{pr}\left(p(d,r)\right)\\ \mathrm{pr}\left(p(d,r)\right)=\frac{\left|\mathrm{sub}\left(p(d,r)\right)\right|}{\left|N\right|}\end{array}\right.$

(p (the d of pr in formula, r)) be attribute p (d, r) probability of occurrence, sub (p (d, r) be) that in instances of ontology figure, all examples from body class d are to the quantity of the attribute instance of the example of body class or data type r, N is the quantity of all class examples in instances of ontology figure.

5. method according to claim 3, is characterized in that, described MI (p (d, r)) circular is:

$\mathrm{MI}\left(p(d,r)\right)=\underset{o\&Element;r}{\mathrm{\&Sigma;}}\underset{s\&Element;d}{\mathrm{\&Sigma;}}\mathrm{pr}(s,o)\&CenterDot;{\log }_2\left(\frac{\mathrm{pr}(s,o)}{\mathrm{pr}\left(s\right)\mathrm{pr}\left(o\right)}\right)$

Pr (s in formula, o) be the probability of occurrence from the example s of body class d to the attribute instance of the example o of body class or data type r in instances of ontology figure, pr (s) is the probability of occurrence from the example s of body class d to the attribute instance of all examples of body class or data type r in instances of ontology figure, and pr (o) is the probability of occurrence from all examples of body class d to the attribute instance of the example o of body class or data type r in instances of ontology figure.

6. method according to claim 2, is characterized in that, described step S2012 circular is:

$w\left(\mathrm{sp}\right)=\left(\underset{p(d,r)\&Element;\mathrm{sp}}{\mathrm{\&Pi;}}w\left(p(d,r)\right)\right)\&CenterDot;{\mathrm{\&delta;}}^{\mathrm{length}\left(\mathrm{sp}\right)-1}$

In formula, sp represents that body class is to the semantic relation path of data type from body frame figure, w (p (d, r)) represent the weight of each attribute that this relation path comprises, δ is interval for (0,1) damped expoential, the number of attributes that length (sp) comprises for semantic path sp, p (d, r) represents that body class d is to the attribute of body class or ontology data type r from body frame figure.

7. method according to claim 2, is characterized in that, described step S2013 circular is:

w(ip)＝w(sp)

In formula ip represent with body frame figure in the corresponding instances of ontology figure of semantic relation path sp in semantic relation path examples.

8. method according to claim 1, is characterized in that, described step S103 circular is:

$R(a,k_i)=\underset{\mathrm{ip}\&Element;\mathrm{IP}(a,k_i)}{\mathrm{\&Sigma;}}(w\left(\mathrm{ip}\right)\&CenterDot;\mathrm{spec}\left(\mathrm{ip}\right))$

In formula, a represents body class example, k _irepresent searching keyword, ip represents semantic relation path examples, IP (a, k _i) represent to be key word k from class example a to data value _isemantic relation path examples set w (ip) represent the weight of semantic relation path ip, spec (ip) represents the specificity of path examples ip, computing formula is as follows:

$\mathrm{spec}\left(\mathrm{ip}\right)=\underset{i=1}{\overset{m}{\mathrm{\&Pi;}}}\frac{1}{\mathrm{degree}(s_i,e_i)}$

In formula, s _irepresent i the body class example that path examples ip comprises, e _irepresent with body class example s _ifor the attribute instance of main body, degree (s _i, e _i) represent with body class example s _ifor all properties example e of main body _iquantity, m represents the quantity of the body class example that path examples ip comprises.

9. method according to claim 3, is characterized in that, described step S104 circular is:

$R(a,K)=1-{\left[\frac{{\mathrm{\&Sigma;}}_{1\&le;i\&le;\left|K\right|}{(D\left(k_i\right)\&CenterDot;(1-\mathrm{NR}(a,k_i)))}^p}{{\mathrm{\&Sigma;}}_{1\&le;i\&le;\left|K\right|}D{\left(k_i\right)}^p}\right]}^{\frac{1}{p}}$

In formula, a represents body class example, and K represents key word of the inquiry set, k _irepresent i key word in K, | K| represents to gather the number of elements in K, D (k _i) expression key word k _iweight, D (k _i) computing formula is as follows:

$D\left(k_i\right)=\log \frac{\left|\mathrm{DV}\right|}{\left|{\mathrm{DV}}_{k_i}\right|}$

In formula, | DV| represents the quantity of all data values in semantic instance graph, | DV _ki| represent all key word k that comprise in semantic instance graph _ithe quantity of data value.NR (a, k _i) represent body class example a and key word of the inquiry k after equalization _ibetween correlativity, NR (a, k _i) computing formula is as follows:

$\mathrm{NR}(a,k_i)=\frac{R(a,k_i)}{\max \{R(a,k_1),R(a,k_2),\&CenterDot;\&CenterDot;\&CenterDot;,R(a,k_{\left|K\right|})\}}$

In formula, R (a, k _i) represent that body class example a is to key word of the inquiry k _icorrelativity, max{} represents to get maximal value.P is for regulating parameter, and p>0.