CN105912721B - RDF data distributed semantic parallel inference method - Google Patents
RDF data distributed semantic parallel inference method Download PDFInfo
- Publication number
- CN105912721B CN105912721B CN201610293055.5A CN201610293055A CN105912721B CN 105912721 B CN105912721 B CN 105912721B CN 201610293055 A CN201610293055 A CN 201610293055A CN 105912721 B CN105912721 B CN 105912721B
- Authority
- CN
- China
- Prior art keywords
- rule
- triple
- trm
- reasoning
- rdfs
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2455—Query execution
- G06F16/24564—Applying rules; Deductive queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of RDF data distributed semantic parallel inference methods, it is first depending on ontology file and RDFS/OWL rule, transitive closure relational matrix (Transitive closure relation matrix, abbreviation TRM) and link variable information are constructed, and then create-rule marks；Then classified according to the type of link variable to RDFS/OWL rule, separately design different inference schemes, the reasoning of RDFS/OWL rule is concurrently completed in conjunction with MapReduce Computational frame.Example triple is filtered by link variable information and regular marks, the transmission loss of a large amount of useless triple data in a distributed system can be reduced.The number of iterations of reasoning can be reduced by constructing transitive closure matrix, improves the efficiency of reasoning.Finally, by inference as a result, deleting duplicate triple data, in real time to further increase the efficiency of successive iterations reasoning.The reasoning of RDFS/OWL rule can efficiently and be correctly realized in the case where data volume increases through the invention.
Description
Technical field
The present invention relates to semantic network technology fields, more particularly to a kind of RDF data distributed semantic parallel inference method.
Background technique
RDF and OWL standard in semantic web has a wide range of applications in every field, such as general knowledge
(DBpedia), medical life science (LODD) and bioinformatics (UniProt), by September, 2014, the total amount of data is
Up to 65,000,000,000 triples.With the rapidly growth of data in semantic net, centralized environment is not suitable for due to the limitation of memory
Reasoning to large-scale data；Research RDFS/OWL distributed parallel reasoning is a newer at present field.As obscured pD*
Reasoning, RDFS parallel inference, distributed rule matching system, distributing inference engine WebPIE, distributing inference engine YARM.
The method of these reasonings is not efficient enough.The scheme of these reasonings largely combines MapReduce Computational frame, passes through RDFS/
OWL inference rule makes inferences data.
With the rapidly growth of data in semantic net, how efficiently to carry out the reasoning of big data is nearly heat of research in 2 years
Point, but also in the initial stage of research.The distributing inference engine of comparative maturity is WebPIE at present, although meeting big data
Parallel inference, but since WebPIE for each rule enables one or more MapReduce task and makes inferences,
And the starting of MapReduce Job is relatively time-consuming, therefore with the increase of RDFS/OWL inference rule, the efficiency of reasoning receives
Limitation.Distributing inference engine YARM uses reasonable data partitioning model, optimizes the execution order of rule, realizes one
The new parallelization reasoning algorithm based on MapReduce of kind, so that RDFS can be completed in a MapReduce task in reasoning
The reasoning of rule.But the algorithm is not suitable for the reasoning of complicated OWL Horst rule.In addition, working as the new of a certain rule generation
When triple repeats, YARM can have excessive redundant computation and generate hash.In order to solve above-mentioned RDFS/OWL distribution
Formula reasoning algorithm there are the problem of, this paper presents SPRM algorithm (Semantic information Parallel
Reasoning on MapReduce), which can efficiently and correctly realize RDFS/ in the case where data volume increases
The reasoning of OWL rule.
The demand of mass data is unable to satisfy under centralized environment, and the reasoning under distributed environment is not efficient enough, reasoning
Parallelization.Presently, there are although parallel inferences that distributing inference engine can be realized data, but MapReduce task opens
Dynamic number is more and time-consuming, there are excessive redundant computation and generates hash, causes it can not be the case where data volume increases
The reasoning of RDFS/OWL rule can efficiently and be correctly realized down.
The technical issues that need to address: how to guarantee that RDFS/OWL distributing inference algorithm is real in 1. solution distributed environments
The correctness of inference data after being now distributed.2. combine the distributed schemes proposed to propose corresponding parallel inference scheme, thus
Meet the demand of the distributed parallel reasoning of large-scale data.
Summary of the invention
In view of this, the object of the present invention is to provide a kind of RDF data distributed semantic parallel inference methods, in data volume
The reasoning of RDFS/OWL rule can efficiently and be correctly realized in the case where increase.
The present invention is realized using following scheme: a kind of RDF data distributed semantic parallel inference method specifically includes following
Step:
Step S1: loading mode triple constructs TRM, while according to RDFS/OWL rule, constructing may in every rule
The link variable information of connection；
Step S2: according to TRM and link variable information, create-rule markup model；
Step S3: being divided into two kinds of forms of single argument and multivariable for link variable, according to the type of TRM and link variable,
RDFS/OWL rule is divided into 5 seed types, separately designs different inference schemes；
Step S4: to Flag_Rulem=1 rule executes the parallel inference of RDFS/OWL rule, and exports intermediate knot
Fruit；
Step S5: the repetition triple in intermediate result is deleted；
Step S6: if including new SchemaTriple in intermediate result, TRM and regular marks model is updated, is returned
Return step S4；Otherwise, terminate.
Further, in the step S1, the relevant parameter of triple is defined as follows:
Define 1: the Subject-Verb and object that mode triple SchemaTriple refers to triple are all in ontology file
It is defined in OntologyFile, i.e., are as follows:
Wherein, the sum of n intermediate scheme triple；
If v ∈ { Si,Pj,Ok,Then,
(Si,Pj,Ok)∈SchemaTriple (1)
Define 2: example triple InstanceTriple refer to Subject-Verb and object at least one in ontology file
It is undefined in OntologyFile, it is specific example, i.e., are as follows:
Wherein, n indicates the sum of example triple；
If v ∈ { Si,Pj,Ok,Then,
(Si,Pj,Ok)∈InstanceTriple
3: triple type mark Flag_TripleType is defined, for identity mode triple and example triple, institute
Triple type mark Flag_TripleType is stated to be defined as follows:
Wherein, n indicate triple total then,
Triple item type mark TripleItem_Flag, it is described for identifying subject or object in triple item
Triple item type mark is defined as follows:
Wherein, n indicates the sum of triple.If v ∈ { Si,Pj,Ok, then
4: mode ternary Groups List SchemaList is defined, for obtaining the mode triple collection of identical predicate or object
It closes, the mode ternary Groups List SchemaList is defined as follows:
Wherein, the sum of n intermediate scheme triple, then,
SchemaList=Om_list∪Pt_list
Wherein, Om_ list expression meets predicate Pj∈'s { rdf:type } and with identical object triplet sets, with
Object name；Pt_ list expression meets predicateAll triplet sets with identical predicate,
With predicate name, it is defined as follows:
Om_ list={ (Si,Pj,Ok)|Pj∈ { rdf:type } &k=m }
Further, it in the step S1, is defined when constructing TRM as follows:
Definition 5: the transitive closure of digraph is defined as: the transitive closure of the vertex a n digraph can be defined as a n
Rank Boolean matrix T={ tik(1≤i, k≤n), if there are an active path, t by vertex i to vertex kik=1, otherwise tik=
0。
Define 6: transitive closure relational matrix Transitive closure relation matrix, referred to as TRM are tied
It closes and defines 4, defines 5 and WarShall algorithm building TRM；TRM is indicated in the transitive closure of digraph, is top with class or attribute
Point, using the predicate of triple as the Boolean matrix T={ t of relationshipik}；
Predicate is PjThe relational matrix of SchemaTriple be expressed as Pj_ TRM, wherein Pj∈{subClassOf,
SubPropertyOf, sameAs, equivalentClass, equivalentProperty },
IfThen Pj_TRM.tik=1, indicate predicate
For PjSchemaTriple subject SiTo object OkThere are direct relation, otherwise Pj_TRM.tik=0；
Wherein, TRM points are class relational matrix CTRM and relation on attributes matrix PTRM；Wherein, CTRM is according to predicate
SubClassOf, equivalentClass, sameAs define SC_TRM, EC_TRM, SAC_TRM respectively；PTRM is according to predicate
SubPropertyOf, equivalentProperty, sameAs define SP_TRM, EP_TRM, SAP_TRM respectively；Due to predicate
For the SchemaTriple relationship containing class and relation on attributes of sameAs, relationship is separated in the building of TRM, constructs SAC_ respectively
TRM and SAP_TRM.
The building TRM specifically includes the following steps:
Step S11: in conjunction with defining 4, obtain in all SchemaTriple respectively predicate be subPropertyOf and
The mode triple data of equivalentProperty, according to the SchemaTriple of acquisition and combine define 6 obtain with
SubPropertyOf and using equivalentProperty as the Boolean matrix of relationship；
Step S12: according to WarShall algorithm obtain it is all with subPropertyOf and with
EquivalentProperty is transitive relation value in the Boolean matrix of relationship, to generate transitive closure relational matrix SP_
TRM, EP_TRM complete the building of SP_TRM and EP_TRM.
Further, it in the step S1, is defined when constructing link variable information as follows:
Define 7: the link variable LinkVar mode triple to be used to connect two former pieces in RDFS/OWL rule
, it is described according to rule, the quantity of link variable is greater than 1；By the link variable information of each rule with<key, value>
Form is stored in table RulemIn _ Table, wherein key stores all mode triple items for former piece connection of the rule,
Value stores the mode triple item of the rule conclusion part.
Further, in the step S3 specifically: right using SPRM algorithm according to the type of TRM and link variable
RDFS/OWL rule is classified, quote RDFS rule when using RDFS- rule numbers form, reference OWL Horst rule
The form of Shi Caiyong OWL- rule numbers；A rule name label is distributed to every rule, rule name label is the rule
Then corresponding title；
5 seed types of RDFS/OWL rule it is as follows:
Class1: the rule that SchemaTriple is combined with SchemaTriple is directly made inferences using TRM；
The rule that type 2:SchemaTriple is combined with InstanceTriple, makes inferences according to TRM, without knot
Close link variable information；
The rule that type 3:SchemaTriple is combined with InstanceTriple, link variable information are single argument,
It needs to make inferences in conjunction with TRM；
The rule that type 4:SchemaTriple is combined with InstanceTriple, link variable information are single argument,
Without being made inferences in conjunction with TRM；
The rule that type 5:SchemaTriple is combined with InstanceTriple, link variable information are multivariable；
To any one rule is defined as: assuming that m rule is expressed as follows:
Rulem:Cm1,Cm2,…,Cmk,…,Cmn→Result
Definition rule flag F lag_Rulem, for marking whether the rule is the rule that impossible activate；
When for Flag_Rulem=0, then the condition for meeting the rule is implied the absence of, without activating the rule；Assuming that Pj
For RulemIn some former piece predicate, wherein Pj∈{subClassOf,subPropertyOf,sameAs,
EquivalentClass, equivalentProperty }, PjCorresponding TRM is set as Pj_TRM；
In conjunction with definition 6 and 7 are defined, regular marks Flag_RulemClass definition according to RDFS/OWL rule is as follows:
If 1) Rulem∈ { Class1, type 2 }, then
If 2) Rulem∈ { type 3, type 4, type 5 }, then
Further, in the step S4, pushing away for RDFS/OWL rule is completed parallel in conjunction with MapReduce Computational frame
Reason just completes reasoning regular in Class1 in the building TRM stage, is directed to type in the parallel inference stage of RDFS/OWL rule
2, type 3, type 4 and type 5 carry out distributed parallel inference, to realize all rule in a MapReduce task
A reasoning then；
For the rule of type 2, type 3 and type 4, according to regular marks model and link variable information, by Map
The parallel inference in stage can obtain the reasoning results triple data；For the rule of type 5, by meeting Inference Conditions
Data carry out redundancy output to ensure the correctness of subsequent parallel data reasoning, i.e., will meet the example triple of same rule
Output is set as identical key key, and to ensure in the Reduce stage, the strictly all rules former piece in same rule can be according to connection
Variable is completed to connect and exports result；
The parallel reasoning algorithm of the RDFS/OWL rule in Map stage specifically:
Input: key is line number；Value is example triple
Output: key is the combination of the reasoning results triple or regular marks and link variable；Value is arbitrary value or band
The triple item of label；
The RDFS/OWL rule parallel reasoning algorithm in Reduce stage specifically:
Input: the output result of duplicate triple data phase is deleted
Output: key is output triple；Value is arbitrary value.
Further, in the step S5, example triple is filtered by link variable information and regular marks,
The transmission loss of a large amount of useless triple data in a distributed system is reduced, reasoning is reduced by construction transitive closure matrix
The number of iterations, to improve the efficiency of reasoning；By inference as a result, deleting duplicate triple data in real time, into
The efficiency of one step raising successive iterations reasoning；
Duplicate triple data are deleted after executing combiner and each reasoning after each Map,
Wherein, combiner is the repeated data exported in the deletion map stage, specifically:
Input: the output result of the parallel reasoning algorithm of the RDFS/OWL rule in Map stage
Output: key is the combination of the reasoning results triple or regular marks and link variable；Value is arbitrary value or band
The triple item of label；
The algorithm for repeating triple data is deleted after each reasoning specifically:
Input: key is line number；Value is triple
Output: the triple after output duplicate removal.
Distributed parallel reasoning proposed by the present invention is first depending on ontology file and RDFS/OWL rule, constructs transitive closure
Relational matrix and link variable information, and then create-rule marks；Then according to the type of link variable to RDFS/OWL rule
Classify, separately design different inference schemes, concurrently completes RDFS/OWL rule in conjunction with MapReduce Computational frame
Reasoning.Example triple is filtered by link variable information and regular marks, a large amount of useless triples can be reduced
The transmission loss of data in a distributed system.The number of iterations of reasoning can be reduced by constructing transitive closure matrix, is improved
The efficiency of reasoning.Finally, by inference as a result, deleting duplicate triple data, in real time to further increase successive iterations
The efficiency of reasoning.Pushing away for RDFS/OWL rule can efficiently and be correctly realized in the case where data volume increases by this programme
Reason.
Compared with prior art, the present invention has the advantage that
1, the present invention constructs TRM, reduces the number of iterations of reasoning.
2, the present invention uses efficient strobe utility, reduces the expense of useless calculating and network transmission.
3, the present invention improves the efficiency of reasoning.
Detailed description of the invention
Fig. 1 is the overall framework figure of SPRM algorithm of the invention.
Fig. 2 is RDFS rule of the invention.
Fig. 3 is OWL Horst rule of the invention.
Fig. 4 is building TRM exemplary diagram of the invention.
Fig. 5 is the link variable hum pattern of building RDFS/OWL rule of the invention.
Fig. 6 is the specific mode data figure of link variable of the invention.
Fig. 7 is the mode data figure in link variable storage table of the invention.
Fig. 8 is the Framework for Reasoning figure of SPRM algorithm of the invention in the t times iteration.
Specific embodiment
The present invention will be further described with reference to the accompanying drawings and embodiments.
The present embodiment provides a kind of RDF data distributed semantic parallel inference methods, as shown in Figure 1, specifically including following
Step:
Step S1: loading mode triple constructs TRM, while according to RDFS/OWL rule, constructing may in every rule
The link variable information of connection；
Step S2: according to TRM and link variable information, create-rule markup model；
Step S3: being divided into two kinds of forms of single argument and multivariable for link variable, according to the type of TRM and link variable,
RDFS/OWL rule is divided into 5 seed types, separately designs different inference schemes；
Step S4: to Flag_Rulem=1 rule executes the parallel inference of RDFS/OWL rule, and exports intermediate knot
Fruit；
Step S5: the repetition triple in intermediate result is deleted；
Step S6: if including new SchemaTriple in intermediate result, TRM and regular marks model is updated, is returned
Return step S4；Otherwise, terminate.
In the present embodiment, in the step S1, the relevant parameter of triple is defined as follows:
Define 1: the Subject-Verb and object that mode triple SchemaTriple refers to triple are all in ontology file
It is defined in OntologyFile, i.e., are as follows:
Wherein, the sum of n intermediate scheme triple；
If v ∈ { Si,Pj,Ok,Then,
(Si,Pj,Ok)∈SchemaTriple (1)
Define 2: example triple InstanceTriple refer to Subject-Verb and object at least one in ontology file
It is undefined in OntologyFile, it is specific example, i.e., are as follows:
Wherein, n indicates the sum of example triple；
If v ∈ { Si,Pj,Ok,Then,
(Si,Pj,Ok)∈InstanceTriple
3: triple type mark Flag_TripleType is defined, for identity mode triple and example triple, institute
Triple type mark Flag_TripleType is stated to be defined as follows:
Wherein, n indicate triple total then,
Triple item type mark TripleItem_Flag, it is described for identifying subject or object in triple item
Triple item type mark is defined as follows:
Wherein, n indicates the sum of triple.If v ∈ { Si,Pj,Ok, then
4: mode ternary Groups List SchemaList is defined, for obtaining the mode triple collection of identical predicate or object
It closes, the mode ternary Groups List SchemaList is defined as follows:
Wherein, the sum of n intermediate scheme triple, then,
SchemaList=Om_list∪Pt_list
Wherein, Om_ list expression meets predicate Pj∈'s { rdf:type } and with identical object triplet sets, with
Object name；Pt_ list expression meets predicateAll triplet sets with identical predicate,
With predicate name, it is defined as follows:
Om_ list={ (Si,Pj,Ok)|Pj∈ { rdf:type } &k=m }
In the present embodiment, it in the step S1, is defined when constructing TRM as follows:
Definition 5: the transitive closure of digraph is defined as: the transitive closure of the vertex a n digraph can be defined as a n
Rank Boolean matrix T={ tik(1≤i, k≤n), if there are an active path, t by vertex i to vertex kik=1, otherwise tik=
0。
Define 6: transitive closure relational matrix Transitive closure relation matrix, referred to as TRM are tied
It closes and defines 4, defines 5 and WarShall algorithm building TRM；TRM is indicated in the transitive closure of digraph, is top with class or attribute
Point, using the predicate of triple as the Boolean matrix T={ t of relationshipik}；
Predicate is PjThe relational matrix of SchemaTriple be expressed as Pj_ TRM, wherein Pj∈{subClassOf,
SubPropertyOf, sameAs, equivalentClass, equivalentProperty },
IfThen Pj_TRM.tik=1, indicate predicate
For PjSchemaTriple subject SiTo object OkThere are direct relation, otherwise Pj_TRM.tik=0；
Wherein, TRM points are class relational matrix CTRM and relation on attributes matrix PTRM；Wherein, CTRM is according to predicate
SubClassOf, equivalentClass, sameAs define SC_TRM, EC_TRM, SAC_TRM respectively；PTRM is according to predicate
SubPropertyOf, equivalentProperty, sameAs define SP_TRM, EP_TRM, SAP_TRM respectively；Due to predicate
For the SchemaTriple relationship containing class and relation on attributes of sameAs, relationship is separated in the building of TRM, constructs SAC_ respectively
TRM and SAP_TRM.
The building TRM specifically includes the following steps:
Step S11: in conjunction with defining 4, obtain in all SchemaTriple respectively predicate be subPropertyOf and
The mode triple data of equivalentProperty, according to the SchemaTriple of acquisition and combine define 6 obtain with
SubPropertyOf and using equivalentProperty as the Boolean matrix of relationship；
Step S12: according to WarShall algorithm obtain it is all with subPropertyOf and with
EquivalentProperty is transitive relation value in the Boolean matrix of relationship, to generate transitive closure relational matrix SP_
TRM, EP_TRM complete the building of SP_TRM and EP_TRM.
In the present embodiment, it in the step S1, is defined when constructing link variable information as follows:
Define 7: the link variable LinkVar mode triple to be used to connect two former pieces in RDFS/OWL rule
, it is described according to rule, the quantity of link variable is greater than 1；By the link variable information of each rule with<key, value>
Form is stored in table Rulem_Table, and wherein key stores all mode triple items for former piece connection of the rule,
Value stores the mode triple item of the rule conclusion part.
In the present embodiment, in the step S3 specifically: SPRM algorithm is right according to the type of TRM and link variable
RDFS/OWL rule is classified；Fig. 2 is indicated using the form of RDFS- rule numbers, such as RDFS-1 when reference RDFS rule
In the 1st rule；It is indicated in Fig. 3 when reference OWL Horst rule using the form of OWL- rule numbers, such as OWL-4
4th rule.Meanwhile a rule name label is distributed to every rule, rule name label is corresponding to the rule
Title (for example, the rule name of rule OWL-4 is labeled as OWL-4).Specific rule classification is as follows:
5 seed types of RDFS/OWL rule it is as follows:
Class1: the rule that SchemaTriple is combined with SchemaTriple is directly made inferences (in Fig. 2 using TRM
Regular RDFS-5, RDFS-11；Rule OWL-6, OWL-7 in Fig. 3, OWL-9, OWL-10, OWL-12a, OWL-12b, OWL-12c,
OWL-13a、OWL-13b、OWL-13c)。
The rule that type 2:SchemaTriple is combined with InstanceTriple, makes inferences according to TRM, without knot
Close link variable information (rule RDFS-7, RDFS-9 in Fig. 2；Rule OWL-11 in Fig. 3).
The rule that type 3:SchemaTriple is combined with InstanceTriple, link variable information are single argument,
It needs that TRM is combined to make inferences (rule RDFS-2, RDFS-3 in Fig. 2).
The rule that type 4:SchemaTriple is combined with InstanceTriple, link variable information are single argument,
Without making inferences (rule OWL-3, OWL-8a, OWL-8b, OWL-14a, OWL-14b in Fig. 3) in conjunction with TRM.
The rule that type 5:SchemaTriple is combined with InstanceTriple, link variable information are multivariable
(rule OWL-1, OWL-2, OWL-4, OWL-15, OWL-16 in Fig. 3).
To any one rule is defined as: assuming that m rule is expressed as follows:
Rulem:Cm1,Cm2,…,Cmk,…,Cmn→Result
Definition rule flag F lag_Rulem, for marking whether the rule is the rule that impossible activate；
When for Flag_Rulem=0, then the condition for meeting the rule is implied the absence of, without activating the rule；Assuming that Pj
For RulemIn some former piece predicate, wherein Pj∈{subClassOf,subPropertyOf,sameAs,
EquivalentClass, equivalentProperty }, PjCorresponding TRM is set as Pj_TRM；
In conjunction with definition 6 and 7 are defined, regular marks Flag_RulemClass definition according to RDFS/OWL rule is as follows:
If 1) Rulem∈ { Class1, type 2 }, then
If 2) Rulem∈ { type 3, type 4, type 5 }, then
Due to RDFS rule 1, only one conditional statement of 4a, 4b, 6,8,10 in Fig. 2, reasoning Parallel Algorithm is transformed
Do not influence [9]；Rule 12 and 13 seldom occurs in RDF data in Fig. 2, in most of reasonings work without
It discusses [5,7~12].Thus, RDFS rule as described herein does not consider these above-mentioned rules.Due to OWL rule 5a, 5b in Fig. 3
The parallelization of reasoning is not influenced, thus, OWL rule-based reasoning as described herein does not consider this two rule in OWL Horst.
In the present embodiment, in the step S4, RDFS/OWL rule is completed parallel in conjunction with MapReduce Computational frame
Reasoning just completes reasoning regular in Class1 in the building TRM stage, is directed to class in the parallel inference stage of RDFS/OWL rule
Type 2, type 3, type 4 and type 5 carry out distributed parallel inference, to realize all in a MapReduce task
Reasoning of rule；
For the rule of type 2, type 3 and type 4, according to regular marks model and link variable information, by Map
The parallel inference in stage can obtain the reasoning results triple data；For the rule of type 5, by meeting Inference Conditions
Data carry out redundancy output to ensure the correctness of subsequent parallel data reasoning, i.e., will meet the example triple of same rule
Output is set as identical key key, and to ensure in the Reduce stage, the strictly all rules former piece in same rule can be according to connection
Variable is completed to connect and exports result；
The parallel reasoning algorithm of the RDFS/OWL rule in Map stage specifically:
Input: key is line number；Value is example triple
Output: key is the combination of the reasoning results triple or regular marks and link variable；Value is arbitrary value or band
The triple item of label；
The RDFS/OWL rule parallel reasoning algorithm in Reduce stage specifically:
Input: the output result of duplicate triple data phase is deleted
Output: key is output triple；Value is arbitrary value.
In the present embodiment, in the step S5, example triple is carried out by link variable information and regular marks
Filtering reduces the transmission loss of a large amount of useless triple data in a distributed system, is subtracted by constructing transitive closure matrix
The number of iterations of few reasoning, to improve the efficiency of reasoning；By inference as a result, deleting duplicate triple data in real time,
To further increase the efficiency of successive iterations reasoning；
Duplicate triple data are deleted after executing combiner and each reasoning after each Map,
Wherein, combiner is the repeated data exported in the deletion map stage, specifically:
Input: the output result of the parallel reasoning algorithm of the RDFS/OWL rule in Map stage
Output: key is the combination of the reasoning results triple or regular marks and link variable；Value is arbitrary value or band
The triple item of label；
The algorithm for repeating triple data is deleted after each reasoning specifically:
Input: key is line number；Value is triple
Output: the triple after output duplicate removal.
In the present embodiment, in conjunction with attached drawing and example to above method explanation specific as follows:
In the building TRM stage: SPRM algorithm is realized by building TRM has regular (Class1) reasoning of transitivity, to subtract
Few reasoning the number of iterations accelerates the efficiency of whole reasoning.According to 1 and definition 4 is defined, obtains SchemaTriple and be loaded into memory
In.Then in conjunction with defining 6, the TRM of each transitivity rule is constructed, in which: matrix dimensionality when CTRM is constructed is data set
Total class number, matrix dimensionality when PTRM is constructed are total attribute number of data set.
By taking predicate is the SchemaTriple of subPropertyOf, equivalentProperty as an example, describe in PTRM
The building process of SP_TRM and EP_TRM, and in conjunction with rule OWL-13a, OWL-13b, OWL- in rule RDFS-5 and Fig. 3 in Fig. 2
13c realizes the reasoning between SchemaTriple, specific steps are as follows:
Step S11: in conjunction with defining 4, obtain in all SchemaTriple respectively predicate be subPropertyOf and
The mode triple data of equivalentProperty according to the SchemaTriple of acquisition and combine definition 6 available
Using subPropertyOf and using equivalentProperty as the Boolean matrix of relationship.
Step S12: according to WarShall algorithm it is available it is all with subPropertyOf and with
EquivalentProperty is transitive relation value in the Boolean matrix of relationship, to generate transitive closure relational matrix SP_
TRM,EP_TRM.The building of SP_TRM and EP_TRM are just completed by the two steps, the process of building is as shown in Figure 4.
In addition, the transmitting sexual norm triplet rules of rule RDFS-5 in Fig. 1 are realized in the building for presenting SP_TRM in Fig. 4
Reasoning；Meanwhile according to the SP_TRM of generation, if SP_TRM.tik=SP_TRM.tki, then can in conjunction with rule OWL-13c in Fig. 3
With the mode triple for generating new predicate as equivalentProperty, for constructing EP_TRM.For EP_TRM, ifThen combine rule OWL-13a, OWL-13b in Fig. 3 that new predicate can be generated as subPropertyOf
Mode triple.Since the quantity of SchemaTriple is few, by constructing TRM in memory, class can be quickly finished
Regular reasoning in type 1, and the number of iterations in RDFS/OWL rule-based reasoning stage MapReduce task can be reduced.
In the link variable information phase that building may connect: needing to pass through change between each former piece of RDFS/OWL rule
The connection of amount, ability reasoning generate new triple data.Since transitivity rule can realize that transmitting pushes away by each TRM
Reason, therefore SPRM algorithm is established connection according to SchemaTriple data and RDFS/OWL rule and is become for non-transitivity rule
Scale, for storing the pattern information for meeting the rule former piece variable.If the link variable table of a certain rule is recorded as sky,
Mean that all mode triples in this data set do not meet the rule former piece, thus the rule can be judged swash
Rule living.The link variable that may be connected refers to the link variable of triggering rule, is each rules and regulations of RDFS/OWL rule
A corresponding table (Rulem_Table) is then established, is stored in Hbase.Wherein, the line unit (key) in every table stores the rule
The variable that may then connect, the variable that train value (value) may be used in the reasoning triple for rule generation.Building
The link variable of RDFS/OWL rule is as shown in Figure 5.
In order to construct the process of link variable in definitely RDFS/OWL rule, to omit prefix in LUBM data set
20 triple data instances, wherein number 1~11 is mode triple data, and number 12~20 is example triple data,
As shown in Figure 6.
In conjunction with defining 7, according to the 1~11 of table 2 SchemaTriple data, obtain specific in link variable storage table
Mode data is as shown in Figure 7.As a result, according to 8 regular marks that can be readily available each rule are defined, such as in this instance, advise
The then Flag_Rule of RDFS-3 (range rule)RDFS-3=0.
In the parallel inference stage of RDFS/OWL rule: due to just completing regular in Class1 push away in the building TRM stage
Reason, therefore distribution is carried out mainly for type 2, type 3, type 4 and type 5 in the parallel inference stage of RDFS/OWL rule
Parallel inference, thus in a MapReduce task realize all rule a reasonings.For 2~type of type 4
Rule can obtain the reasoning results three by the parallel inference in Map stage according to regular marks model and link variable information
Tuple data.For the rule of type 5 (multivariable), need to export by carrying out redundancy to the data for meeting Inference Conditions with true
Protect the correctness of subsequent parallel data reasoning；The example triple output that same rule will be met is set as identical key
(key), to ensure in the Reduce stage, it is same rule in strictly all rules former piece can according to link variable complete connection and it is defeated
Result out.
Specific step is as follows for the parallel reasoning algorithm of RDFS/OWL rule:
The RDFS/OWL rule parallel reasoning algorithm in algorithm 1.Map stage:
Input: key is line number；Value is example triple
Output: key is the combination of the reasoning results triple or regular marks and link variable；Value is arbitrary value or band
The triple item of label.
Specific algorithm is as shown in Table 1:
The RDFS/OWL rule parallel reasoning algorithm in algorithm 2.Reduce stage:
Input: the output result of algorithm 3
Output: key is output triple；Value is arbitrary value
Specific algorithm is as follows:
SPRM algorithm is as shown in Figure 8 for the Framework for Reasoning of the t times iteration of RDFS/OWL rule.
It is other kinds of for definitely RDFS/OWL rule since Class1 directly makes inferences in TRM matrix
Reasoning shows the process of reasoning in conjunction with LUBM data slot in Fig. 6 by way of example, and the arbitrary value output in example takes
“irrelevant”。
By taking rule RDFS-7 (subProperty example rule) in Fig. 2 as an example, show the reasoning process of 2 rule of type: root
According to Flag_Rule known to definition 8RDFS-7=1.It is attribute according to all parent attributes of the available headOf of SP_TRM
worksFor,memberOf；There was only (FullProfessor7, headOf, Department0) available activation in Fig. 6
RDFS-7 rule, thus can with reasoning generate triple (FullProfessor7, worksFor, Department0) and
(FullProfessor7,memberOf,Department0).If irrelevant can use arbitrary value, then in the key in Map stage
Value to output shaped like:
<" ResultFlag "+(FullProfessor7, worksFor, Department0), irrelevant>
<" ResultFlag "+(FullProfessor7, memberOf, Department0), irrelevant>
By taking rule RDFS-2 in Fig. 2 (domain rule) as an example, show the reasoning process of 3 rule of type: can according to defining 8
Know Flag_RuleRDFS-2=1；According to according to the example triple data inputted in Fig. 6, using the predicate of triple as line unit
(rowkey) train value (value) in inquiry Domain_Table, example triple (Lecturer0, teacherOf,
GraduateCourse5) enter the rule-based reasoning.Then, it is according to all parents of the available Faculty of SC_TRM
Employee, thus, it is possible to reasoning generate triple (Lecturer0, type, Faculty) and (Lecturer0, type,
Employee).Then exported in the key-value pair in Map stage are as follows:
<" ResultFlag "+(Lecturer0, type, Faculty), irrelevant>
<" ResultFlag "+(Lecturer0, type, Employee), irrelevant>
By taking rule OWL-8a (inverseOf) in Fig. 3 as an example, show the reasoning process of 4 rule of type: can according to defining 8
Know Flag_RuleOWL-8a=1.According to the example triple data inputted in Fig. 6, using the predicate of triple as line unit (rowkey)
Train value (value) in inquiry Inverse_8a_Table, only example triple (FullProfessor0, degreeFrom,
University1) can obtain train value (value) of its predicate in Inverse_8a_Table is hasAlumnus, thus
Reasoning generates triple (University1, hasAlumnus, FullProfessor0).Then exported in the key-value pair in Map stage
Shaped like:
<" ResultFlag "+(University1, hasAlumnus, FullProfessor0), irrelevant>
By taking rule OWL-4 (transitiveProperty) in Fig. 3 as an example, show the reasoning process of 5 rule of type: according to
SP_Flag_Ru le known to table 1OWL-4=1.According to the example triple data inputted in Fig. 6, using the predicate of triple as line unit
(rowkey) train value (value) in TransitiveProperty_Table, example triple are inquired
(ResearchGroup0, subOrganizationOf, Department0) and (Department0,
SubOrganizationOf, University0) column of its predicate in TransitiveProperty_Table can be obtained
Being worth (value) is subOrganizationOf.The key-value pair in Map stage exports are as follows:
<oWL-4+subOrganizationOf+Department0,0+ResearchGroup0>
<oWL-4+subOrganizationOf+Department0,1+University0>
In the Reduce stage, the data that key (key) is OWL-4, value (values) root in the same key (key) are obtained
It is subject or object that triple item caused by reasoning is obtained according to label 0 or 1, generates the triple data of final reasoning
Are as follows: (ResearchGroup0, subOrganizationOf, University0).Regular marks and its respective value table such as 1 institute of table
Show.
1 regular marks of table and its respective value table
The triple stage is repeated deleting: a large amount of duplicate triple numbers can be generated during executing distributing inference
According to if the data de-duplication that this node and this reasoning are not generated, these identical triples will generate more
Repetition triple data, cause the waste that system resource is meaningless.This algorithm by after each map execute combiner with
And duplicate triple data are deleted after each reasoning, to reduce the expense of network, promote whole Reasoning Efficiency.
Algorithm 3.Combiner deletes the repeated data that this node map stage exports
Input: the output result of map function in algorithm 1
Output: key is the combination of the reasoning results triple or regular marks and link variable；Value is arbitrary value or band
The triple item of label.
Specific algorithm is as follows:
This method is completed to delete by a Map/Reduce repeats triple algorithm, the specific steps are as follows:
Algorithm 4., which is deleted, repeats triple algorithm
map(key,value)
Input: key is line number；Value is triple
Output: the triple after output duplicate removal
Specific algorithm is as follows:
reduce(key,valueList)
Input: key is the key of map output；Value is corresponding valueList
Output: the triple after output duplicate removal
In the present embodiment, this algorithm analysis is as follows: the complexity and centralized algorithm complexity of SPRM algorithm
Analysis it is less identical, by the time complexity under the worst case of SPRM algorithm be divided into the Map stage time complexity and
The time complexity in Reduce stage.Assuming that the scale of data set is N, wherein mode triple is n,
In MapReduce the Map stage and line number be k, Reduce stage incoming example triple number is m, the Reduce stage
And line number is t.
Since SPRM algorithm is classified RDFS/OWL rule, after TRM building is completed, SPRM algorithm is parallel again
2~type of type, 5 rule-based reasoning is executed, therefore the complexity of SPRM algorithm is analyzed according to the type of RDFS/OWL rule,
Wherein the complexity of a reasoning iteration is respectively as follows:
1) complexity of Class1 rule-based reasoning is mainly the WarShall algorithm complexity for constructing TRM, is O (n3)
2) complexity of 2 rule-based reasoning of type:
The time complexity in Map stage are as follows: O (n2*N/k)
The time complexity in Reduce stage are as follows: O (m/t)
3) complexity of 3 rule-based reasoning of type:
The time complexity in Map stage are as follows: O (n*N/k)
The time complexity in Reduce stage are as follows: O (m/t)
4) complexity of 4 rule-based reasoning of type:
The time complexity in Map stage are as follows: O (N/k)
The time complexity in Reduce stage are as follows: O (m/t)
5) 5 rule-based reasoning complexity of type:
The time complexity in Map stage are as follows: O (n*N/k)
The time complexity in Reduce stage are as follows: O (m2/t)
In summary, the time complexity of reasoning iteration of SPRM algorithm is O (n)=O (n3)+O(n2*N/k)+O(n*
N/k)+O(m/t)+O(m2/t).Compared to data set, the number of mode triple n is considerably less, it can be considered that its magnitude is
Constant, and SPRM uses efficient strobe utility, reduction m as much as possible, to reduce overall time complexity.
The foregoing is merely presently preferred embodiments of the present invention, all equivalent changes done according to scope of the present invention patent with
Modification, is all covered by the present invention.
Claims (4)
1. a kind of RDF data distributed semantic parallel inference method, it is characterised in that: specifically includes the following steps:
Step S1: loading mode triple constructs TRM, while according to RDFS/OWL rule, and constructing may connection in every rule
Link variable information；
Step S2: according to TRM and link variable information, create-rule markup model；
Step S3: being divided into two kinds of forms of single argument and multivariable for link variable, will according to the type of TRM and link variable
RDFS/OWL rule is divided into 5 seed types, separately designs different inference schemes；
Step S4: to Flag_Rulem=1 rule executes the parallel inference of RDFS/OWL rule, and exports intermediate result；
Step S5: the repetition triple in intermediate result is deleted；
Step S6: if in intermediate result including new SchemaTriple, TRM and regular marks model are updated, returns to step
Rapid S4；Otherwise, terminate；
Wherein, in the step S1, the relevant parameter of triple is defined as follows:
Define 1: the Subject-Verb and object that mode triple SchemaTriple refers to triple are all in ontology file
It is defined in OntologyFile, i.e., are as follows:
Wherein, the sum of n intermediate scheme triple；
If v ∈ { Si,Pj,Ok,Then,
(Si,Pj,Ok)∈SchemaTriple (1)
Define 2: example triple InstanceTriple refer to Subject-Verb and object at least one in ontology file
It is undefined in OntologyFile, it is specific example, i.e., are as follows:
Wherein, n indicates the sum of example triple；
If v ∈ { Si,Pj,Ok,Then,
(Si,Pj,Ok)∈InstanceTriple
3: triple type mark Flag_TripleType is defined, for identity mode triple and example triple, described three
Tuple type flag F lag_TripleType is defined as follows:
Wherein, n indicate triple total then,
Triple item type mark TripleItem_Flag, for identifying subject or object in triple item, the ternary
Group item type mark is defined as follows:
Wherein, n indicates the sum of triple；If v ∈ { Si,Pj,Ok, then
4: mode ternary Groups List SchemaList is defined, for obtaining the mode triplet sets of identical predicate or object,
The mode ternary Groups List SchemaList is defined as follows:
Wherein, the sum of n intermediate scheme triple, then,
SchemaList=Om_list∪Pt_list
Wherein, Om_ list expression meets predicate Pj∈'s { rdf:type } and with identical object triplet sets, with the guest
Language name；Pt_ list expression meets predicateAll triplet sets with identical predicate, with this
Predicate name, is defined as follows:
Om_ list={ (Si,Pj,Ok)|Pj∈ { rdf:type } &k=m }
Wherein in the step S1, it is defined when constructing TRM as follows:
Definition 5: the transitive closure of digraph is defined as: the transitive closure of the vertex a n digraph can be defined as a n rank cloth
You are matrix T={ tik(1≤i, k≤n), if there are an active path, t by vertex i to vertex kik=1, otherwise tik=0；
Definition 6: transitive closure relational matrix Transitive closure relation matrix, referred to as TRM, in conjunction with fixed
Justice 4 defines 5 and WarShall algorithm building TRM；TRM is indicated in the transitive closure of digraph, using class or attribute as vertex, with
The predicate of triple is the Boolean matrix T={ t of relationshipik}；
Predicate is PjThe relational matrix of SchemaTriple be expressed as Pj_ TRM, wherein Pj∈{subClassOf,
SubPropertyOf, sameAs, equivalentClass, equivalentProperty },
IfThen Pj_TRM.tik=1, expression predicate is Pj
SchemaTriple subject SiTo object OkThere are direct relation, otherwise Pj_TRM.tik=0；
Wherein, TRM points are class relational matrix CTRM and relation on attributes matrix PTRM；Wherein, CTRM according to predicate subClassOf,
EquivalentClass, sameAs define SC_TRM, EC_TRM, SAC_TRM respectively；PTRM is according to predicate
SubPropertyOf, equivalentProperty, sameAs define SP_TRM, EP_TRM, SAP_TRM respectively；Due to predicate
For the SchemaTriple relationship containing class and relation on attributes of sameAs, relationship is separated in the building of TRM, constructs SAC_ respectively
TRM and SAP_TRM；
The building TRM specifically includes the following steps:
Step S11: in conjunction with defining 4, obtain in all SchemaTriple respectively predicate be subPropertyOf and
The mode triple data of equivalentProperty, according to the SchemaTriple of acquisition and combine define 6 obtain with
SubPropertyOf and using equivalentProperty as the Boolean matrix of relationship；
Step S12: it is obtained according to WarShall algorithm and all is with subPropertyOf and with equivalentProperty
Transitive relation value in the Boolean matrix of relationship completes SP_TRM to generate transitive closure relational matrix SP_TRM, EP_TRM
With the building of EP_TRM；
Wherein, it in the step S1, is defined when constructing link variable information as follows:
Define 7: the link variable LinkVar mode triple item to be used to connect two former pieces in RDFS/OWL rule, root
It is described according to rule, the quantity of link variable is greater than 1；By the link variable information of each rule with<key, value>form
It is stored in table Rulem_Table, wherein key stores all mode triple items for former piece connection of the rule, and value is deposited
Store up the mode triple item of the rule conclusion part.
2. a kind of RDF data distributed semantic parallel inference method according to claim 1, it is characterised in that:
In the step S3 specifically: using SPRM algorithm according to the type of TRM and link variable, carried out to RDFS/OWL rule
Classification, quote RDFS rule when using RDFS- rule numbers form, quote OWL Horst rule when using OWL- rule compile
Number form；A rule name label is distributed to every rule, rule name label is title corresponding to the rule；
5 seed types of RDFS/OWL rule it is as follows:
Class1: the rule that SchemaTriple is combined with SchemaTriple is directly made inferences using TRM；
The rule that type 2:SchemaTriple is combined with InstanceTriple, makes inferences according to TRM, without the company of combination
Connect variable information；
The rule that type 3:SchemaTriple is combined with InstanceTriple, link variable information are single argument, are needed
It is made inferences in conjunction with TRM；
The rule that type 4:SchemaTriple is combined with InstanceTriple, link variable information are single argument, are not necessarily to
It is made inferences in conjunction with TRM；
The rule that type 5:SchemaTriple is combined with InstanceTriple, link variable information are multivariable；
To any one rule is defined as: assuming that m rule is expressed as follows:
Rulem:Cm1,Cm2,…,Cmk,…,Cmn→Result
Definition rule flag F lag_Rulem, for marking whether the rule is the rule that impossible activate；
When for Flag_Rulem=0, then the condition for meeting the rule is implied the absence of, without activating the rule；Assuming that PjFor
RulemIn some former piece predicate, wherein Pj∈{subClassOf,subPropertyOf,sameAs,
EquivalentClass, equivalentProperty }, PjCorresponding TRM is set as Pj_TRM；
In conjunction with definition 6 and 7 are defined, regular marks Flag_RulemClass definition according to RDFS/OWL rule is as follows:
If 1) Rulem∈ { Class1, type 2 }, then
If 2) Rulem∈ { type 3, type 4, type 5 }, then
3. a kind of RDF data distributed semantic parallel inference method according to claim 1, it is characterised in that:
In the step S4, the reasoning of RDFS/OWL rule is completed parallel in conjunction with MapReduce Computational frame, in building TRM rank
Section just completes reasoning regular in Class1, is directed to type 2, type 3,4 and of type in the parallel inference stage of RDFS/OWL rule
Type 5 carries out distributed parallel inference, to realize a reasoning of all rules in a MapReduce task；
For the rule of type 2, type 3 and type 4, according to regular marks model and link variable information, by the Map stage
Parallel inference, the reasoning results triple data can be obtained；For the rule of type 5, by the number for meeting Inference Conditions
According to redundancy output is carried out to ensure the correctness of subsequent parallel data reasoning, i.e., the example triple for meeting same rule is exported
It is set as identical key key, to ensure in the Reduce stage, the strictly all rules former piece in same rule can be according to link variable
It completes to connect and exports result；
The parallel reasoning algorithm of the RDFS/OWL rule in Map stage specifically:
Input: key is line number；Value is example triple
Output: key is the combination of the reasoning results triple or regular marks and link variable；Value is arbitrary value or tape label
Triple item；
The RDFS/OWL rule parallel reasoning algorithm in Reduce stage specifically:
Input: the output result of duplicate triple data phase is deleted
Output: key is output triple；Value is arbitrary value.
4. a kind of RDF data distributed semantic parallel inference method according to claim 1, it is characterised in that:
In the step S5, example triple is filtered by link variable information and regular marks, is reduced a large amount of useless
Triple data transmission loss in a distributed system, the number of iterations of reasoning is reduced by construction transitive closure matrix,
To improve the efficiency of reasoning；It is subsequent to further increase by inference as a result, deleting duplicate triple data in real time
The efficiency of iteration reasoning；
Duplicate triple data are deleted after executing combiner and each reasoning after each Map,
Wherein, combiner is the repeated data exported in the deletion map stage, specifically:
Input: the output result of the parallel reasoning algorithm of the RDFS/OWL rule in Map stage
Output: key is the combination of the reasoning results triple or regular marks and link variable；Value is arbitrary value or tape label
Triple item；
The algorithm for repeating triple data is deleted after each reasoning specifically:
Input: key is line number；Value is triple
Output: the triple after output duplicate removal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610293055.5A CN105912721B (en) | 2016-05-05 | 2016-05-05 | RDF data distributed semantic parallel inference method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610293055.5A CN105912721B (en) | 2016-05-05 | 2016-05-05 | RDF data distributed semantic parallel inference method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105912721A CN105912721A (en) | 2016-08-31 |
| CN105912721B true CN105912721B (en) | 2019-06-07 |
Family
ID=56753523
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610293055.5A Active CN105912721B (en) | 2016-05-05 | 2016-05-05 | RDF data distributed semantic parallel inference method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105912721B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106445913A (en) * | 2016-09-06 | 2017-02-22 | 中南大学 | MapReduce-based semantic inference method and system |
| CN107016110B (en) * | 2017-04-15 | 2019-12-17 | 福州大学 | OWLHorst rule distributed parallel reasoning algorithm combined with Spark platform |
| CN108763451B (en) * | 2018-05-28 | 2022-03-11 | 福州大学 | Streaming RDF data parallel reasoning algorithm based on Spark Streaming |
| CN108875953A (en) * | 2018-06-08 | 2018-11-23 | 福州大学 | A kind of complex rule reasoning design method extending DL operator |
| CN111309928A (en) * | 2020-02-21 | 2020-06-19 | 广东电网有限责任公司 | Operation and maintenance knowledge base construction method based on Bayes and semantic analysis |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104615703A (en) * | 2015-01-30 | 2015-05-13 | 福州大学 | RDF data distributed parallel inference method combined with Rete algorithm |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101008691B1 (en) * | 2008-12-15 | 2011-01-19 | 한국과학기술정보연구원 | System and Method for Hybrid Rete Reasoning based on In-Memory and DBMS |
-
2016
- 2016-05-05 CN CN201610293055.5A patent/CN105912721B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104615703A (en) * | 2015-01-30 | 2015-05-13 | 福州大学 | RDF data distributed parallel inference method combined with Rete algorithm |
Non-Patent Citations (4)
| Title |
|---|
| MPPIE:基于消息传递的RDFS并行推理框架;吕小玲 等;《计算机科学与探索》;20150812;第10卷(第4期);第451-465页 |
| Scalable RDF store based on HBase and MapReduce;Sun Jianling 等;《International Conference on Advanced Computer Theory and Engineering IEEE》;20101231;第633-636页 |
| 基于云计算的海量语义信息并行推理方法研究;施惠俊;《中国优秀硕士学位论文全文数据库 信息科技辑》;20120715(第07期);第I138-2683页 |
| 基于大规模模糊RDF数据的推理引擎;刘畅;《中国优秀硕士学位论文全文数据库 信息科技辑》;20120715(第07期);第I138-2700页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105912721A (en) | 2016-08-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105912721B (en) | RDF data distributed semantic parallel inference method | |
| US11763175B2 (en) | Systems and methods for semantic inference and reasoning | |
| CN105320756B (en) | A kind of database association rule digging method based on improvement Apriori algorithm | |
| Zerhari et al. | Big data clustering: Algorithms and challenges | |
| CN104050202B (en) | Method and apparatus for searching for database | |
| Liu et al. | An incremental and distributed inference method for large-scale ontologies based on mapreduce paradigm | |
| KR20130095250A (en) | Processing related datasets | |
| US20140244657A1 (en) | Dynamic graph system for a semantic database | |
| Nohuddin et al. | Finding “interesting” trends in social networks using frequent pattern mining and self organizing maps | |
| US20160328433A1 (en) | Representing Large Body of Data Relationships | |
| Wang et al. | Distributed Pregel-based provenance-aware regular path query processing on RDF knowledge graphs | |
| CN108520035A (en) | SPARQL parent map pattern query processing methods based on star decomposition | |
| Terdjimi et al. | Hylar+ improving hybrid location-agnostic reasoning with incremental rule-based update | |
| CN106445645A (en) | Method and device for executing distributed computation tasks | |
| Peng et al. | Optimization research of decision support system based on data mining algorithm | |
| Xie et al. | Active zero-shot learning: a novel approach to extreme multi-labeled classification | |
| Ravichandran | Big Data processing with Hadoop: a review | |
| CN104239581A (en) | Database-system-oriented replicated data provenance tracing method | |
| Li et al. | A rule-based knowledge discovery engine embedded semantic graph knowledge repository for retail business | |
| Jing et al. | SPARQL graph pattern rewriting for OWL-DL inference queries | |
| Alotaibi et al. | Big data mining: A classification perspective | |
| Lee et al. | Ontology management for large-scale e-commerce applications | |
| Beheshti et al. | Bp-sparql: A query language for summarizing and analyzing big process data | |
| Kim et al. | Optimizing RDF (S) queries on cloud platforms | |
| Bellomarini et al. | Model-Independent Design of Knowledge Graphs-Lessons Learnt From Complex Financial Graphs. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |