Patent Application
20190108452
Embodiments of the present specification relate generally to management of domain knowledge, and more particularly to systems and methods for capturing, modelling, and using domain knowledge in industrial applications.
Industrial applications in areas such as the healthcare sector or manufacturing sector require management of domain knowledge. Improper processing of information in such applications may lead to substantial increase in costs. As one example, in the healthcare system, insurance companies may deny reimbursement of medical expenditure claims for reasons such as incomplete data provided by the patients. As another example, in a manufacturing setup, parts specifications may not be reproducible due to infrastructure limitations or lack of process maturity. In such instances, multiple resubmissions of insurance claims or repeated redesigning of parts may be required, thereby necessitating higher costs and prolonged cycle times for industrial processes.
Establishing efficient knowledge management processes requires semantic models representing domain knowledge provided by various stake holders including subject matter experts (SMEs). Domain knowledge is represented in semantic based techniques using one of a variety of description logic languages such as a Web Ontology Language (OWL), which is a recommendation of the World Wide Web consortium (W3C). Capturing requirements from SMEs and encoding the captured requirements into a description language is an iterative process. Also, encoding the captured requirements into the description language involves human interactions and associated problems such as manual review, probability of introduction, and oversight of logical and syntactical errors. Disadvantageously, such errors may not be detectable during early stages such as during development of industrial applications. Such delay in encoding of the captured requirements may lead to undesirable time delays and project costs. Typically, description languages are not natural languages and knowledge encoding requires programming skills. Further, the encoded requirements may not be easily verifiable by the SMEs. Additionally, use of variables in the syntax of rules adversely affects human readability and prevents the SMEs from authoring domain rules.
Recent advances in formal Controlled Natural Languages (CNLs) have improved the process of capturing the requirements in a way such that a subject matter expert (SME) can more easily evaluate and verify the captured requirements. An open source controlled English language called Semantic Application Design Language (SADL) licensed under Eclipse Public License has successfully been used to express constructs of the OWL. SADL also supports rules representative of implications in a First Order Logic (FOL). However, existing CNLs employ syntax that requires use of one or more variables for authoring rules. Inductive logic programming technique is employed to capture domain knowledge. However, SMEs still need to interact with programmers in designing or modifying the semantic model.
In accordance with one aspect of the present specification, a method is disclosed. The method includes receiving event data corresponding to an industrial application and generating at least one inference concept based on the event data. The method also includes obtaining a semantic model comprising a plurality of inference concepts, a plurality of relationships among the plurality of inference concepts and a plurality of concept rules representative of domain knowledge. The plurality of concept rules is authored using the plurality of inference concepts and the plurality of relationships. Moreover, the method includes processing the at least one inference concept based on the semantic model to generate inferential data. The inferential data is representative of an inference corresponding to the event data. Additionally, the method includes controlling the industrial application based on the inferential data.
In accordance with another aspect of the present specification, a system is disclosed. The system includes a data input unit configured to receive event data corresponding to an industrial application. Further, the system includes an inference engine having a semantic model that includes a plurality of inference concepts, a plurality of relationships among the plurality of inference concepts, and a plurality of concept rules representative of domain knowledge. The plurality of concept rules is authored using the plurality of inference concepts and the plurality of relationships. The inference engine further includes a knowledge encoder unit communicatively coupled to the data input unit and configured to generate at least one inference concept based on the event data. The inference engine also includes an evaluation unit communicatively coupled to the knowledge encoder and configured to process the at least one inference concept based on the semantic model to generate inferential data. The inferential data is representative of an inference corresponding to the event data. In addition, the system includes an output unit communicatively coupled to the inference engine and configured to control the industrial application based on the inferential data.
In accordance with another aspect of the present specification, a non-transitory computer readable medium having instructions to enable at least one processor unit is disclosed. The instructions enable the at least one processor unit to receive event data corresponding to an industrial application and generate at least one inference concept based on the event data. Moreover, the instructions enable the at least one processor unit to obtain a semantic model comprising a plurality of inference concepts, a plurality of relationships among the plurality of inference concepts, and a plurality of concept rules representative of domain knowledge. The plurality of concept rules is authored using the plurality of inference concepts and the plurality of relationships. Also, the instructions enable the at least one processor unit to process the at least one inference concept based on the semantic model to generate inferential data. The inferential data is representative of an inference corresponding to the event data. The instructions also enable the at least one processor unit to control the industrial application based on the inferential data.
These and other features and aspects of embodiments of the present invention will become better understood when the following detailed description is read with reference to the accompanying drawings in which like characters represent like parts throughout the drawings, wherein:
As will be described in detail hereinafter, systems and methods for management of domain knowledge are presented. More particularly, systems and methods for capturing, modelling, and utilizing domain knowledge in industrial applications are presented.
The term “semantic web” refers to a web of data that can be processed directly or indirectly by machines. The term “semantic model” refers to one or more logical and mathematical expressions formulated from variables representative of inference concepts, relationships, and rules. The semantic web organizes information and enables knowledge processing using ontology based semantic models incorporating web resources. The term “industrial application” as used herein refers to an application related to an industrial system such as a healthcare management system or a manufacturing system. The term “knowledge management system” as used herein refers to an analytical engine or an inference engine used in the industrial application.
Further, the term “ontology” refers to a specification of a plurality of inference concepts along with relationships between two or more of the plurality of inference concepts. In object-oriented programming, ontologies may be considered as domain classes having logical statements describing inference concepts, their properties, and relationships between the inference concepts. Tools referred to herein as reasoners, are employed by the semantic web to process rules to perform advanced queries and extract implicit relationships among resources. The term “ontology language” refers to a formal language used for constructing ontologies. An ontology is a formal explicit specification of a shared conceptualization of a domain of interest. Ontology languages are capable of encoding knowledge about specific domains and including reasoning rules that support processing of domain specific knowledge. Ontology languages may be interchangeably and equivalently referred as “declarative languages.” The term “first order logic” or “FOL” refers to a prepositional logic combined with objects, properties, relations, and functions.
In a presently contemplated configuration, the system 100 includes a data input unit 102, the knowledge processing subsystem 118, and an output unit 114. The knowledge processing subsystem 118 includes a knowledge encoder unit 104, a semantic model 106, an evaluation unit 108, a processor unit 110, and a memory unit 112. The various components of the knowledge processing subsystem 118 may be interconnected with each other by a communications bus 116. The communication bus 116 may represent a wired or wireless connection.
The data input unit 102 is configured to receive event data from a user. The data input unit 102 may be a part of an integrated development environment (IDE) used to design and implement the knowledge management system 100. In an embodiment where the industrial application is a healthcare system, the event data may be a reimbursement claim submission. In an embodiment where the industrial application is a manufacturing system, the event data may be a specification of a part to be manufactured. The data input unit 102 may be a keyboard, a display, a file reader, a microphone, a video camera, or any other suitable input device, or combinations thereof. The event data may be either processed in real-time by the knowledge processing system 118 or stored in the memory for off-line processing.
In one embodiment, the data input unit 102 is also configured to receive domain knowledge from one or more subject matter experts (SMEs) for example, using natural languages. The data input unit 102 is also configured to accept declarations using keywords, phrases, and prepositions as first-order logical statements. In one embodiment, the data input unit 102 is configured to assist the user in authoring concept rules involving inference concepts and relationships related to a specific domain of knowledge. In particular, the data input unit 102 enables the SMEs to author concept rules using a ‘Crule language.’ The term ‘Crule language’ is used herein to refer to a rule authoring language based on natural language constructs without the need of using variables. The Crule language includes constructs such as, but not limited to, cardinality, disjunction, chaining of relations, and existential quantifier to represent complex knowledge in the rules. Consequently, the SMEs are not required to use variables while capturing domain knowledge and authoring the concept rules.
The concept rules are representative of domain knowledge. In one example, in the domain of education, inference concepts such as ‘Professor,’ ‘Administrator,’ and ‘Class’ are employed. Further, a plurality of properties such as ‘teaching’ and ‘availableToTeach’ are also employed. The domain of education is also defined by a plurality of relationships between a plurality of education related topics. Specifically, a domain specific rule such as a condition of availability of a ‘Professor’ to teach a ‘Class’ to be authored by a subject matter expert (SME) may be expressed as “if a professor teaches a class and the professor is not an administrator then that professor is available to teach another class.”. When the SME starts authoring the rule via the data input unit 102, a concept rule characterizing the domain specific rule is represented as:
if a Professor teaches a Class and
the Professor is not an Administrator
then
the Professor is availableToTeach another Class
In the example of the domain specific rule, Professor-Classes is the name of the rule. Also, the italicized words are fillers and keywords, and the other words are concepts and properties. The SME may specify a name for the rule such as ‘Professor-Classes.’ The data input unit 102 may assist the SME in authoring rules by analyzing the text already entered through the data input unit 102. In one embodiment, the data input unit 102 may be provided with a library of keywords and syntax to use the keywords usable for authoring concept rules. Further, the data input unit 102 may also provide indentations on a display used by the SME to author the rules. In one embodiment, the data input unit 102 may be provided with a text parsing software for analyzing the text entered by the SME and assisting the SME to author rules. The text parsing software may also be configured to verify the syntax of the rules and accept the rules after confirmation by the SME. In one embodiment, the SME may complete authoring a concept rule and submit the concept rule to be stored in the memory unit 112 as part of the semantic model 106.
The knowledge encoder unit 104 is communicatively coupled to the data input unit 102 and configured to receive the concept rule from the memory unit 112 and the event data from the data input unit 102 or from the memory unit 112. The knowledge encoder unit 104 is configured to generate a description language version of the concept rule based on a description logic language. In one embodiment, the knowledge encoder unit 104 is configured to process the event data received from the data input unit 102 and generate at least one inference concept based on the event data. In certain embodiments, the knowledge encoder unit 104 is configured to generate one or more inference concepts and one or more relationships based on the event data. It may be noted that the generated one or more inference concepts is a subset of the plurality of inference concepts and the generated one or more relationships is a subset of the plurality of relationships provided by or stored in the semantic model 106.
Further, the knowledge encoder unit 104 is also configured to process the text entered by the user or text stored in the memory unit 112 using a natural language processing technique. In one embodiment, the knowledge encoder unit 104 is configured to parse the received concept rule. Moreover, the knowledge encoder unit 104 is also configured to translate the concept rule specified by the user into the description logic language based on the parsed text. In one embodiment, Semantic Application Design Language (SADL) is used to represent rules. In another embodiment, PROLOG, a general purpose logic programming language is used to represent rules and other user requirements. For example, the concept rule Professor-Classes presented hereinabove is represented in SADL as:
Rule Professor-Classes:
c1 is a Class and
p1 teaches c1 and
p1 is not an Administrator and
c2 is a Class and
c1 != c2
p1 is availableToTeach of c2.
where, the terms p1, c1, and c2 are variables.
In some embodiments, the knowledge encoder unit 104 may include a plurality of translators. In some of these embodiments, the user may be required to specify a preferred translator among the plurality of translators based on compatibility of system hardware and legacy software components. In one embodiment, the plurality of translators may include one or more of a SADL translator, a PROLOG translator, and the like.
The semantic model 106 refers to a domain knowledge represented as a plurality of inference concepts, a plurality of relationships among two or more of the plurality of inference concepts, a plurality of rules for processing the inference concepts, and relationships, and combinations thereof. An existing domain specific ontology is used to develop the semantic model 106 based on description logic syntax. The domain specific ontology may be shared across systems in a specific domain and may be prepared or updated in a collaborative way by a SMEs. In a non-limiting embodiment, a semantic modelling language includes one or more of SADL and OWL. The semantic model 106 is generated by integrating logical inference provided by the ontology and the rule-based inference specified by the SMEs.
The evaluation unit 108 is communicatively coupled to the knowledge encoder unit 104 and the semantic model 106 and configured to process the at least one generated inference concept based on the semantic model 106 to generate an inferential data. In one embodiment, the inferential data may include a recommendation by the semantic model 106 to the user for suitably modifying the event data. In the example of the healthcare system, the recommendation may suggest modification to one or more aspects of an insurance claim to reduce or minimize denial of the insurance claim. In another embodiment, the inferential data may include an evaluation report on manufacturability of a part specification. The evaluation unit 108 is further configured to modify the semantic model 106 based on the inferential data and a corresponding desired inferential data. The desired inferential data may be available from a user or a memory location. The desired inferential data corresponds to expected inferential data from the semantic model 106. The evaluation unit 108 is configured to process the event data through the use of the one or more generated inference concepts based on the semantic model 106. The processing of the event data may also include use of one or more generated relationships derived by the knowledge encoder unit 104.
The output unit 114 is communicatively coupled to the evaluation unit 108 and configured to present the inferential data to an industrial application or to a user. In one embodiment, the output unit 114 is configured to control the industrial application based on the inferential data. Specifically, the output unit 114 may retrieve one or more recommendations corresponding to the event data from the inferential data based on desired inferential data. The recommendations are generally representative of suggestions to modify the event data or actions to modify some aspect of the knowledge management system. Further, the output unit 114 is configured to modify the event data based on the recommendation. In the application of a manufacturing system, the output unit 114 may process the inferential data to retrieve a recommendation such as an indication of a design change and one or more parameters to be considered by a computer-aided design (CAD) designer for modifying the design. In an insurance approval system, the output unit 114 may process the inferential data to retrieve a recommendation such as a binary variable indicating an approval or a denial decision about an insurance claim. In one embodiment, the output unit 114 may generate recommendations to modify a medical reimbursement claim. The recommendation may include suggestions of additional documents to be provided and/or procedural steps to be followed while preparing or resubmitting the insurance claim.
The processor unit 110 includes at least one of a general-purpose computer, a graphics processing unit (GPU), a digital signal processor, and a controller. In other embodiments, the processor unit 110 includes a customized processor element such as, but not limited to, an application-specific integrated circuit (ASIC) and a field-programmable gate array (FPGA). In some embodiments, the processor unit 110 may perform one or more functions of at least one of the knowledge encoder unit 104, the evaluation unit 108, and the data input unit 102. In one embodiment, the processor unit 110 may be configured to receive commands, parameters from an operator via a console that has a keyboard or a mouse or data from the data input unit 102. The processor unit 110 may also be configured to receive clauses of a concept rule from the data input unit 102 and store the clauses of the concept rules in the memory unit 112. The processor unit 110 may include more than one processor co-operatively working with one another for performing intended functionalities. The processor unit 110 is further configured to store and retrieve contents into and from the memory unit 112. In one embodiment, the processor unit 110 is configured to initiate and control the functionality of at least one of the data input unit 102, the knowledge encoder unit 104, and the evaluation unit 108.
In one embodiment, the memory unit 112 may be a random-access memory (RAM), read only memory (ROM), flash memory or any other type of computer readable memory accessible by at least one of the data input unit 102, the knowledge encoder unit 104, and the evaluation unit 108. The memory unit 112 is also configured to store the semantic model 106 corresponding to the industrial application. In one embodiment, the memory unit 112 may be a non-transitory computer readable medium encoded with a program having a plurality of instructions to instruct at least one of the data input unit 102, the knowledge encoder unit 104, and the evaluation unit 108 to perform a sequence of steps to generate the inferential data corresponding to an industrial application. The program may be used to further instruct the processor unit 110 to control the industrial application.
In one embodiment, a non-transitory computer readable medium is encoded with instructions that enable the processor unit 110 to assist SMEs to author domain rules. The instructions enable the processor unit 110 to receive event data corresponding to an industrial application from the data input unit 102. Further, the instructions enable the processor unit 110 to perform the functionality of the knowledge encoder unit 104. The instructions enable the processor unit 110 to parse the event data to generate one or more inference concepts and instance relationships. The instructions are also configured to evaluate the one or more inference concepts and instance relationships to determine an inferential data by performing the functions of the evaluation unit 108. In one embodiment, the instructions enable the processor unit 110 to accept new concept rules authored by the SME, update the semantic model 106, generate an inferential data, and control the industrial application based on the inferential data.
The method further includes obtaining a semantic model, such as the semantic model 106 of
The method at step 206, includes generating at least one inference concept based on the event data. The generation of at least one inference concept is performed by knowledge encoder unit 104 of
Also, the at least one generated inference concept is processed based on the semantic model 106 to generate inferential data, as indicated by step 208. At step 208, clauses of a concept rule are evaluated by performing one or more knowledge processing steps to generate the inferential data. Processing of the at least one generated inference concept includes parsing of clauses of a concept rule to generate inference concepts, relationships, and keywords. The parsing further includes identification and removal of filler words in the clauses. The parsing also includes associating the inference concepts and the relationships with knowledge processing steps directed by keywords.
In one embodiment, the processing of step 208 includes generating the inferential data by processing the event data via evaluating the concept rule using one or more generated inference concepts and one or more generated relationships. The evaluation unit 108 of
The method also includes controlling the industrial application based on the inferential data, as depicted in step 210. In the application of the manufacturing system, the processor unit 108 may generate an indication to the CAD designer to modify a design of a part. In another embodiment, controlling the industrial application may include generating parameters or aspects to be modified in the CAD design before proceeding towards subsequent steps of manufacturing. In the insurance approval system, controlling includes an approval or a denial decision about an insurance claim. In one embodiment, controlling the industrial application includes generating recommendations to modify the medical reimbursement claim with suggestions of additional documents to be provided and/or procedural steps to be followed.
The schematic 300 includes a parser 302 configured to parse a plurality of concept rules of the semantic model 106 of
The schematic 300 also includes a plurality of translators 308 such as a SADL translator 310 and a PROLOG translator 312. It may be noted that the SADL translator 310 is configured to introduce one or more variables in an SADL rule so that the concept rule may be handled by existing knowledge based platforms. In one embodiment, choice of a description language translator is specified by a user.
In another embodiment, the choice of the descriptor language translator is determined based on at least one of a target language and software platform used by the inference engine. By way of example, the knowledge management system 100 built using SADL and a Jena engine uses an SADL translator.
Further, the schematic 300 includes a target language translator 314 configured to translate the description logic language rules to a target language based on the OWL file 306. In one embodiment, the SADL rules are translated to Jena rules by a Jena translator. It may be noted that in one embodiment, the target language rules may be translated to one of the description logic language rules by the target language translator. In another embodiment, the description logic language rules may be translated to a concept rule by a corresponding description language translator.
In the illustrated embodiment of
As another example in the manufacturing system, a concept rule for identifying a pad fillet is given as:
Rule findPadFillet1:
if a BlendingFace has edge an IntersectionEdge
with edgeAdjacencyType TANGENT and
the BlendingFace has edge a second IntersectionEdge
with edgeAdjacencyType TANGENT and
the first IntersectionEdge != second IntersectionEdge and
the second IntersectionEdge has connectedFaces a PartFace and
the PartFace is a CylindricalFace or ConicalFace and
the PartFace is not concave and
the PartFace is floorFace and
a second PartFace has edge the first IntersectionEdge and
the second PartFace != the BlendingFace and
the second PartFace != the first PartFace and
the second PartFace does not sharesVertex
with the first PartFace and
the second PartFace has edge a Part Edge
with edgeAdjacencyType CONVEX and
the PartEdge != the first IntersectionEdge
then
there exists a PadFillet
satisfying (PadFillet has featureFace the BlendingFace)
such that
where findPadFillet1 in the header (first line) of the concept rule is the name of the rule, the words in italics are fillers and keywords, and the other words in the body (second line onwards) of the concept rule are domain specific inference concepts and relationships.
In one embodiment, the Crule language constructs are formulated to enable avoiding use of variables in authoring rules. Within a rule, the indefinite articles ‘a’ and ‘an’ are used to introduce a concept or a relationship, and the definite article ‘the’ is used to refer to the same concept or relationship subsequently within the rule. As an example, in the rule of findPadFillet1, a first instance of the concept is recited as ‘a BlendingFace’ in the second line, and subsequent instances of the concept are recited as ‘the BlendingFace’ in the fourth line, the twelfth line, and the twentieth line. In one embodiment, multiple instances of the same type are distinguished by referring to the first instance as ‘a first . . . ’, the second instance as ‘a second . . . ,’ and so on. Subsequent references to the same instances may take the form of ‘the first . . . ’, ‘the second . . . ,’ and the like. When there are two instances of the same type, the second one may be referred to as ‘another’ for the first time. In the example of findPadFillet1 rule, a first PartFace is referred as ‘a PartFace’ on line seven and as a second PartFace is referred to as ‘a second PartFace’ on line eleven.
It may be noted herein that the same rule may be authored in Crule language in different ways. These different Crule language formats may be translated to the same rule in a description language. For example, the clause ‘a PartialTurnedFace has featureFace some PartFace’ may also be authored as ‘a PartialTurnedFace has featureFace a PartFace’. In some instances, two clauses of a rule may be combined as a single clause. As an example, the clauses ‘the PartFace has edge a CircularEdge’ and ‘the CircularEdge has edgeAdjacencyType CONCAVE’ may be combined in a single clause as ‘the PartFace has edge a CircularEdge with edgeAdjacencyType CONCAVE.’
The Crule language is configured to accommodate disjunction in a natural way in contrast to Jena and other semantic web rules languages. As an example, on line eight of the ‘findPadFillet1’ rule, “PartFace is a CylindricalFace or ConicalFace” is representative of a disjunction which gets translated to an ‘oneOf’ construct of the SADL. The Crule language is further configured to represent the getInstance( . . . ) construct of SADL as ‘there exists’ which is easily relatable to first order logic. The Crule language is also configured to represent the ‘countMatches’ construct of SADL by defining a property ‘sharesVertex.
As another example, an SADL version of the concept rule ‘findPadFillet1’ is given by:
Rule findPadFillet1:
fface is a BlendingFace
fface has edge oedge
oedge has edgeAdjacencyType TANGENT
oedge is a IntersectionEdge
fface has edge bedge
oedge != bedge
bedge has edgeAdjacencyType TANGENT
bedge is a IntersectionEdge
bedge has connectedFaces bface
bface is a t1
oedge has connectedFaces oface
fface != oface
oface != bface
bface, edge, xe2, xe2, vertex, xv1) = 0
oface has edge e3
e3 != oedge
e3 has edgeAdjacencyType CONVEX
fillet1 = getInstance(PadFillet, featureFace, fface)
where words in bold are representative of variables and other words are representative of inference concepts, keywords, relationships, and filler words.
The SADL version of the concept rule includes disjunction construct ‘oneOf( ),’ a construct getInstance( ), and a construct countMatches( ), which were more intuitively expressed in the Crule language.
In one embodiment, the Crule language is configured as a controlled natural language (CNL) enabling the SMEs to represent domain knowledge. A framework, referred herein as PENS framework, for classifying and comparing the controlled natural languages (CNLs) on four parameters of precision, expressiveness, naturalness, and simplicity on a scale of 1 to 5 is used to characterize the CNLs. The Crule language is characterized as being P5E3N5S4 indicating a highest score of 5 for precision and naturalness, a score of 4 for simplicity, and a good score of 3 for expressiveness. In comparison, the English language is characterized as P1E5N5S1, propositional logic is classified as P5E1N1S5, Attempto controlled language (ACE) is classified as P4E3N4S3, and SADL is characterized as P5E3N4S4. It may be noted that the Crule language receives higher scores for individual parameters and receives a higher average score compared to other conventional controlled natural languages.
The technical effect of the disclosed systems and methods is that domain knowledge in a knowledge based system is accurately represented in an efficient manner. The disclosed systems and methods enable a subject matter expert to interact with the knowledge based system to author knowledge processing rules without requiring training in descriptive programming languages or assistance from programmers. This feature in a knowledge processing system reduces delay in modifying the knowledge base to include most recently acquired knowledge and insights to process the domain knowledge. Disclosed embodiments of the knowledge management system employ natural language based Crule language to author rules in a natural way. New rules may be added or existing rules in the knowledge management system may be modified in shorter time periods at reduced cost. The concept rules captured are independent of the target executable rule language. Reuse of domain expertise and deployment of domain knowledge across apparently different knowledge systems becomes easier.
It is to be understood that not necessarily all such objects or advantages described above may be achieved in accordance with any particular embodiment. Thus, for example, those skilled in the art will recognize that the systems and techniques described herein may be embodied or carried out in a manner that achieves or improves one advantage or group of advantages as taught herein without necessarily achieving other objects or advantages as may be taught or suggested herein.
While the technology has been described in detail in connection with only a limited number of embodiments, it should be readily understood that the specification is not limited to such disclosed embodiments. Rather, the technology can be modified to incorporate any number of variations, alterations, substitutions or equivalent arrangements not heretofore described, but which are commensurate with the spirit and scope of the claims. Additionally, while various embodiments of the technology have been described, it is to be understood that aspects of the specification may include only some of the described embodiments. Accordingly, the specification is not to be seen as limited by the foregoing description, but is only limited by the scope of the appended claims.
