Plane Thermoelastic Waves in Infinite Half-Space Caused FACTA UNIVERSITATIS Series: Mechanical Engineering Vol. 11, N o 2, 2013, pp. 181 - 192 INTERPRETING THE MEANING OF GEOMETRIC FEATURES BASED ON THE SIMILARITIES BETWEEN ASSOCIATIONS OF SEMANTIC NETWORK  UDC 004.8 Milan Trifunović, Milos Stojković, Miroslav Trajanović, Dragan Mišić, Miodrag Manić University of Niš, Faculty of Mechanical Engineering, Niš, Serbia Abstract In this paper the method of semantic network analysis which enables semantic network data categorization is presented. The main goal of the network analysis is determination of the similarity of the semantic network associations based on their attribute values. The method allows highly efficient semantic categorization of new concepts, which does not depend on pre-planned inputs and predefined inference rules. Also, the method allows different semantic interpretations of the same concept in different semantic contexts. As a specific example for demonstration of the semantic categorization process, a part of mold manufacturing workflow is chosen. Key Words: Artificial Intelligence, Semantic Features, Cognitive Data Processing, Active Semantic Model, Manufacturing 1. INTRODUCTION The semantic categorization as a term from cognitive psychology is used to express the process of assigning meaning to a (new) item [1, 2]. With the semantic data models, this process is usually performed by setting up a semantic relation between the concepts (nodes) in the semantic network. Semantic categorization represents one of the biggest challenges faced by modern information technologies. Computer networking in the Inter- net has enabled an instant access to huge amounts of data and information. However, it is often rather difficult to find appropriate information, namely the one for which the user  Received December 3, 2013 Corresponding author: Milan Trifunović Faculty of Mechanical Engineering, A. Medvedeva 14, 18000 Niš, Serbia E-mail: milant@masfak.ni.ac.rs Acknowledgements. This paper is part of the project "The research of modern non-conventional technologies application in manufacturing companies with the aim of increase efficiency of use, product quality, reduce of costs and save energy and materials" (project id TR35034), funded by the Ministry of Education, Science and Technological Development of Republic of Serbia. 182 M. TRIFUNOVIĆ, M. STOJKOVIĆ, M. TRAJANOVIĆ, D. MIŠIĆ, M. MANIĆ searches. The information access quality does not equally follow increases of available information [3]. Research in cognitive psychology suggest that the process of searching information contents should include the search of meaning (or according to the meaning) of what is required in order to conduct it in a significantly more precise and even faster way [4, 5]. Following these recommendations, the current research trend in the field of information technologies includes two main subdirections which complement each other: a) research and creation of semantic data models and b) research and modeling of cogni- tive processes. A large piece of knowledge is already embedded into modern Computer Aided Prod- uct Development (CAPD) systems. Actually, parts of mathematics, physics, chemistry, and other fundamental sciences are turned into algorithms as well as programming code of these systems. On the other side, tacit knowledge [6] with its origin in experience and intuition as well as its liability to permanent updating and changes through a direct touch with the product development process, is often very difficult to turn into strongly defined algorithms. This kind of knowledge has stayed, more or less, out of CAPD systems. Nev- ertheless, it plays a crucial role in making important assessments, judgments, inferences and decisions. The tacit knowledge products have a huge impact on the strategic product development process features. Exploration of the methods for embedding and using tacit knowledge is an important and unavoidable evolutionary step forward in the CAPD sys- tems development and, certainly, one of the greatest challenges in the field of IT applica- tion in the product development process. In this paper an approach is described where a new semantic model named Active Semantic Model (ASM) is used for modeling and semantic interpretation of geometry elements in a CAD model. The ASM structure and the method are explained and demon- strated in a given case. The focus is on the method description. 2. ACTIVE SEMANTIC MODEL ASM [7] is a new semantic model which has been developed in-house. Its primary aim is to capture and interpret semantics of the design features related to manufacturability issues [8]. ASM intends to introduce an alternative approach to knowledge representation in comparison with the existing semantic models by moving the focus of data structuring from concepts to semantic relations or associations (the term which is used in ASM). This idea of structuring the meaning in associations is chosen to support the thesis stating that the knowledge that people have about items (visual representations, objects, situations, etc.) is contained in inter-concept associations that abstractly represent those items [9]. Furthermore, ASM has proved itself as a more flexible and productive in capturing and interpreting semantics of data compared to the existing semantic models [7]. Here, we explain an algorithm for determining the similarity of associations as a core process in determining semantic inter-concept similarity. 2.1 Structure The ASM structure consists of: Concepts, Associations between concepts, Concept bodies, and Contexts. Interpreting the Meaning of Geometric Features Based on Similarities between Associations of Semantic Network 183 ASM concepts are at the same time nodes of the ASM semantic network and abstract representations of objects, features, situations, etc. The concept data structure consists of only one parameter – Name. Concept bodies are their realizations. The ASM association structure is characterized by eleven parameters [7]: names (cpti, cptj) of two concepts (or contexts) that are associated (network nodes are part of associa- tion); topological parameters: roles (ri, rj) of concepts (i.e. type, subtype), type (t) of asso- ciating (i.e. classifying), direction (d) of associating (←, ↔, →) and character (c) of as- sociating (+, -); weight parameters: accuracy (h) of associating for given context (0; 0.25; 0.5; 0.75; 1) and significance (s) of associating for given context (0; 0.25; 0.5; 0.75; 1); and affiliation parameters: context id to which an association belongs, and user id to identify who has created the association. There can be only one concept with given name, but there can be many associations belonging to different contexts associating it with other concepts. Type of association states the relationship between the two concepts, i.e. the role of each concept in the association. In this way the values for parameters roles of concepts and type of association are coupled and can be considered as ordered triplets. ASM also introduces contexts which are sets of semantically close associations. The context brings abstract meaning of a certain object, situation or event, and, therefore, it is semantically designated. General context is defined and built into ASM structure inde- pendently of the user. As for others, particular contexts require additional semantic de- scription by the user. All the associations from particular contexts are assigned (usually with different parameters) to the general one. Association plexus in ASM is, in general, a context subset (mathematical structure) and can be considered without specific abstract meaning. This difference between association plexus and the context has no effect on data processing and their use for the data semantic interpretation in ASM. The ASM structure is not domain-specific and can be used for knowledge representa- tion in diverse fields. The knowledge from specific domain should be represented through context(s). Semantic relations between contexts allow knowledge from one context to be applicable to others. 3. INTERPRETING THE SEMANTICS OF DATA IN ASM Cognitive data processing (CDP) algorithms represent a set of data processing proce- dures with which ASM attempts to perform data semantic interpretation. They enable ASM to acquire new knowledge independently and make meaningful decisions and re- sponses. This group of algorithms consists of: 1. the procedure of determining: a. the class of similarity of associations, and, b. the degree of semantic similarity of concepts 2. the algorithm for Determining the Similarity of Associations (DSA) (the core of this algorithm is the procedure for Determining the Parameters of Association (DPA)) 3. the algorithm for Determining the Similarity of Contexts (DSC) 4. the algorithm or the Procedure of Contexts Upgrading (CUP) 5. the algorithm for the creation of heuristics and knowledge "crystallization." 184 M. TRIFUNOVIĆ, M. STOJKOVIĆ, M. TRAJANOVIĆ, D. MIŠIĆ, M. MANIĆ 3.1. Procedure for determining association parameters Determination of semantic similarity of two concepts is usually performed for the con- cepts which are not directly connected by associations (CPTX and CPTN), i.e. in the cases when there exists at least one layer of concepts between CPTX and CPTN known to ASM (layer of "connectional" concepts CPTi (Fig. 1)) through which they are connected. Pairs of associations through which concepts CPTX and CPTN are connected to the same con- nectional concepts CPTi are called common associations (Fig. 1). The process of deter- mining semantic similarity of two concepts results in the creation of association(s) be- tween these two concepts. The parameters (type, direction, character, accuracy, and sig- nificance) of this (these) association(s) are determined through the procedure of deter- mining association parameters. Concepts CPTX and CPTN which are directly connected with one or more associations (Fig. 1) are already semantically categorized. However, determination of semantic similarity for these concepts is possible and needed after every modification of semantic network (after the creation of new associations). Fig. 1 Schema of semantic relations and concepts with illustration of used terminology The association parameters values, and, therefore, the semantic features of a new con- cept, generally depend on the set of parameter values of all the association pairs. The asso- ciations parameters also depend to a large extent on the contexts containing these associa- tions and semantic elements (concepts and contexts) connected by them. In other words, two concepts can be in completely or partially different semantic relations, depending on the context wherein they are observed. The characteristic of each association in terms of its be- longing to a certain context implicitly connects that context with members of this association (concepts or contexts that are connected by this association). This implicit connection be- tween association members and context to which this association belongs is determined by the values of association parameters that are appropriate for given context (Fig. 2). Therefore, it can be stated that the association parameters values, and, consequently, the new concept’s semantic features in the given context, depend on the set of parameter values of all the association pairs that belong to that context. CPTX CPTi CPTN CPTi+3 CPTi+2 CPTi+1 ASM space Common associations Connectional concepts Direct association Known concept New concept Interpreting the Meaning of Geometric Features Based on Similarities between Associations of Semantic Network 185 Fig. 2 Connection between semantic relations (associations) and sets of semantically close associations (contexts) In order to take into account the influence of accuracy and significance of associa- tions, as well as relevance of associations in accordance with context, ASM, in all the cases, determines the values of the association parameters between concepts CPTX and CPTN as follows: 1. ASM determines parameters of association for general context (CTX0): a. ASM separates a subset of accurate and significant associations of general con- text. All association pairs that belong to general context whose product of arithmetic mean values of accuracy and significance is smaller than 0.25 are excluded from further process of determining the association parameters val- ues. b. ASM performs the procedure of determining the values of parameters of association according to the established algorithm (which will be described in detail later in this section). c. ASM creates association between concepts CPTX and CPTN with the deter- mined values of association parameters. 2. ASM determines parameters of association for every particular context (CTXS) which includes associations and concepts connected by them: a. In the first step of refinement for every particular context, ASM takes into ac- count the whole set of associations by which concepts CPTX and CPTN are connected, including associations that are excluded in previous step. ASM ex- cludes pairs of associations in which associations belong to different contexts. Furthermore, ASM excludes a subset of association pairs in which associations belong to the general context, but keeps their variations in particular contexts. At the end of this step, only association pairs in which both associations belong to the same particular context and association pairs in which associations be- long only to general context are kept. b. ASM groups pairs of associations according to specific context. For every sub- set, grouped according to specific context, ASM selects a subset of accurate and significant association pairs. All association pairs that belong to specific context whose product of arithmetic mean values of accuracy and significance is smaller than 0.25 are excluded from further process of determining the asso- ciation parameters values. CPT1 CPT2 | r1 | t | c | s | h | d | r2 | CTX1 CTX2 CTX3 { | r1 | t | c | s | h | d | r2 | }CTX3 { | r1 | t | c | s | h | d | r2 | }CTX1 { | r1 | t | c | s | h | d | r2 | }CTX2 186 M. TRIFUNOVIĆ, M. STOJKOVIĆ, M. TRAJANOVIĆ, D. MIŠIĆ, M. MANIĆ c. ASM performs the procedure of determining the values of parameters of associations according to the established algorithm for every subset of associa- tion pairs. d. ASM creates associations between concepts CPTX and CPTN with determined values of association parameters for every specific context separately. 3.1.1 The procedure of determining the type of association There are three characteristic cases that determine what type will be assigned to the association created during the categorization: 1. ASM determines complete match of all parameters of all association pairs (fifth class of similarity). In this case ASM creates an association of "synonymous" type between these two concepts (CPTX is synonym for CPTN). 2. ASM determines that new concept CPTX and known concept CPTN have associa- tions of at least the same type toward connectional concepts CPTi. In other words, these associations can be of the same or different character, direction, accuracy and significance. Therefore, similarity of at least first class is achieved. In this case ASM creates association of "similarity" type between these two concepts (CPTX is similar to CPTN) (Fig. 3). Fig. 3 Determination of the type of association for the case of the second, third or fourth class of similarity 3. ASM determines that concepts CPTX and CPTN share associations of zero class of similarity with concepts CPTi. Hence, in this case, concepts CPTX and CPTN have associations of different types with the same (connectional) concepts CPTi, i.e. roles of concepts CPTX and CPTN in these associations are different. The type of association that will connect concepts CPTX and CPTN will be determined de- pending on the class of association plexus in which new concept CPTX is found. For example, an easily recognizable class of association plexus is the one that de- CPTX CPTk CPTN End mill Milling cutter Face mill tXk tNk tXN = similarity ASM space Step 1: Description – Association created by user Existing association Step 2: Categorization – Association created by ASM [subtype] [similar concept] [type] [subtype] [type] [similar concept] CPTj Milling [activity] [activity] [argument] [argument] Step 3: Knowledge widening Interpreting the Meaning of Geometric Features Based on Similarities between Associations of Semantic Network 187 scribes some activity (Fig. 4). In this kind of plexus, besides the concept that represents the activity itself, there exists motive for performing that activity, activ- ity subject or subjects (who is performing the activity), activity object or objects (against which the activity is performed), activity argument (with which the activ- ity is performed), product, and result of activity. Therefore, it is sufficient to intro- duce part of association plexus to enable ASM to recognize the class of association plexus for the description of activity. In the next step ASM starts directed search procedure over the semantic network in order do determine the rest of the associa- tions from the plexus. This procedure is intended for pre-planned types of plexuses (association plexuses that describe activity, assembly, etc.). This procedure is based on the CASE algorithm which contains explicitly defined causal mechanism (rule) for every pre-planned CASE. Fig. 4 Association plexus describing the activity "Surface milling" 3.1.2 The procedure of determining accuracy and significance of association The values for accuracy and significance of the association being created between concepts CPTX and CPTN are determined according to the following equations: 1 1      k Xi Nii xn s s s k (1) 1 1      k Xi Nii xn h h h k (2) where k is the number of association pairs formed between concepts CPTX and CPTN and the same concepts CPTi from the semantic network. CPTX CPTk CPTN АSM Oval surface Surface milling tXl tNl tXN = subject-activity ASM space [concept] [subject] [motive] [result] [object] [activity] CPTj Milling [argument] [part] [assembly] [product] CPTl CPTm CPTp CPTq CPTr [attribute] Operation OP10 Surface Identifying NC sequence 188 M. TRIFUNOVIĆ, M. STOJKOVIĆ, M. TRAJANOVIĆ, D. MIŠIĆ, M. MANIĆ Since association accuracy and significance take values from a finite set of standard values (0; 0,25; 0,5; 0,75; 1) for the sake of simplicity, it is necessary (though not required) to adopt values for these parameters that will be different from the calculated values. These values are adopted as follows: Table 1 Adopting the values for accuracy and significance of association Conditional part Result sxn ≥ 0.5 sxn = first bigger standard value sxn < 0.5 sxn = first smaller standard value hxn ≥ 0.5 hxn = first bigger standard value sxn < 0.5 hxn = first smaller standard value sxn = 0 or hxn = 0 Remove Acptx↔cptn 3.1.3 The procedure of determining the character of association Determination of the association character, in the case when there is only one association pair between concepts CPTX and CPTN, is carried out using the following matrix (which is illustrated in Fig. 5): Table 2 Matrix for determining the character of association (only one association pair) cXk cNk cXN + + + +     +  +  Fig. 5 Typical schema of associations during the determination of association character (only one association pair) In the case when there are multiple association pairs (and, consequently, more connectional concepts) between concepts CPTX and CPTN, the function for determination of the character of association which should be created between concepts CPTX and CPTN varies depending on the case: CPTX CPTk CPTN Fillet Sharp edge Оval cXk = + cNk =  cXN =  ASM space Association created by user Existing association Association created by ASM Interpreting the Meaning of Geometric Features Based on Similarities between Associations of Semantic Network 189 1. When there are multiple association pairs between concepts CPTX and CPTN which have the same characters. In this case ASM creates association of positive character between concepts CPTX and CPTN. Concepts CPTX and CPTN associate the same connectional concepts CPTi in the same way – affirmative or nonaffirmative. 2. When there are multiple association pairs between concepts CPTX and CPTN which have different characters. In this case ASM creates association of negative character between concepts CPTX and CPTN. If all association pairs have the same types, then ASM creates the association of negative character and "antonymous" type. 3.1.4 The procedure of determining the direction of association Determination of the association direction in the case when there is only one association pair between concepts CPTX and CPTN is carried out using the following matrix (which is illustrated in Fig. 6): Table 3 Matrix for determining the direction of association (only one association pair) dXk dNk dXN                            Fig. 6 Typical schema of association during the determination of association direction (only one association pair) In the case when there are multiple association pairs (and, consequently, more con- nectional concepts) between concepts CPTX and CPTN the function for determination of the direction of association which should be created between concepts CPTX and CPTN is also defined according to the matrix: CPTX CPTk CPTN Fillet Оval dXk =  dNk =  dXN =  ASM space Association created by user Existing association Аssociation created by ASM Oval surface 190 M. TRIFUNOVIĆ, M. STOJKOVIĆ, M. TRAJANOVIĆ, D. MIŠIĆ, M. MANIĆ Table 4 Matrix for determining the direction of association (multiple association pairs) dX1 d1N dX2 d2N dX3 d3N dXN                 … … … … …   … … … … … 4. CASE A simplified example of the mold manufacturing workflow will be used for the dem- onstration of the approach. Instead of usual mold cooling canals, the tool constructor has planned conformal cooling canals. The usual mold cooling canals are straight, have a cir- cular cross section, and are manufactured by drilling (deep drilling). The conformal cooling canals extend on a spatial curve (trajectory), and usually have a circular cross section. By following the form of lateral ribs, the conformal cooling canals can approach certain parts of the mold and provide a more complete air outlet. However, the conformal cooling canals, because of their shape, can not be manufactured by drilling. Therefore, the regular technological process, which implies manufacturing of cooling canals by drilling, cannot be performed. Thus, an exception related to the mold geometry causes disturbance of manufacturing process workflow, preventing the regular technological process. The scenario that shows the ASM application in this situation is the following: The ASM semantic network already contains a semantic description of the following concepts: opening, hole, canal, cooling canal, straight line, curve, trajectory, circular cross section, drilling, etc. Some concepts are part of the "mold" context, while the others be- long to the general context. During the introduction of the concept "Spatial curve" DPA algorithms will (by using the CASE algorithm presented in Fig. 7) connect concepts "Spatial curve" and "Straight line" with the "subtype-type" association, whose character will be harmonized with the character of the association between concepts "Curve" and "Straight line" (negative character). This means that these two concepts are associated nonaffirmative, i.e. one excludes the other. The next association pair that will be proc- essed by DPA algorithms is "Spatial curve"↔"Straight line"↔"Axis". This pair will in- duce ASM to conclude that "Spatial curve" and "Axis" are similar, but again with nega- tive character. In the last iteration of reasoning ASM will connect concept "Spatial curve" with concept "Drilling" modeled on the association between concepts "Axis" and "Drill- ing" with the association of negative character (Fig. 8). Fig. 7 The schema of decision making for the case with one "similarity" type association CPTX CPTk CPTN ASM space [subtype] [subtype] [type] [similar concept] [similar concept] [type] Interpreting the Meaning of Geometric Features Based on Similarities between Associations of Semantic Network 191 Fig. 8 Part of the ASM semantic network representing geometric and technological features of the mold cooling canal. Green associations are added based on the analysis of blue associations In the following reasoning iterations, negative character of the association between concepts "Spatial curved opening with circular cross section" and "Straight opening with circular cross section" will be reflected, in a similar way, on semantic relation between the concepts "Spatial curved opening with circular cross section" and "Drilling". Finally, negative character of the association between concepts "Spatial curved opening with cir- cular cross section" and "Drilling" will be reflected on semantic relation between the con- cepts "Mold conformal cooling canals" and "Drilling" (Fig. 8). Canal subtype type Opening Mold cooling canal subtype type h=0.5 Hole similar concept similar concept h=0.5 Straight opening with circular cross section subtype type h=1.0 Circular cross section Bottom part assembly Axis concept attribute part assembly c =  Mold cooling opening subtype type attribute Drilling Drill product activity argument h=1.0 h=1.0, s=1.0 h=1.0 h=1.0 Vent synonym synonym h=1.0 h=0.75 Curve Spatial curve Mold conformal cooling canal Spatial curved opening with circular cross section attribute attribute h=1.0 h=1.0 subtype type h=0.75 synonym synonym h=1.0 concept s=1.0 Straight line subtype type c =  similar concept similar concept type subtype c =  c =  c =  attribute concept product activity c =  c =  product concept concept similar concept subtype similar concept c =  product activity type subtype similar concept similar concept c =  concept similar concept 192 M. TRIFUNOVIĆ, M. STOJKOVIĆ, M. TRAJANOVIĆ, D. MIŠIĆ, M. MANIĆ 5. CONCLUSION Active Semantic Model (and its procedures of cognitive data processing) have suc- ceeded in carrying out the first three steps of semantic interpretation autonomously: 1. Introduction – by the initial associating of a new set of data with the data that exist in the semantic network of ASM, 2. Recognition – by determining of similarity of associations between the new and the known elements of the ASM’s semantic network, 3. Categorization – by generating associations between the new and the known ele- ments of the ASM’s semantic network, The developed functionalities of ASM provide for new possibilities for the autono- mous generation of relatively precise and useful assessments and predictions in the be- forehand unpredicted or insufficiently precise defined input data sets. REFERENCES 1. Deutch, J. A., Deutch, D. , 1963, Attention: Some Theoretical Considerations, Psychological Review, 70, 80-90. 2. Norman, D. A., 1968, Toward a theory of memory and attention, Psychological Review, 75, 522-536. 3. Krill P., 2000, Overcoming Information Overload, InfoWorld. 4. Peterson, L.R., Peterson, M.J., Miller, G.A., 1961, Short term retention and meaningfulness, Canadian Journal of Psychology, 15, 143-147. 5. Kostić, A., 2006, Kognitivna psihologija, Zavod za udžbenike i nastavna sredstva, Beograd. 6. Nonaka, I., Takeuchi, H., 1995, The knowledge creating company: how Japanese companies create the dynamics of innovation, Oxford University Press, New York. 7. Stojkovic, M., Manic, M., Trifunovic, M., Misic, D., 2011, Semantic categorization of data by determining the similarities of associations of the semantic network, E-Society Journal Research and Applications, 2, 3-13. 8. Stojkovic, M., 2011, Analysis of the Manufacturability Parameters Based on Semantic Structures of the Digital Product Model, PhD Thesis, University of Nis, Faculty of Mechanical Engineering in Nis, Nis, Serbia. 9. Anderson, J.R., Bower, G. H., 1973, Human Associative Memory, Winston, Washington, DC. INTERPRETACIJA ZNAČENJA GEOMETRIJSKIH ODLIKA BAZIRANA NA SLIČNOSTI ASOCIJACIJA SEMANTIČKE MREŽE U radu je predstavljen koncept analize semantičke mreže koji omogućava semantičku kategorizaciju podataka. Glavni cilj analize je utvrđivanje sličnosti asocijacija semantičke mreže na osnovu sličnosti vrednosti atributa asocijacija. Koncept omogućava efikasnu semantičku kategorizaciju novih pojmova, koja ne zavisi od unapred planiranih ulaza i definisanih pravila zaključivanja. Pristup omogućava i različite semantičke interpretacije istog pojma u različitim semantičkim kontekstima. Kao primer za demonstraciju procesa semantičke kategorizacije je iskorišćen deo radnog toka izrade kalupa. Ključne reči: veštačka inteligencija, semantičke odlike, kognitivna obrada podataka, Aktivni semantički model, izrada