Academia.eduAcademia.edu

Figure 15 – uploaded by Abolfazl Mir

See full PDFdownloadDownload figure
Figure 9. Zone of influence for the best location of query Q4. Fig. 9 shows a polygon (in orange) that represents the commercial zone that influences the locations found and evaluated as good candidates for query Qy.

Figure 9 Zone of influence for the best location of query Q4. Fig. 9 shows a polygon (in orange) that represents the commercial zone that influences the locations found and evaluated as good candidates for query Qy.

Related Figures (53)

Figure 1. Domain decomposition based parallelisation of the Random Indexing algorithm.
Figure 1. Domain decomposition based parallelisation of the Random Indexing algorithm.
TABLE II. PERFORMANCE CHARACTERISTICS GROUPED BY DATASET AND NUMBER OF COMPUTING NODES. TABLE III. PERFORMANCE CHARACTERISTICS FOR FRAGMENTED DATASETS, THE NUMBER OF FRAGMENTS CORRESPONDS TO THE NUMBER OF COMPUTING NODES.
TABLE II. PERFORMANCE CHARACTERISTICS GROUPED BY DATASET AND NUMBER OF COMPUTING NODES. TABLE III. PERFORMANCE CHARACTERISTICS FOR FRAGMENTED DATASETS, THE NUMBER OF FRAGMENTS CORRESPONDS TO THE NUMBER OF COMPUTING NODES.
Figure 3. Performance results for decomposed datasets.
Figure 3. Performance results for decomposed datasets.
TABLE IV. PERFORMANCE RESULTS FOR PARALLEL AIRHEAD SEARCH ALGORITHMWITH VARYING JVM SETTING ON WIKII DATASET.
TABLE IV. PERFORMANCE RESULTS FOR PARALLEL AIRHEAD SEARCH ALGORITHMWITH VARYING JVM SETTING ON WIKII DATASET.
RISO-TT is a rules-based system and since it was conceived to become extensible and flexible it uses a configuration file which determines the connection from the internal logic to the rules-base. To insert a new standard (or rule) in the rules base, the new standard is simply added to the configuration files. For this change to be realized, it just needs that RISO-TT be run again with a document that contains, in its content, the corresponding expressions. Figure | presents the RISO-TT architecture. Figure 1. RISO-TT architecture.
RISO-TT is a rules-based system and since it was conceived to become extensible and flexible it uses a configuration file which determines the connection from the internal logic to the rules-base. To insert a new standard (or rule) in the rules base, the new standard is simply added to the configuration files. For this change to be realized, it just needs that RISO-TT be run again with a document that contains, in its content, the corresponding expressions. Figure | presents the RISO-TT architecture. Figure 1. RISO-TT architecture.
We notice that the p-value obtained from the Heideltime data characterizes a non-normalized sample. However, evaluating the data, we noticed that they concern three samples which, in the document marking process, had not a good result. The format in question is for dates with three characters (e.g., 200 AD). This format was not recognized by Heideltime and, so, due to the non- Right after tests, the first task carried out was the evaluation of normality of the resulting data. For this activity, we ran the Shapiro-Wilk test, obtaining the results displayed in Table I.
We notice that the p-value obtained from the Heideltime data characterizes a non-normalized sample. However, evaluating the data, we noticed that they concern three samples which, in the document marking process, had not a good result. The format in question is for dates with three characters (e.g., 200 AD). This format was not recognized by Heideltime and, so, due to the non- Right after tests, the first task carried out was the evaluation of normality of the resulting data. For this activity, we ran the Shapiro-Wilk test, obtaining the results displayed in Table I.
Based on the BoxPlot presented in Figure 2, it is possible to conclude that we cannot state whether there is superiority of one of the tools Heideltime and SUTime, since the trust intervals intersect each other. This occurs maybe due to the fact that both tools use the same marking standard and are based only on the knowledge based available in the standard. However, it is possible to state that there is a statistically proved superiority of the results of RISO-TT, compared to the others.
Based on the BoxPlot presented in Figure 2, it is possible to conclude that we cannot state whether there is superiority of one of the tools Heideltime and SUTime, since the trust intervals intersect each other. This occurs maybe due to the fact that both tools use the same marking standard and are based only on the knowledge based available in the standard. However, it is possible to state that there is a statistically proved superiority of the results of RISO-TT, compared to the others.
ABLE III. ACCURACY, COVERAGE AND F-MEASURE RESULTS
ABLE III. ACCURACY, COVERAGE AND F-MEASURE RESULTS
Fig. 1. Ecosystem for Statistics-Driven Ontology Modeling Support The approach chosen aims at being generic, in order to cater the needs of different use cases that expose a strong need for dynamic adaptation and the demand for a deep analysis of interactions with the models. Figure | depicts the influences relevant for the considerations made in this approach:
Fig. 1. Ecosystem for Statistics-Driven Ontology Modeling Support The approach chosen aims at being generic, in order to cater the needs of different use cases that expose a strong need for dynamic adaptation and the demand for a deep analysis of interactions with the models. Figure | depicts the influences relevant for the considerations made in this approach:
Fig. 3. Relations between different wildcard concepts and their temporal likelihood to occur before/after each other Figure 3 shows how temporal relationships are discovered levaring the depicted Markov chain.
Fig. 3. Relations between different wildcard concepts and their temporal likelihood to occur before/after each other Figure 3 shows how temporal relationships are discovered levaring the depicted Markov chain.
Figure 1. Taxonomy of Ontology. Moreover, we identify properties and relationships of each one of the concepts. Then, we obtain the taxonomy and classes shown in Fig. 1.
Figure 1. Taxonomy of Ontology. Moreover, we identify properties and relationships of each one of the concepts. Then, we obtain the taxonomy and classes shown in Fig. 1.
The query is received and the processing starts: the que is analyzed and contextualized through the exploration of t ontology. Next, the rules of the context are applied (market analysis constraints). This results in the parameters that define a geographic query in market domain. Then, t ontology is explored to find a matching between query elements and the concepts stored into ontology. Each element of a query is searched within the ontology (e.g., label names). When a concept is found into the ontology, then its context is extracted. A context is formed by t ry he ne by ne neighboring concepts of a matched concept and its semantic relations are extracted and stored into a vector. The process works according to the following algorithm depicted pseudo code (see Fig. 2): in Figure 2. Ontology exploration algorithm.
The query is received and the processing starts: the que is analyzed and contextualized through the exploration of t ontology. Next, the rules of the context are applied (market analysis constraints). This results in the parameters that define a geographic query in market domain. Then, t ontology is explored to find a matching between query elements and the concepts stored into ontology. Each element of a query is searched within the ontology (e.g., label names). When a concept is found into the ontology, then its context is extracted. A context is formed by t ry he ne by ne neighboring concepts of a matched concept and its semantic relations are extracted and stored into a vector. The process works according to the following algorithm depicted pseudo code (see Fig. 2): in Figure 2. Ontology exploration algorithm.
TABLE II. SEMANTIC VALUES FOR MARKET PARAMETERS In Table 2, the possible values or range of values for each parameter is showed. For example, when a business requires a form of transportation, the evaluation is done in two ways: fast, in terms of time and distance, considering the traffic during the transportation. Table 2 shows the possible semantic values for some parameters in a market context.
TABLE II. SEMANTIC VALUES FOR MARKET PARAMETERS In Table 2, the possible values or range of values for each parameter is showed. For example, when a business requires a form of transportation, the evaluation is done in two ways: fast, in terms of time and distance, considering the traffic during the transportation. Table 2 shows the possible semantic values for some parameters in a market context.
Qi, = “find a business location in Mexico City with market competition in a radius of 10 km located in corner, or into the street, rent or buy". The result is the list of properties that meet the criteria, one of which is shown in Fig. 4. Now, if the query Q; is modified on the relevance of the parameters: parking and roads near. Then, the new Q;, query is generated. The result is now a new list of locations.
Qi, = “find a business location in Mexico City with market competition in a radius of 10 km located in corner, or into the street, rent or buy". The result is the list of properties that meet the criteria, one of which is shown in Fig. 4. Now, if the query Q; is modified on the relevance of the parameters: parking and roads near. Then, the new Q;, query is generated. The result is now a new list of locations.
Figure 3. Input parameters (interface Web system).
Figure 3. Input parameters (interface Web system).
Figure 6. Results for new generated query Qua.
Figure 6. Results for new generated query Qua.
Figure 7. Best commercial locations for query Q3. Figure 5. The best location for commercial business from Qi.
Figure 7. Best commercial locations for query Q3. Figure 5. The best location for commercial business from Qi.
Figure 8. Rated building or property for query Quy. Another test conducted for the query Q,: "Find location for a stationer’ shop, with a low floating population, located to 3 km from a zone influenced commercially." The results obtained show 10 possible locations, the most relevant being located in: Calle 15 No 100. This is shown in Fig. 8.
Figure 8. Rated building or property for query Quy. Another test conducted for the query Q,: "Find location for a stationer’ shop, with a low floating population, located to 3 km from a zone influenced commercially." The results obtained show 10 possible locations, the most relevant being located in: Calle 15 No 100. This is shown in Fig. 8.
Figure 1. Schema of Social Communication [24] (German) Figure 1 shows Aufermann’s [24] model for social communication in which two parties are involved. The sender encodes the statement he intends to submit in a message. Therefore, he uses his own character set to encode the message. The message is sent via a medium to the recipient whereby spatial and temporal distance is overcome. When receiving the message the recipient will use his own character set for the decoding of the message.
Figure 1. Schema of Social Communication [24] (German) Figure 1 shows Aufermann’s [24] model for social communication in which two parties are involved. The sender encodes the statement he intends to submit in a message. Therefore, he uses his own character set to encode the message. The message is sent via a medium to the recipient whereby spatial and temporal distance is overcome. When receiving the message the recipient will use his own character set for the decoding of the message.
Figure 2. Knowledge Transfer through four layers The four layers that influence the transfer of a message from one person to another are code, syntactic, semantic and pragmatic layer. The concept of a knowledge transfer through different layers was influenced by the OSI Reference Model [27]. Figure 2 illustrates the transfer of a message from the sender to the receiver passing through the four layers. At the lowest level of the layer for transfer is the code. The code consists of symbols or signs that represent the smallest unit, which forms the basis of the higher layers. In the case of written language, which is the focus here, the smallest elements are the characters, o, taken from an alphabet X. In the case of spoken language it would be phonemes, or in sign language gestures.
Figure 2. Knowledge Transfer through four layers The four layers that influence the transfer of a message from one person to another are code, syntactic, semantic and pragmatic layer. The concept of a knowledge transfer through different layers was influenced by the OSI Reference Model [27]. Figure 2 illustrates the transfer of a message from the sender to the receiver passing through the four layers. At the lowest level of the layer for transfer is the code. The code consists of symbols or signs that represent the smallest unit, which forms the basis of the higher layers. In the case of written language, which is the focus here, the smallest elements are the characters, o, taken from an alphabet X. In the case of spoken language it would be phonemes, or in sign language gestures.
Knowledge about the receiver is an important requirement for a successful and lossless transfer of a message. The better the sender knows the receiver, the easier he can encode the message. A proper encoding of t message can be done by using elements that exist identica in the personal knowledge base of the sender as well as in t personal knowledge base of the receiver. If the receiver unknown, only assumptions can be made to support t selection. The other way around it is easier for the receiver decode the message if he knows the sender of the message very well. Figure 3 visualizes the overlapping of t knowledge in different layers. ne ly ne iS ne to ne
Knowledge about the receiver is an important requirement for a successful and lossless transfer of a message. The better the sender knows the receiver, the easier he can encode the message. A proper encoding of t message can be done by using elements that exist identica in the personal knowledge base of the sender as well as in t personal knowledge base of the receiver. If the receiver unknown, only assumptions can be made to support t selection. The other way around it is easier for the receiver decode the message if he knows the sender of the message very well. Figure 3 visualizes the overlapping of t knowledge in different layers. ne ly ne iS ne to ne
Besides a few pre-defined primitive types UML allows the definition of additional datatypes in class diagrams. These can be primitive types, complex datatypes, and enumerations. In UML, datatypes—similar to classes—can have owned at- tributes (as well as operations which are not discussed here). Therefore, they can be used to describe structure. Figure 1 shows examples for the three kind of datatypes. Fig. 1. Examples for datatypes in UML. Left: user-defined datatype witt two components. Center: user-defined primitive datatype. Right: Enumeratior with three allowed values.
Besides a few pre-defined primitive types UML allows the definition of additional datatypes in class diagrams. These can be primitive types, complex datatypes, and enumerations. In UML, datatypes—similar to classes—can have owned at- tributes (as well as operations which are not discussed here). Therefore, they can be used to describe structure. Figure 1 shows examples for the three kind of datatypes. Fig. 1. Examples for datatypes in UML. Left: user-defined datatype witt two components. Center: user-defined primitive datatype. Right: Enumeratior with three allowed values.
Fig. 2. Extract from the UML meta model, showing the difference between classes and datatypes. In contrast to instances of classes “any instances of that data type with the same value are considered to be equal instances.’[9, p. 63] Although the graphical representations of datatypes in general (instances of DataType) as well as primitive types (instances of PrimitiveType and enumerations (instances of Enumeration) in particular look similar to the representation of classes (instances of Class) they are different elements of the meta model as shown in Figure 2.
Fig. 2. Extract from the UML meta model, showing the difference between classes and datatypes. In contrast to instances of classes “any instances of that data type with the same value are considered to be equal instances.’[9, p. 63] Although the graphical representations of datatypes in general (instances of DataType) as well as primitive types (instances of PrimitiveType and enumerations (instances of Enumeration) in particular look similar to the representation of classes (instances of Class) they are different elements of the meta model as shown in Figure 2.
Fig. 3. Extract relevant for datatypes from the OWL-2 meta model. An OWL-2 datatype is defined by assigning an Inter- nationalized Resource Identifier (RI) to a DataRange us- ing a DatatypeDefinition axiom. According to the OWL-2 DL specification this IRI must have been declared as the name of a datatype.
Fig. 3. Extract relevant for datatypes from the OWL-2 meta model. An OWL-2 datatype is defined by assigning an Inter- nationalized Resource Identifier (RI) to a DataRange us- ing a DatatypeDefinition axiom. According to the OWL-2 DL specification this IRI must have been declared as the name of a datatype.
Fig. 6. Example for the transformation of a generalization relation between datatypes. no internal strcuture (e.g., enumerations) a transformation is possible: While the generalization of UML classes can be transformed into an OWL-2 ObjectUnionOf class expres- sion this is not possible for datatypes. As the name suggest, an ObjectUnionOf can only be used for classes. Instead an instance of DataUnionOf is used. The sub-datatypes combined in the DataUnionOf constitute a new data range. Using a DatatypeDefinition axiom a name is assigned to this set of datatypes. This name is the name of the super- datatype from UML. Figure 6 shows an example for such a transformation.
Fig. 6. Example for the transformation of a generalization relation between datatypes. no internal strcuture (e.g., enumerations) a transformation is possible: While the generalization of UML classes can be transformed into an OWL-2 ObjectUnionOf class expres- sion this is not possible for datatypes. As the name suggest, an ObjectUnionOf can only be used for classes. Instead an instance of DataUnionOf is used. The sub-datatypes combined in the DataUnionOf constitute a new data range. Using a DatatypeDefinition axiom a name is assigned to this set of datatypes. This name is the name of the super- datatype from UML. Figure 6 shows an example for such a transformation.
Fig. 7. Example of a QVT-Relations rule from the OWL-2 — UML transformation that maps OWL-2 Literals to UML EnumerationLiterals.
Fig. 7. Example of a QVT-Relations rule from the OWL-2 — UML transformation that maps OWL-2 Literals to UML EnumerationLiterals.
Many methodologies identify the domain expert as a crucial participant in the ontology engineering process. For example, Methodology 6 discusses that “... an interorganizational ontology needs to be modeled not by external knowledge engineers, but by domain experts themselves. Only they have the tacit knowledge about the domain and can sufficiently assess the real impact of the conceptualizations and derived collaborative services on their organization. ...” However, it is interesting to observe that only Methodology 3, Methodology 4, and Methodology 6 offer explicit support for domain experts to model the respective ontologies, or at least parts of it, autonomously.
Many methodologies identify the domain expert as a crucial participant in the ontology engineering process. For example, Methodology 6 discusses that “... an interorganizational ontology needs to be modeled not by external knowledge engineers, but by domain experts themselves. Only they have the tacit knowledge about the domain and can sufficiently assess the real impact of the conceptualizations and derived collaborative services on their organization. ...” However, it is interesting to observe that only Methodology 3, Methodology 4, and Methodology 6 offer explicit support for domain experts to model the respective ontologies, or at least parts of it, autonomously.
Fig. 1: Profile Methodology and structures that are defined inside M1 level: Profile, FOAF definition, Location and Food ontologies. Profile ontology uses MetaProfile’s ontological definitions. Metaprofile is indepen- dent from the domain. So, it consists of the basic properties to represent a social profile and its ancestors, a behavioral profile and its properties, and demographic properties of a person. These representations need to be designed inside a person’s profile. Thus, we aim to define indicators. A profile indicator, pi , is the key property that defines a profile. p € P, pi € PI, dy € D, on € O;
Fig. 1: Profile Methodology and structures that are defined inside M1 level: Profile, FOAF definition, Location and Food ontologies. Profile ontology uses MetaProfile’s ontological definitions. Metaprofile is indepen- dent from the domain. So, it consists of the basic properties to represent a social profile and its ancestors, a behavioral profile and its properties, and demographic properties of a person. These representations need to be designed inside a person’s profile. Thus, we aim to define indicators. A profile indicator, pi , is the key property that defines a profile. p € P, pi € PI, dy € D, on € O;
Fig. 2: Policy components of the model
Fig. 2: Policy components of the model
Fig. 3: Class hierarchy of the food domain ontology
Fig. 3: Class hierarchy of the food domain ontology
Fig. 4: Object and Data type properties of the food ontology connection between a place and this place’s food menu. For this purpose, we selected the schema.org’s [10] ontology and adapted this ontology to our case study. Schema.org’s ontology has a property to describe a menu item, but it is a genera definition, which ranges to a String or the Thing class. Furthermore, a connection between the menu and the food domain ontology is a necessity. The relationship between the location ontology and the food domain ontology can be seen in Figure 6. This connection gives us the opportunity to build a profile-based policy description to handle the problems in our case study examples.
Fig. 4: Object and Data type properties of the food ontology connection between a place and this place’s food menu. For this purpose, we selected the schema.org’s [10] ontology and adapted this ontology to our case study. Schema.org’s ontology has a property to describe a menu item, but it is a genera definition, which ranges to a String or the Thing class. Furthermore, a connection between the menu and the food domain ontology is a necessity. The relationship between the location ontology and the food domain ontology can be seen in Figure 6. This connection gives us the opportunity to build a profile-based policy description to handle the problems in our case study examples.
Fig. 6: Location and Food Ontology Relations Fig. 5: Lasagna individual of MealCourse class that we described in Section 2.
Fig. 6: Location and Food Ontology Relations Fig. 5: Lasagna individual of MealCourse class that we described in Section 2.
Fig. 9: General view of the architecture
Fig. 9: General view of the architecture
Fig. 10: Mobile application example
Fig. 10: Mobile application example
By defining ontologies based on a particular domain [26], the algorithm [27][28] within the Intelligence Gathering tool will facilitate the return of intelligence in a structured manner and only for information predefined within our ontologies for BI. Figure 1 presents a breakdown of the thinking required behind ontology design and will form the foundations for developing our BI ontologies. Fig. 1. Visualisation of layers for Ontology Creation [2][29]
By defining ontologies based on a particular domain [26], the algorithm [27][28] within the Intelligence Gathering tool will facilitate the return of intelligence in a structured manner and only for information predefined within our ontologies for BI. Figure 1 presents a breakdown of the thinking required behind ontology design and will form the foundations for developing our BI ontologies. Fig. 1. Visualisation of layers for Ontology Creation [2][29]
Fig. 2. IGUSMON Project System Architecture [22] The research is in its infancy and the following architecture and design described is the conceptualisation of our algorithm for intelligence gathering using semantic metadata. Figure 2 illustrates the conceptual design of the Information Gathering tool, which demonstrates how the web spiders will act as a mechanism for gathering the raw data, before sending the extracted semantic metadata back to the database for validation with the predefined ontologies. Once the extracted data is verified, a data-mining [34] algorithm structures the data into information before returning it as a knowledgeable answer to the Query Management System.
Fig. 2. IGUSMON Project System Architecture [22] The research is in its infancy and the following architecture and design described is the conceptualisation of our algorithm for intelligence gathering using semantic metadata. Figure 2 illustrates the conceptual design of the Information Gathering tool, which demonstrates how the web spiders will act as a mechanism for gathering the raw data, before sending the extracted semantic metadata back to the database for validation with the predefined ontologies. Once the extracted data is verified, a data-mining [34] algorithm structures the data into information before returning it as a knowledgeable answer to the Query Management System.
Fig. 3. Implementation of Swarming Intelligence Behaviour with the IGUSMON Project System Architecture signal the release of the web spiders from the spider deployment module via the database and a swarming agent. Collected information will be verified for relevant inte ligence within the Validation Module via ontology checks. However, before the semantic metadata reaches the Vali wor dation Module, a final check will be conducted via a ker agent against the Irrelevant Data module, where discarded information from previous extractions, that did not prod uce positive intelligence results relating to a query, are stored.
Fig. 3. Implementation of Swarming Intelligence Behaviour with the IGUSMON Project System Architecture signal the release of the web spiders from the spider deployment module via the database and a swarming agent. Collected information will be verified for relevant inte ligence within the Validation Module via ontology checks. However, before the semantic metadata reaches the Vali wor dation Module, a final check will be conducted via a ker agent against the Irrelevant Data module, where discarded information from previous extractions, that did not prod uce positive intelligence results relating to a query, are stored.
benelits of this focus are anticipated to ensure vast datasets can be quickly referenced and utilised for extraction. The direct integration of the semantic metadata with the ontologies will ensure that relevant knowledge can be extracted. An obvious limitation to this method will be determined by how much of the relevant data is attributed with semantic metadata. Even though semantic web methods have been proposed for over a decade now, data does exist that was created before and after, which seldom or minimally focuses on semantics. However this does not mean that semantic metadata is limited; with the technological growth and vast amounts of growing data, this limitation is becoming less finite. The other limitation that will impact our research will be the reach of the algorithm. When the conceptualisation of the algorithm is developed, the testing will focus on a finite number of websites for extraction, due to available computing power and time constraints. As mentioned, the focus of the IGUSMON project is currently immersed within this area and development is in progression; some of the design elements proposed may change as the modules are created and tested for feasibility. Fig. 4. Mathematical Theory of Communication [5][22]
benelits of this focus are anticipated to ensure vast datasets can be quickly referenced and utilised for extraction. The direct integration of the semantic metadata with the ontologies will ensure that relevant knowledge can be extracted. An obvious limitation to this method will be determined by how much of the relevant data is attributed with semantic metadata. Even though semantic web methods have been proposed for over a decade now, data does exist that was created before and after, which seldom or minimally focuses on semantics. However this does not mean that semantic metadata is limited; with the technological growth and vast amounts of growing data, this limitation is becoming less finite. The other limitation that will impact our research will be the reach of the algorithm. When the conceptualisation of the algorithm is developed, the testing will focus on a finite number of websites for extraction, due to available computing power and time constraints. As mentioned, the focus of the IGUSMON project is currently immersed within this area and development is in progression; some of the design elements proposed may change as the modules are created and tested for feasibility. Fig. 4. Mathematical Theory of Communication [5][22]
RESULTS OF THE SECOND EXPERIMENTATION
RESULTS OF THE SECOND EXPERIMENTATION
RESULTS OF THE FIRST EXPERIMENTATION
RESULTS OF THE FIRST EXPERIMENTATION
COMPARISON OF RESULTS FOR THE POLARITY
COMPARISON OF RESULTS FOR THE POLARITY
favie Vit EXPERIMENTS ON THE POLARITY WITHOUT CLASSIFICATION OF TOPICS
favie Vit EXPERIMENTS ON THE POLARITY WITHOUT CLASSIFICATION OF TOPICS
One reason why a significant improvement is obtained in these classes may be due to the number of topics in the training data sets. Topics number of class "Issue" and class "Person" are, respectively, 22 and 21. [HE RESULT OF FEW TOPIC (TREC) FOR POSITIVE WORDS
One reason why a significant improvement is obtained in these classes may be due to the number of topics in the training data sets. Topics number of class "Issue" and class "Person" are, respectively, 22 and 21. [HE RESULT OF FEW TOPIC (TREC) FOR POSITIVE WORDS
Figure 1. The result of Positive MAP in the various topics of TREC, using the two approaches: with categorization "MAP-AC" and without categorization "MAP-SC".
Figure 1. The result of Positive MAP in the various topics of TREC, using the two approaches: with categorization "MAP-AC" and without categorization "MAP-SC".
Figure 2. The result of Negative MAP in the various topics of TREC, using the two approaches: with categorization "MAP-AC" and without categorization "MAP-SC".
Figure 2. The result of Negative MAP in the various topics of TREC, using the two approaches: with categorization "MAP-AC" and without categorization "MAP-SC".
Figure 1. Pay-As-You-Go Framework for OWL ontology management in the cloud
Figure 1. Pay-As-You-Go Framework for OWL ontology management in the cloud
Figure 2. Pay-As-You-Go Architecture of the proposed system
Figure 2. Pay-As-You-Go Architecture of the proposed system
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved