Academia.eduAcademia.edu

Figure 34 – uploaded by Laure Berti-Equille

See full PDFdownloadDownload figure
Figure 3.6: Decision Costs for Rule Selection with Different Data Quality Variations without Misclassification for the a Priori Probability 7° = 0.200

Figure 3 6: Decision Costs for Rule Selection with Different Data Quality Variations without Misclassification for the a Priori Probability 7° = 0.200

Related Figures (67)

Figure 2: Applications and Data Types Coverage Beyond the defense of the “Habilitation a Diriger des Recherches (HDR)”, a wish would be that this dissertation may be useful for one- or two-semester courses on data quality in advanced undergraduate or graduate courses on database manage- ment and data warehouse design that are usually required in Information System, Business IT, or Computer Science curriculum.
Figure 2: Applications and Data Types Coverage Beyond the defense of the “Habilitation a Diriger des Recherches (HDR)”, a wish would be that this dissertation may be useful for one- or two-semester courses on data quality in advanced undergraduate or graduate courses on database manage- ment and data warehouse design that are usually required in Information System, Business IT, or Computer Science curriculum.
Contents
Contents
Figure 1.1: Taxonomy of Existing Techniques for Entity Resolution
Figure 1.1: Taxonomy of Existing Techniques for Entity Resolution
Table 1.3: Decision Models for Handling Duplicates (Batini & Scannapieco, 2006)
Table 1.3: Decision Models for Handling Duplicates (Batini & Scannapieco, 2006)
such that x, < aj; + %;, (triangular inequality) for all i, 7,4 vertices of the graph, «;; € {0,1}: if 7 and j are in the same cluster x;,; is 1; otherwise 2;; is 0. A method for online clustering in (Sarawagi & Kirpal, 2004) avoids the problem of data skew and redundant computation while creating clusters based on similar- ity along multiple strings rather than single strings, and it improves the efficiency when index searching with increasing threshold to return the most similar cluster. Fa eh le ee, ee a i: a eo cs ee i a ee, Ee et on i Se es: ae ee ee. ee
such that x, < aj; + %;, (triangular inequality) for all i, 7,4 vertices of the graph, «;; € {0,1}: if 7 and j are in the same cluster x;,; is 1; otherwise 2;; is 0. A method for online clustering in (Sarawagi & Kirpal, 2004) avoids the problem of data skew and redundant computation while creating clusters based on similar- ity along multiple strings rather than single strings, and it improves the efficiency when index searching with increasing threshold to return the most similar cluster. Fa eh le ee, ee a i: a eo cs ee i a ee, Ee et on i Se es: ae ee ee. ee
Table 1.4: Main Data Transformation Operators for ETL Table 1.4 presents the main ETL operators (e.g., format, add, merge, etc.) sup- ported by Potter’s Wheel and AJAX systems with, in the first column, the name of the operator and, in the second column, its definition.
Table 1.4: Main Data Transformation Operators for ETL Table 1.4 presents the main ETL operators (e.g., format, add, merge, etc.) sup- ported by Potter’s Wheel and AJAX systems with, in the first column, the name of the operator and, in the second column, its definition.
Based on the online study of Galhardas*, Table 1.5 presents the research proto- types and open source systems developed for data cleaning, data preparation and analysis.
Based on the online study of Galhardas*, Table 1.5 presents the research proto- types and open source systems developed for data cleaning, data preparation and analysis.
Contents
Contents
Table 2.2: Join over FK attributes considered as correct Join over FK attributes considered as correct. Ignoring these observations, we as- sume that the data are correct, fresh, consistent, complete and without duplicates, the join of tables PRODUCT and CUSTOMER over the CUST_ID attribute would result in a table consisting of 3 rows, as shown in Table 2.2. For the sake of simplic- ity, only PROD_ID, CUST_IDs, P_DESTINATION and CUST_CITY are projected in the join result of tables PRODUCT and CUSTOMER, respectively renamed as P and C.
Table 2.2: Join over FK attributes considered as correct Join over FK attributes considered as correct. Ignoring these observations, we as- sume that the data are correct, fresh, consistent, complete and without duplicates, the join of tables PRODUCT and CUSTOMER over the CUST_ID attribute would result in a table consisting of 3 rows, as shown in Table 2.2. For the sake of simplic- ity, only PROD_ID, CUST_IDs, P_DESTINATION and CUST_CITY are projected in the join result of tables PRODUCT and CUSTOMER, respectively renamed as P and C.
Join over incorrect FK attributes. If a value of the joining field is incorrect, then, assuming that referential integrity constraints have been enforced, the join would result in a row that, while incorrect, at least should be there. To see this, suppose that in the P3 row of PRODUCT table, the correct value C'2 is replaced by an incor- rect value, say C3. Then, the join operation would generate an incorrect row for a correct one, as shown in Table 2.3. Join over non FK attributes considered as correct. If the joining field is not a foreign key, then the result could contain for each incorrect joining value multiple rows that would not exist. For instance, in Table 2.1, consider joining the two tables over the P_DESTINATION field of PRODUCT table and the CUST_CITY field of CUSTOMER table. The result would yield five rows, as illustrated in Table 2.4.
Join over incorrect FK attributes. If a value of the joining field is incorrect, then, assuming that referential integrity constraints have been enforced, the join would result in a row that, while incorrect, at least should be there. To see this, suppose that in the P3 row of PRODUCT table, the correct value C'2 is replaced by an incor- rect value, say C3. Then, the join operation would generate an incorrect row for a correct one, as shown in Table 2.3. Join over non FK attributes considered as correct. If the joining field is not a foreign key, then the result could contain for each incorrect joining value multiple rows that would not exist. For instance, in Table 2.1, consider joining the two tables over the P_DESTINATION field of PRODUCT table and the CUST_CITY field of CUSTOMER table. The result would yield five rows, as illustrated in Table 2.4.
Join over incorrect non FK attributes. Now assume that the value ‘Reading, NJ, US’ in the P3 row of PRODUCT table is incorrectly replaced by the value ‘Reading UK’. The resulting join would now have six rows as shown in Table 2.5, two o which would be incorrect. Another type of error in that field could lead to multiple missing rows. To see this, consider the case where, in P3 row of the PRODUCT table, the value ‘Reading, NJ, US’ is incorrectly replaced by ‘London’. In this case, the resulting join would four rows instead of five.
Join over incorrect non FK attributes. Now assume that the value ‘Reading, NJ, US’ in the P3 row of PRODUCT table is incorrectly replaced by the value ‘Reading UK’. The resulting join would now have six rows as shown in Table 2.5, two o which would be incorrect. Another type of error in that field could lead to multiple missing rows. To see this, consider the case where, in P3 row of the PRODUCT table, the value ‘Reading, NJ, US’ is incorrectly replaced by ‘London’. In this case, the resulting join would four rows instead of five.
row of PRODUCT table. The correct answer is illustrated in Table 2.6 instead o! the one given in Table 2.2 for P1. Join over non FK attributes of duplicates. The result of joining PRODUCT and CUSTOMER over the non FK attributes P_DESTINATION and the CUST_CITY once the duplicates have been removed in CUSTOMER table would be composed of three rows (instead of five rows) as illustrated in Table 2.7.
row of PRODUCT table. The correct answer is illustrated in Table 2.6 instead o! the one given in Table 2.2 for P1. Join over non FK attributes of duplicates. The result of joining PRODUCT and CUSTOMER over the non FK attributes P_DESTINATION and the CUST_CITY once the duplicates have been removed in CUSTOMER table would be composed of three rows (instead of five rows) as illustrated in Table 2.7.
Figure 2.2: Relational Metamodel of CRM_DB
Figure 2.2: Relational Metamodel of CRM_DB
Modeling the relational resource instances, Figure 2.3 shows how a Rowset in- herits from Extent, from the Foundation package. It represents all the data com- prised in a ColumnSet. A RowSet can only be owned by a ColumnSet or any derived class. A RowSet contains Rows. Row inherits from Object. Its structure is defined by the corresponding ColumnSet and its Columns from the relational metamodel given in Annexes, Figure A.1 and A.2. (pages 152 et 153). Each Row is divided into ColumnValues, which match the value of a relational table, at the intersection of a row and a column. ColumnValue inherits from DataValue from ObjectModel.
Modeling the relational resource instances, Figure 2.3 shows how a Rowset in- herits from Extent, from the Foundation package. It represents all the data com- prised in a ColumnSet. A RowSet can only be owned by a ColumnSet or any derived class. A RowSet contains Rows. Row inherits from Object. Its structure is defined by the corresponding ColumnSet and its Columns from the relational metamodel given in Annexes, Figure A.1 and A.2. (pages 152 et 153). Each Row is divided into ColumnValues, which match the value of a relational table, at the intersection of a row and a column. ColumnValue inherits from DataValue from ObjectModel.
Figure 2.4: QoD Extension to CWM Data Instance Metamodel Quality dimensions are used as the vocabulary to define abstractly various as- pects of quality, as the user may perceive it. Of course, each user might have a different vocabulary, different preferences and goals underlying quality specifica- tions in the various dimensions of quality. In order to aggregate the QoD measures related to one quality dimension, we define the QoDDimension class composed of the associated QoDMeasures. QoD dimension has a name, a scoring method, and a acceptability value. The scoring method is used to compute the acceptability value with possibly weighting the degree of constraint satisfaction of the QoD measures associated to the QoD dimension and compared to expected values. Acceptability of a QoD dimension evaluated from a data object instance is the probability that each value of its associated QoD measures satisfies the specified constraint.
Figure 2.4: QoD Extension to CWM Data Instance Metamodel Quality dimensions are used as the vocabulary to define abstractly various as- pects of quality, as the user may perceive it. Of course, each user might have a different vocabulary, different preferences and goals underlying quality specifica- tions in the various dimensions of quality. In order to aggregate the QoD measures related to one quality dimension, we define the QoDDimension class composed of the associated QoDMeasures. QoD dimension has a name, a scoring method, and a acceptability value. The scoring method is used to compute the acceptability value with possibly weighting the degree of constraint satisfaction of the QoD measures associated to the QoD dimension and compared to expected values. Acceptability of a QoD dimension evaluated from a data object instance is the probability that each value of its associated QoD measures satisfies the specified constraint.
Figure 2.5: Example of QoD Metadata Associated to CRM_DB PRODUCT table Each QoD measure is associated to only one QoD dimension and one database object in- stance. For example, for PRICE column, nullValues% is a QoD measure associated to Completeness QoD dimension. This QoD measure is calculated at a given time (May 15th, 2006) by procedure plsql_id6 and returns the value 75% whereas the expected value for this measure equals 100%.
Figure 2.5: Example of QoD Metadata Associated to CRM_DB PRODUCT table Each QoD measure is associated to only one QoD dimension and one database object in- stance. For example, for PRICE column, nullValues% is a QoD measure associated to Completeness QoD dimension. This QoD measure is calculated at a given time (May 15th, 2006) by procedure plsql_id6 and returns the value 75% whereas the expected value for this measure equals 100%.
Generic examples of Level II Functions we use for characterizing aspects of consistency (CT) and accuracy (AC) dimensions are given in Table 2.9. Table 2.9: Examples of Level IT Functions
Generic examples of Level II Functions we use for characterizing aspects of consistency (CT) and accuracy (AC) dimensions are given in Table 2.9. Table 2.9: Examples of Level IT Functions
The larger the (”)-statistic, the farther away, in the generalized distance sense, the point (a1;, %2;,+++ , @mi) is from the centroid (£1;, £2;,:-- , Zmi) of the reference group. Thus, the tuple may be said to be more deviant from the average member of the group.
The larger the (”)-statistic, the farther away, in the generalized distance sense, the point (a1;, %2;,+++ , @mi) is from the centroid (£1;, £2;,:-- , Zmi) of the reference group. Thus, the tuple may be said to be more deviant from the average member of the group.
Table 2.12 presents the description of the main classification techniques (with the underlined ones we used). Table 2.12: Examples of Level IV Classification and Partitioning Methods
Table 2.12 presents the description of the main classification techniques (with the underlined ones we used). Table 2.12: Examples of Level IV Classification and Partitioning Methods
Figure 2.6: Example of Analytic Workflow for QoD evaluation of CRM_DB
Figure 2.6: Example of Analytic Workflow for QoD evaluation of CRM_DB
Table 2.14: Assigning Probabilities to QoD Dimension for a DB Object Instance
Table 2.14: Assigning Probabilities to QoD Dimension for a DB Object Instance
Table 2.15: Range-Encoded Bitmap Index with Binning for QoD Measures As illustrated in Table 2.15, the values of the QoD measure m,; are given in the second leftmost column. The identifier of the DB object instance which the m;; measure value is associated to is given in the fist column. The range of possible values of mj; is partitioned into five bins [0, 20), [20,40), etc. A “1-bit” indicates that the measure value falls into a specific bin. On the contrary, a “O-bit” indicates that the measure value does not fall into the specific bin.
Table 2.15: Range-Encoded Bitmap Index with Binning for QoD Measures As illustrated in Table 2.15, the values of the QoD measure m,; are given in the second leftmost column. The identifier of the DB object instance which the m;; measure value is associated to is given in the fist column. The range of possible values of mj; is partitioned into five bins [0, 20), [20,40), etc. A “1-bit” indicates that the measure value falls into a specific bin. On the contrary, a “O-bit” indicates that the measure value does not fall into the specific bin.
Figure 2.7: Syntax of Quality Contract Type Creation 2.7. EXTENDING THE SYNTAX OF A QUERY LANGUAGE
Figure 2.7: Syntax of Quality Contract Type Creation 2.7. EXTENDING THE SYNTAX OF A QUERY LANGUAGE
Figure 2.8: Syntax of Call Specification in Quality Contract Type Declaration 2.7. EXTENDING THE SYNTAX OF A QUERY LANGUAGE
Figure 2.8: Syntax of Call Specification in Quality Contract Type Declaration 2.7. EXTENDING THE SYNTAX OF A QUERY LANGUAGE
Table 2.16: Examples of Quality Contract Type Declaration
Table 2.16: Examples of Quality Contract Type Declaration
Checking Freshness on CELLS, ROWS and TABLE of PRODUCT At this stage, when the set of constraints C’ declared in the contracts are checked on the DB object instance 0, a temporary matrix temp is generated from the quality matrix QoD(o) associated to o. Each element of the temporary matrix indicates the degree of satisfaction of the related constraint: 6;; in [0,1].
Checking Freshness on CELLS, ROWS and TABLE of PRODUCT At this stage, when the set of constraints C’ declared in the contracts are checked on the DB object instance 0, a temporary matrix temp is generated from the quality matrix QoD(o) associated to o. Each element of the temporary matrix indicates the degree of satisfaction of the related constraint: 6;; in [0,1].
2.7. EXTENDING THE SYNTAX OF A QUERY LANGUAGE
2.7. EXTENDING THE SYNTAX OF A QUERY LANGUAGE
Table 2.19: Examples of Simple QWITH Queries with EXACT Constraint Checking Mode of the query. Of course, these queries lead to different results depending on the constraints imposed on the QoD dimensions evaluated from database object instances.
Table 2.19: Examples of Simple QWITH Queries with EXACT Constraint Checking Mode of the query. Of course, these queries lead to different results depending on the constraints imposed on the QoD dimensions evaluated from database object instances.
Figure 2.12: Quality-Extended Query Processing
Figure 2.12: Quality-Extended Query Processing
Figure 3.1: General Framework of Data Quality Awareness for the KDD Process metadata on local data sources may be used to drive the data loading process (for example, for loading only the most accurate and up-to-date data in the data ware- housing system). In the second case (post-loading), QoD measures are computed a posteriori from the data that have been already integrated and stored in the system.
Figure 3.1: General Framework of Data Quality Awareness for the KDD Process metadata on local data sources may be used to drive the data loading process (for example, for loading only the most accurate and up-to-date data in the data ware- housing system). In the second case (post-loading), QoD measures are computed a posteriori from the data that have been already integrated and stored in the system.
3.2. DATA QUALITY AWARENESS FOR KDD PROCESS
3.2. DATA QUALITY AWARENESS FOR KDD PROCESS
The average quality of the association rule R denoted ¢(R) can be computed by a weighted sum of the quality vector components of the rule as follows: Definition 3.3.1. The quality of the association rule R is defined by a fusion function denoted ®,; specific for each quality dimension j that merges the components of the quality vectors of the data sets constituting the extension of the right-hand and left-hand sides of the rule. The quality of the rule R is defined as a k-dimensional vector such as:
The average quality of the association rule R denoted ¢(R) can be computed by a weighted sum of the quality vector components of the rule as follows: Definition 3.3.1. The quality of the association rule R is defined by a fusion function denoted ®,; specific for each quality dimension j that merges the components of the quality vectors of the data sets constituting the extension of the right-hand and left-hand sides of the rule. The quality of the rule R is defined as a k-dimensional vector such as:
Based on the main data quality dimensions (e.g., freshness, accuracy, complete ness, and consistency), Table 3.2 presents several examples of fusion function al lowing the combination of quality measures per quality dimension for two dat sets noted x and y in the rule x = y.
Based on the main data quality dimensions (e.g., freshness, accuracy, complete ness, and consistency), Table 3.2 presents several examples of fusion function al lowing the combination of quality measures per quality dimension for two dat sets noted x and y in the rule x = y.
In the case of misclassification these inequalities give rise to three different threshold values 4, p and v (respectively for legitimately, potentially and not interest- ing rules) in the decision space as illustrated in Figure 3.4.
In the case of misclassification these inequalities give rise to three different threshold values 4, p and v (respectively for legitimately, potentially and not interest- ing rules) in the decision space as illustrated in Figure 3.4.
Figure 3.4: Decision Areas for Rule Post-Selection
Figure 3.4: Decision Areas for Rule Post-Selection
Table 3.4: Quality Measures and Estimated Probabilities of Selected Attributes of the KDD-Cup-98 Data Set fac in Table 3.4 (also noted f4c(q(Z)) in our formalism) is the probability den- sity that the data set Z has acceptable quality and correct data when the average quality score of the variable Z is G(Z). fun (also noted fy (qG(Z))) is the probability density that the data set Z has unacceptable quality due to erroneous data when the average quality score of T is q(Z). The top tena priori association rules discovered by (Wang et al., 2005) are given in Table 3.5 with the confidence, the support (in number of records), and the quality scores. Table 3.5 shows the score per quality dimension and the average quality score for each association rule. The scores are computed from the definitions of the quality dimensions given in Table 3.2 and the data quality scores previously given per attribute in Table 3.4.
Table 3.4: Quality Measures and Estimated Probabilities of Selected Attributes of the KDD-Cup-98 Data Set fac in Table 3.4 (also noted f4c(q(Z)) in our formalism) is the probability den- sity that the data set Z has acceptable quality and correct data when the average quality score of the variable Z is G(Z). fun (also noted fy (qG(Z))) is the probability density that the data set Z has unacceptable quality due to erroneous data when the average quality score of T is q(Z). The top tena priori association rules discovered by (Wang et al., 2005) are given in Table 3.5 with the confidence, the support (in number of records), and the quality scores. Table 3.5 shows the score per quality dimension and the average quality score for each association rule. The scores are computed from the definitions of the quality dimensions given in Table 3.2 and the data quality scores previously given per attribute in Table 3.4.
3.4. EXPERIMENTS ON QUALITY-AWARE RULE MINING
3.4. EXPERIMENTS ON QUALITY-AWARE RULE MINING
We observe that only 5 rules (i.e., R1, R5, R7, R9, R10) are potentially interesting among the top ten rules considering the quality of data they are computed from. With data quality-awareness, the other rules (R2, R3, R4, R6, R8) are not interesting despite a good rank in the top ten list. It’s also interesting to notice that the profit per rule predicted by (Wang et al., 2005) may be considerably counterbalanced by the cost of the rule computed from low-quality data (although it depends from initial costs defined in Table 3.3). The second best rule R2 whose predicted profit is $61.73 has a cost of $109.5 and thus is classified as not interesting due to the low quality of its data sets.
We observe that only 5 rules (i.e., R1, R5, R7, R9, R10) are potentially interesting among the top ten rules considering the quality of data they are computed from. With data quality-awareness, the other rules (R2, R3, R4, R6, R8) are not interesting despite a good rank in the top ten list. It’s also interesting to notice that the profit per rule predicted by (Wang et al., 2005) may be considerably counterbalanced by the cost of the rule computed from low-quality data (although it depends from initial costs defined in Table 3.3). The second best rule R2 whose predicted profit is $61.73 has a cost of $109.5 and thus is classified as not interesting due to the low quality of its data sets.
Figure 3.7: Amplitude of Cost Variations Depending on Data Quality Variations without Misclassification for the a priori Probability 7° = 0.200
Figure 3.7: Amplitude of Cost Variations Depending on Data Quality Variations without Misclassification for the a priori Probability 7° = 0.200
Another interesting result is shown in Figure 3.8 where the decisions for rule selection change simultaneously with the data quality variations. Among the top ten interesting rules discovered by (Wang et al., 2005) with the initial data quality (noted Init Qual), 5 rules (R1, R5, R7, R9 and R10) are potentially worth being selected based on their average data quality and 5 rules are not interesting (R2, R3, R4, R6 and R8). While increasing data quality up to +30%, 3 rules become legitimately interesting (R5, R7 and R9).
Another interesting result is shown in Figure 3.8 where the decisions for rule selection change simultaneously with the data quality variations. Among the top ten interesting rules discovered by (Wang et al., 2005) with the initial data quality (noted Init Qual), 5 rules (R1, R5, R7, R9 and R10) are potentially worth being selected based on their average data quality and 5 rules are not interesting (R2, R3, R4, R6 and R8). While increasing data quality up to +30%, 3 rules become legitimately interesting (R5, R7 and R9).
Figure 3.9: Decision Costs for Rule Selection with a Priori Probability in [0.1,0.5] with Misclassification 3.4. EXPERIMENTS ON QUALITY-AWARE RULE MINING While keeping the a priori probability 7° = 0.200 and using Eq. 3.13, 3.12 and 3.14, we compute the values of the three decision thresholds for rule selection with misclassification and we obtain: = 0.1053, p = 0.1667, and v = 4.6667. Based on the values for these thresholds, we can assign the rules to one of the three decision areas (see Table 3.6).
Figure 3.9: Decision Costs for Rule Selection with a Priori Probability in [0.1,0.5] with Misclassification 3.4. EXPERIMENTS ON QUALITY-AWARE RULE MINING While keeping the a priori probability 7° = 0.200 and using Eq. 3.13, 3.12 and 3.14, we compute the values of the three decision thresholds for rule selection with misclassification and we obtain: = 0.1053, p = 0.1667, and v = 4.6667. Based on the values for these thresholds, we can assign the rules to one of the three decision areas (see Table 3.6).
Figure 3.10: Decision Costs for Rule Selection with Different Data Quality Varia- tions with Misclassification for the a Priori Probability 7° = 0.200 Figure 3.10 shows that the costs are relatively stable with smaller amplitudes and more distinct and staggered cost ranges than in the case without misclassi- fication (see Figures 3.10 and 3.11 compared to Figures 3.6 and 3.7) except at the maxima of data quality variations (.e., 4 impact on decision costs. + 50%) when the misclassification has more
Figure 3.10: Decision Costs for Rule Selection with Different Data Quality Varia- tions with Misclassification for the a Priori Probability 7° = 0.200 Figure 3.10 shows that the costs are relatively stable with smaller amplitudes and more distinct and staggered cost ranges than in the case without misclassi- fication (see Figures 3.10 and 3.11 compared to Figures 3.6 and 3.7) except at the maxima of data quality variations (.e., 4 impact on decision costs. + 50%) when the misclassification has more
In Figure 3.12 only R7 is legitimately interesting among the top ten rules discov- ered by (Wang et al., 2005) with the initial data quality (noted Init Qual). Three rules are not interesting (R8, R3 and R4) and the other 6 rules are potentially inter- esting. Misclassification globally attenuates the “verdict” that classifies each rule correspondingly to one of the decision areas for the legitimately, potentially or not interesting rules. Some rules (e.g., R8) keep the same behavior with or without misclassification when data quality varies.
In Figure 3.12 only R7 is legitimately interesting among the top ten rules discov- ered by (Wang et al., 2005) with the initial data quality (noted Init Qual). Three rules are not interesting (R8, R3 and R4) and the other 6 rules are potentially inter- esting. Misclassification globally attenuates the “verdict” that classifies each rule correspondingly to one of the decision areas for the legitimately, potentially or not interesting rules. Some rules (e.g., R8) keep the same behavior with or without misclassification when data quality varies.
Figure 3.12: Decision Status on Rule Selection for Data Quality Variations with Misclassification for 7° = 0.200
Figure 3.12: Decision Status on Rule Selection for Data Quality Variations with Misclassification for 7° = 0.200
Contents
Contents
data domain, i.e., gene expression measurements through several physiopatholog- ical conditions, ti) genomic data domain, i.e., DNA gene, mRNA, protein sequences and their annotations, and iii) biomedical knowledge, i.e., biological and medical concepts that annotate the genes with concepts of ontologies like GO’ and UMLS”, using BioMeKE"', the application developed at INSERM U522 by Marquet et al. (2003). ; °Gene Ontology, GO: http://www.geneontology.org /
data domain, i.e., gene expression measurements through several physiopatholog- ical conditions, ti) genomic data domain, i.e., DNA gene, mRNA, protein sequences and their annotations, and iii) biomedical knowledge, i.e., biological and medical concepts that annotate the genes with concepts of ontologies like GO’ and UMLS”, using BioMeKE"', the application developed at INSERM U522 by Marquet et al. (2003). ; °Gene Ontology, GO: http://www.geneontology.org /
Figure 4.2: Mediation Architecture with XQuaL the router also broadcasts the QWITH statement with the regular part of the query to all available wrappers, it waits for their response. Finally, the one: of selected wrappers that maximize the satisfaction degrre of quality con straints specified in the contract will be chosen to execute the SQL statement Assume that the constraints are too strict and no wrapper can satisfy them in this case the relaxation module will renegotiate the contract terms witl selected wrappers; it will actually reduce the strictnes of the constraints de fined in the contract of the QWITH query and broadcasts the modified par of the contract to the wrappers able to answer. The wrappers that conforn the most to the modified contract will be selected to answer the query.
Figure 4.2: Mediation Architecture with XQuaL the router also broadcasts the QWITH statement with the regular part of the query to all available wrappers, it waits for their response. Finally, the one: of selected wrappers that maximize the satisfaction degrre of quality con straints specified in the contract will be chosen to execute the SQL statement Assume that the constraints are too strict and no wrapper can satisfy them in this case the relaxation module will renegotiate the contract terms witl selected wrappers; it will actually reduce the strictnes of the constraints de fined in the contract of the QWITH query and broadcasts the modified par of the contract to the wrappers able to answer. The wrappers that conforn the most to the modified contract will be selected to answer the query.
Table A.3: PRICE Outlier Detection in PRODUCT table (PL/SQL)
Table A.3: PRICE Outlier Detection in PRODUCT table (PL/SQL)
Related topics:
Information SystemsInformation TechnologyData Quality (Computer Science)Database Management Systems
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved