1. A reference is made to the applicants' earlier Indian patent application titled “System and Method for an Influence based Structural Analysis of a University” with the application number 1269/CHE2010 filed on 6 May 2010.
2. A reference is made to another of the applicants' earlier Indian patent application titled “System and Method for Constructing a University Model Graph” with an application number 1809/CHE/2010 and filing date of 28 Jun., 2010.
3. A reference is made to yet another of the applicants' earlier Indian patent application titled “System and Method for University Model Graph based Visualization” with the application number 1848/CHE/2010 dated 30 Jun. 2010.
The present invention relates to the analysis of the information about a university in general, and more particularly, the analysis of the university based on the structural representations. Still more particularly, the present invention relates to a system and method for what-if analysis based on a model graph associated with the university.
A what-if analysis is typically a hypothetical analysis in which the parameters of a system being analyzed are hypothetically changed so as to determine the new system behavior. Such an analysis helps in determining what happens if the system parameters change in a particular manner. Again, typically, this is done in a simulated environment that uses a model of the system being what-if analyzed. What-if analysis is common and has been used to obtain practical insights in many domains: financial, industrial, process, and business domains to name just a few.
An Educational Institution (EI) (also referred as University) comprises of a variety of entities: students, faculty members, departments, divisions, labs, libraries, special interest groups, etc.
University portals provide information about the universities and act as a window to the external world. A typical portal of a university provides information related to (a) Goals, Objectives, Historical Information, and Significant Milestones, of the university; (b) Profile of the Labs, Departments, and Divisions; (c) Profile of the Faculty Members; (d) Significant Achievements; (e) Admission Procedures; (f) Information for Students; (g) Library; (h) On- and Off-Campus Facilities; (i) Research; (j) External Collaborations; (k) Information for Collaborators; (I) News and Events; (m) Alumni; and (n) Information Resources. The educational institutions are positioned in a very competitive environment and it is a constant endeavor of the management of the educational institution to ensure to be ahead of the competition. This calls for a critical analysis of the overall functioning of the university and help suggest improvements so as enhance the overall strength aspects and overcome the weaknesses. Consider as a typical scenario involving an allocation of funds to the various laboratories of the institution: it makes sense to allocate funds to those labs that provide opportunities for more faculty members to undertake their research work; this in turn would involve more students as research assistants; this double headed improvement leads to the overall enhanced assessment of the institution. Similarly, consider a scenario of enhancing the overall assessment of a faculty member: in this case, encouraging the faculty member to attend a technical conference and present their work would help enhance the influencing factors with respect to both peer faculty members and students. These illustrative scenarios call for what-if analysis based on a model of the institution to obtain better and practical insights into the institution.
U.S. Pat. No. 7,606,165 to Qiu; Lili (Bellevue, Wash.), Bahl; Paramvir (Sammamish, Wash.), Zhou; Lidong (Sunnyvale, Calif.), Rao; Ananth Rajagopala (El Cerrito, Calif.) for “What-if analysis for network diagnostics” (issued on Oct. 20, 2009 and assigned to Microsoft Corporation (Redmond, Wash.)) describes a network troubleshooting framework for performing what-if analysis of wired and wireless networks.
United States Patent Application 20100198958 titled “Real-Time Feedback for Policies for Computing System Management” by Cannon; David M.; (Tucson, Ariz.); Humphries; Marshall L.; (Tucson, Ariz.) (filed on Apr. 14, 2010 and assigned to International Business Machines Corporation, Armonk, N.Y.) describes a method for providing real-time feedback regarding the effect of applying a policy definition used for management in a computing system.
“Adding Change Impact Analysis to the Formal Verification of C Programs” by Autexier; Serge and Luth; Christoph (appeared in Dominique Mery and Stephan Merz (Eds.), Proceedings 8th
International Conference on integrated Formal Methods (IFM2010), LNCS, Nancy, France, Springer, October, 2010) describes a framework based on document graph model to handle changes to programs and specifications efficiently as part of formal software verification.
“Modularity-Driven Clustering of Dynamic Graphs” by Gorke; Robert, Maillard; Pascal, Staudt; Christian, and Wagner; Dorothea (appeared in Experimental Algorithms, Lecture Notes in Computer Science, 2010, Volume 6049/2010, 436-448) describes graph analysis algorithms for efficiently maintaining a modularity based clustering of a graph that changes dynamically.
“A Graph-Theory Framework for Evaluating Landscape Connectivity and Conservation Planning” by Minor; Emily and Urban; Dean (appeared in Conservation Biology (Wiley-Blackwell), Volume 22, Issue 2, Pages 297-307, April 2008) describes a graph-theoretic approach to characterize multiple aspects of landscape connectivity in a habitat network and uses the notions of graph measures such as compartmentalization and clustering for the purposes of analysis.
The known systems do not address the issue of what-if analysis based on a comprehensive modeling of an educational institution at various levels in order to be able to provide for introspective analysis. The present invention provides for a system and method for what-if analysis based on a university model graph of the educational institution.
The primary objective of the invention is to achieve what-if analysis based on a university model graph (UMG) associated with an educational institution to help the educational institution in an introspective analysis.
One aspect of the present invention is to analyze a what-if analysis request and to derive a revised optimized university model graph.
Another aspect of the present invention is to interpret the revised optimized university model graph and generate recommendations.
Yet another aspect of the present invention is to find an optimal sub-UMG based on the university model graph.
Another aspect of the present invention is to minimally change to tune the university model graph so as achieve the set base scores of the select nodes of the university model graph.
Yet another aspect of the present invention is to determine the best set of entities and entity-instances among a few sets based on the university model graph.
Another aspect of the present invention is to select a sub-UMG and tune the sub-UMG.
Yet another aspect of the invention is to achieve the tuning of the university model graph based on a set of influence values.
Another aspect of the present invention is to achieve combining of two or more university model graphs.
a depicts an illustrative University Model Graph. 140 describes UMG as consisting of two main components: Entity Graph (142) and Entity-Instance Graph (144). Entity graph consists of entities of the university as its nodes and an abstract edge (146) or abstract link is a directed edge that connects two entities of the entity graph. Note that edge and link are used interchangeably. The weight associated with this abstract edge is the influence factor or influence value indicating nature and quantum of influence of the source entity on the destination entity. Again, influence factor and influence value are used interchangeably. Similarly, the nodes in the entity-instance graph are the entity instances and the edge (148) or the link between two entity-instances is a directed edge and the weight associated with the edge indicates the nature and quantum of influence of the source entity-instance on the destination entity-instance.
b provides the elements of a University Model Graph. The fundamental elements are nodes and edges. There are two kinds of nodes: Abstract nodes (160 and 162) and Nodes (164 and 166); There are three kinds of directed edges or links: Abstract links (168), links (170 and 172), and semi-abstract links (174 and 176). As part of the modeling, the abstract nodes are mapped onto entities and nodes are mapped onto the instances of the entities; Each node is associated with an entity-specific instantiated model and a node score that is a value between 0 and 1 is based on the entity-specific instantiated model; This score is called as Base Score; the weight associated with an abstract link corresponds to an entity influence value (EI-Value), the weight associated with a semi-abstract link corresponds to either an entity-entity-instance influence value (EIEI-Value) or an entity-instance-entity influence value (IEEI-Value), and finally, the weight associated with a link corresponds to an entity-instance influence value (I-Value). Note that edges and links are used interchangeably. Further, each entity is associated with a model and an instance of an entity is associated with a base score and an instantiated model, wherein the base score is computed based on the associated instantiated model and denotes the assessment of the entity instance. The weight associated with a directed edge indicates the nature and quantum of influence of the source node on the destination node and is a value between −1 and +1; This weight is called as Influence Factor.
About What-If Scenarios (300):
1. There are several scenarios that are of interest with respect to a university.
2. Analyzing these scenarios based on University Model Graph provides an opportunity for the university under consideration to have a better operational control.
3. How is UMG suited for What-If analysis?
UMG brings out an impact of an entity-instance on one or more of the entity instances;
This impact indicates how positiveness and negativeness spread throughout the university;
By controlling these two impacts, the university gets an opportunity to manage its internal operations and resources in an efficient manner;
Further, as the UMG captures impacts at both entity and entity-instance levels, it allows for a very fine-grained control on the university.
4. Illustrative scenarios:
A. How to allocate CAPEX—Determining the best way to distribute the annual budget keeping in mind to optimize on the overall and particular assessments;
B. How to improve the industry participation and sponsorships—Identifying of key faculty members and helping them improve their overall profile;
C. What is the impact of organizing seminars and conferences—In particular, helps in student and faculty member participation enhancing the overall assessment;
D. What is the impact of improving library infrastructure—In general, this has a wide ranging impact helping in faculty members and students, and on projects and seminars; and
E. What is the impact of a faculty member moving out—a faculty member has an influencing impact on peer faculty members and students.
400 provides an illustrative parametric model of STUDENT entity. Note that the generated recommendations are based on parameter values where there seems to be a scope for improvement. The computations are illustrative in nature with the overall score arrived based on the weighted summation.
Similarly, 420 provides a few recommendations based on a hierarchical model associated with LIBRARY entity. Please note that the computations are for illustrative purposes and combined as a weighted summation at each level in the hierarchy.
Again, 440 provides a few recommendations based on an activity based model associated with FACULTY MEMBER entity. Please note that the computations are for illustrative purposes and combined as a weighted summation at each level in the activity hierarchy.
Means for (analysis of a what-if request) Generic Techniques for What-If Analysis (500):
1. Given a UMG, find an optimal sub-UMG.
2. Given a set S of entities and entity-instances along with the base scores, find out the minimal changes to UMG to achieve the scores as per S.
3. Given a few sets, S1, S2, . . . , and Sn, and a UMG, find out which Si is the best.
4. Local analysis: Select a sub-UMG, and perform Techniques 2 and 3 above.
5. Given a set PS of paired entities/entity-instances, and a UMG, change the I-Values minimally within plus or minus threshold, and determine the optimal UMG.
6. Change the I-Values minimally of as many entities/entity-instances as possible so that the base scores of entities/entity-instances change minimally by a given percentage.
7. Given two or more UMGs, combine them to generate a merged-UMG.
These techniques play an important role in the analysis and processing of a what-if request.
Consider an entity-instance EIj;
Looking from this node perspective, EIj influences positively some nodes, negatively some nodes, gets positively influenced by some nodes, and negatively influenced by some nodes;
As depicted in 620, the node EIj has influences shown by arrow marks: Dotted incoming arrows indicate negative incoming influences, dotted outgoing arrows indicate negative outgoing influences, thick incoming arrows indicate positive incoming influences, and thick outgoing arrows indicate positive outgoing influences.
The objective is that when a negative influence value is reduced, effort should be made to increase the positive influence by a similar factor.
As described above, there are four distinct cumulative influence values (640): N1 nodes negatively influence EIj with an aggregated value of InNI and this value is denoted by −I3; Similarly, EIj influences N2 nodes negatively with an aggregated value of OutNI and this value is denoted by −I1; N3 nodes positively influence EIj with an aggregated value of InPI and this value is denoted by +I4; and, EIj influences N4 nodes positively with an aggregated value of OutPI and this value is denoted by +I2.
Balance −I1 by +I2 and similarly, balance −I3 by +I4.
What it means is that more negatives in UMG provide more opportunities for improvement.
A way is to distribute negatives equally on the positive entity instance influences.
Means for an approach for determining an optimal sub-UMG (660):
Output—an Optimal sub-UMG
Step 2: For each node Nj, Compute the following:
InNI—Sum of incoming negative influences;
N1—Number of nodes collectively influencing InNI;
OutNI—Sum of outgoing negative influences;
N2—Number of nodes collectively influencing OutNI;
InPI—Sum of incoming positive influences;
N3—Number of nodes collectively influencing InPI;
OutPI—Sum of outgoing positive influences;
N4—Number of nodes collectively influencing OutPI;
Here, the node denotes either an entity or entity-instance.
Increment each influence value (edge value) due to OutPI by OutNI/N4;
Set the negative influence value (edge value) due to OutNI as 0;
Case N3>0:
Increment each influence value (edge value) InPI by InNI/N3;
Set the negative influence value (edge value) due to InNI as 0;
Case N4=0: //No OutPI
// No OutPI—nobody being positively influenced
// Take a quantum of InPI and reduce OutNI;
Let Alpha be a pre-defined threshold;
InPIAlpha=InPI*Aplha;
Increment each influence value (edge value) due to OutNI by InPIAlpha/N2;
Increment each influence value (edge value) due to InPI by InPIAlpha/N3
Case N3=0; //No InPI;
// No InPI—nobody influences positively;
// Take a quantum of OutPI and reducen InNI;
Let Beta be a pre-defined threshold;
OutPIBeta=OutPI*Beta;
Increment each influence value (edge value) due to InNI by OutPIBeta/N1;
Increment each influence value (edge value0 due to OutPI by OutPIBeta/N4;
Case N3=0 and N4=0:
// Nobody being positively influenced and nobody influences positively;
Remove the node;
Step 1: Input: A set S of nodes (entities/entity-instances);
Input: A UMG;
Output: A tuned UMG
Step 2: Base score of a node is affected by (a) change in parameter values of Parametric Function (PF) of the node; (b) change in I-Values (influence values) directly or indirectly leading to the node;
Step 3: Approach—Change the base scores and I-values of nodes minimally to achieve the result;
Realistically, a small epsilon changes to the base scores and I-Values are indeed possible;
Step 4: For each node N1 in S, find the nearest neighbors N1NN based on UMG;
For each N2 in N1NN,
Recompute the base scores by propagation of influence values;
Check whether each node of S has attained the required base score;
If NOT, expand the nearest neighbor set and Repeat.
Means for an Approach for Selecting the best Set given UMG (800):
Step 1: Input—A few sets S1, S2, . . . , Sk;
Input—A UMG
Output—Select the best set Sj
Step 2: Approach—Combine each Si with the UMG and determine SUM of (BaseScore across the nodes of the UMG);
Select Sj that maximizes the above SUM;
Step 3: Combining Si with UMG
Case 1: Si is a node and the corresponding node exists in the UMG;
Replace the node in UMG and compute the base scores and the sum of the base scores;
Si is a node and the corresponding node does not exist in the UMG;
Note: A new entity-instance needs to be created;
Based on Parametric Function and available data values,
Determine the Base Score of the node;
Based on positive and negative influencers, determine the possible I-Values with select nodes (entities/entity-instances) of the UMG;
Compute the base score and the sum of the base scores;
Case 2: Si is a set of nodes;
Repeat Case 1 for each Node of Si;
Case 3: Si is a sub-graph with I-Values;
Case 31: No common nodes;
Merge Si and UMG, and Recompute the sum of the base scores;
Case 32: Some nodes are common;
Replace the common nodes; take the better I-Value for each of the matching edge;
Merge the remaining nodes;
Recompute the base scores and the sum of the base scores;
Case 33: All nodes are common;
Replace and Recompute the sum of the base scores;
Step 2: Obtain the conditions for the selection of a sub-UMG;
Obtain the set S;
Step 3: Selection of Sub-UMG based on semantic conditions and semantic neighbors;
For example, consider the entity FACULTY MEMBER; for each such entity, define semantic neighbors; and continue in the same manner; As an illustration, FACULTY MEMBER, all courses offered by FACULTY MEMBER (nearest neighbors NNs), STUDENTS who have enrolled for each course, LAB where FACULTY MEMBER is an investigator, FUNDS allocated to LAB, FACULTY MEMBER co-working in LAB, . . .
Step 4: Perform Sub-UMG tuning based on 5;
Step 5: Obtain the sets S1, S2, . . . , Sk;
Step 6: Perform the selection of the best Sj based on Sub-UMG;
Means for an Approach for tuning UMG based on I-Values—1 (1000):
Step 1: Input—A set PS of entity-instance pairs;
Input—A UMG;
Step 2: For each edge E in PS,
Locate the corresponding edge in the UMG;
Increase the I-Value by an Epsilon;
Step 3: Recompute the base scores by I-Value propagation;
Means for an Approach for tuning UMG based on I-Values—2 (1020):
Step 3 (P1): Obtain a node N;
Change I-Values leading to N by Epsilon (a pre-defined threshold);
Check whether base score of N has changed by a given percentage;
Step 4 (P4): How to select N? Based on number of in-degrees, Sum of I-Values, . . . ;
Step 5 (P2): Select nearest neighbors NN of N;
For each N1 of NN, Perform P1;
Step 6: If the UMG has still more nodes left to be covered,
Select a new node based on P4 and Repeat;
Step 2: Output—A combined UMG (CUMG)
Step 4: Obtain the Next UMG from S;
Step 5: Case 1: Obtain the common nodes between the Next UMG and CUMG;
For each common node, replace with the best of base scores;
For each common edge, replace with the best of the I-Values;
Case 2: For each non-common node, suitably introduce into the CUMG;
Repeat until there are no more UMGs to be combined.
Interpreting What-IF analysis Results (1200):
Means and an approach for generating recommendations based on Parametric Model:
1. The interpretation is based on the model associated with a node of UMG that is a part of what-if analysis.
2. There are three kinds of models: Parametric model, Hierarchical model, and Activity-Based model.
3. Consider a parametric model: This model comprises of a set of parametric functions (PFs); Each PF is labeled with 1 or 0 indicating whether it is manipulable or not. That is, whether the parameter is amenable for reflecting any improvement.
4. Let SPF be a set of such manipulable parameters;
As an illustration, consider three parameters of SPF, X1, X2, and X3;
Define,
S=W1*X1+W2*X2+W3*X3;
Let Delta be the proposed to change to S; S′=S+Delta
The problem is to find changes in X1 (X1′), X2 (X2′), and X3 (X3′) such that
S′=W1*X1′+W2*X2′+W3*X3′
How do we solve this problem?
5. Each parameter X is a normalized value between 0 and 1;
With respect to each parameter, define a lower threshold (LT) and an upper threshold (UT) (1220);
If the value of X<LT, then it is difficult to demand an improvement; (Under Performance)
If the value of X>UT, then again, it is difficult to demand an improvement (Over Saturation)
If the value LT<X<UT, then there is a scope for improvement, with the expected improvements to increase from LT to 0.5 and then drop;
Interpreting What-IF analysis Results (Contd.) (1240):
Means and an approach for generating recommendations based on Parametric Model (Contd.):
For each Xi: If Xi<LT, Then Epsilon1=0;
Else If Xi>UT, Then Epsilon1=0;
Else If Xi<=0.5, Then, Epsilon1=(X−LT)/(0.5−LT);
Else If Xi>0.5, Then Epsilon1=(UT−X)/(UT−0.5);
9. Affect changes to parameters based Delta1, Delta2, and Delta3.
10. Suggest changes based on Delta1 and description associated with the each parameter;
Interpreting What-IF analysis Results (Contd.) (1260):
Means and an approach for generating recommendations based on Hierarchical Model:
1. Consider an illustrative hierarchical model (1270):
2. Let Base score of E1 be S; As an illustration, What-If analysis requires the value to be changed to S′;
Let Beta=S′−S;
3. Get the child nodes of E1; With respect to the illustrative model, N1, N2, and N3 are the child nodes;
4. Let X1, X2, and X3 be the Non-Leaf-values associated with the child nodes N1, N2, and N3;
6. Based on the semantic description of a node and the corresponding change, provide the recommendations;
7. Repeat the above steps for each of the child nodes.
Interpreting What-IF analysis Results (Contd.) (1280):
Means and an approach for generating recommendations based on Activity-Based Model:
1. Consider an illustrative Activity-based model (1290):
2. Let Base score of E1 be S; As an illustration, What-If analysis requires the value to be changed to S′;
Let Beta=S′−S;
3. Get the child nodes of E1; With respect to the illustrative model, N1, N2, and N3 are the child nodes;
4. Let X1, X2, and X3 be the Non-Leaf-values associated with the child nodes N1, N2, and N3;
6. Based on the semantic description of a node and the corresponding change, provide the recommendations;
7. Repeat the above steps for each of the child nodes.
The illustrated UMG (1300) is shown in two forms: A graph based depiction (1320) displays how the various nodes (that stand for entities/entity-instances) N1, N2, . . . , N11 are interconnected; further, the edges are indicated with the illustrative influence values that are a value between −1 and +1. An equivalent representation is in the form of adjacency matrix (1340). In this representation, the element values depict the influence values as shown. Further, the base score associated with each of the nodes is also indicated under the column “Base Score.” The depicted UMG is in its stable form after the influence values have been propagated. An illustrative propagation is shown wherein the influence values of the child nodes along with base scores are used in arriving at the updated base score of a parent node.
Thus, a system and method for what-if analysis based on a university model graph is disclosed. Although the present invention has been described particularly with reference to the figures, it will be apparent to one of the ordinary skill in the art that the present invention may appear in any number of systems that provide for what-if analysis of influence based structural representation. It is further contemplated that many changes and modifications may be made by one of ordinary skill in the art without departing from the spirit and scope of the present invention.
Number | Date | Country | Kind |
---|---|---|---|
3203/CHE/2010 | Oct 2010 | IN | national |