Semantic Discovery System User Survey

Slide 1: 

The Semantic Discovery System What Works for You – What Doesn’t?

Can you please help us to prioritise which of the Semantic Web “benefits” has... (what amount of /or any) ...relevance for you by clicking through the next few slides and casting an opinion? : 

Can you please help us to prioritise which of the Semantic Web “benefits” has... (what amount of /or any) ...relevance for you by clicking through the next few slides and casting an opinion? After one more explanatory slide we show images of a typical large corporate set of Oracle Databases connected via our Semantic Query product – SDS. At first sight these screens look like many other ‘traditional’ (pre Semantic) Business Intelligence Graphical screens. Our request to you is to guide us with feedback on the merits (we hope there will be some) of: What we show SDS ‘doing’ - is this something you would care about? What we haven’t shown SDS ‘doing’ but are capabilities in development – ditto? Slide 1

The next few slides show the practical difference experienced when wishing to construct ad hoc queries of multiple distributed Oracle tables (AND/OR Excel Spreadsheets ) using the SDS Semantic Query approach versus the pre-Semantic ‘Query Builder Wizard’ type of tool. : 

The next few slides show the practical difference experienced when wishing to construct ad hoc queries of multiple distributed Oracle tables (AND/OR Excel Spreadsheets ) using the SDS Semantic Query approach versus the pre-Semantic ‘Query Builder Wizard’ type of tool. Slide A shows what one actually sees when connected to a large Oracle system – many, many tables with hard to understand table names and relationships. Slide B shows the capability of Semantics to ‘just’ select a small, meaningful and visualizable subset/model of all that data with names that make sense. Slide C is the key “ground-floor” differentiator – in our opinion – of the Semantic approach. Here you see that just by clicking on (emboldening) 2 or 3 attributes you have in effect constructed a query – which in turn brings back the data. Our assertion then is that you, the scientists, can get to Slide C, i.e. phrase and answer a very sophisticated question in a few seconds and moreover, repeat that process to ask many such questions, with a much faster cycle time than having to use query ‘wizards’. Our derivative assertion is that with any one small model (of 10-15 concepts say) one can ask a very large number of different questions very quickly, and then move on to any other models of interest to do the same. Slide 2

A: This slide merely serves to show a typical “as is” RDBMS - dozens of hard to understand tables and their inter-relationships – our aim with SDS is to get from there to the next slide ASAP so that useful work can start to be done : 

A: This slide merely serves to show a typical “as is” RDBMS - dozens of hard to understand tables and their inter-relationships – our aim with SDS is to get from there to the next slide ASAP so that useful work can start to be done Slide 3

B: Here, SDS has chosen a (could be any subset) collection of Concepts that naturally connect together and made their names more intelligible so that we can formulate questions that make sense – this is of course still dynamically connected to the large Oracle database ‘underneath’. : 

B: Here, SDS has chosen a (could be any subset) collection of Concepts that naturally connect together and made their names more intelligible so that we can formulate questions that make sense – this is of course still dynamically connected to the large Oracle database ‘underneath’. Slide 4

C: Note now, that from this same diagram/model, we are forming a Semantic Query simply by highlighting the attributes shown below - here this highlighting is saying “Find me the treatment descriptions corresponding to Donors in this Group …” : 

C: Note now, that from this same diagram/model, we are forming a Semantic Query simply by highlighting the attributes shown below - here this highlighting is saying “Find me the treatment descriptions corresponding to Donors in this Group …” Slide 5

C(i): This screen is the same as the last one after we have pressed the Green ‘get me the answers’ button. We believe that the ‘flow’ of the logic is self evident and we assert that because you only had to click on three attributes to achieve a 6-table Multi-million row Oracle Join you saved considerable query formulation time. : 

C(i): This screen is the same as the last one after we have pressed the Green ‘get me the answers’ button. We believe that the ‘flow’ of the logic is self evident and we assert that because you only had to click on three attributes to achieve a 6-table Multi-million row Oracle Join you saved considerable query formulation time. Slide 6

Given the three preceding slides, do you think the following assertions hold true? : 

Given the three preceding slides, do you think the following assertions hold true? [QUESTION 1] I can see that I could use a ‘model’ expressed as a set of high level Concept names (DONOR …) and their inter-relationships ( has sample) to express a statement of what I want to find out – this would be of value if it was connected to live data and would let me get answers as shown – (Y/N) [QUESTION 2] I can see that at this high level, I would not need to specify that Treatment “ID” has to equal Donor Group “ID/whatever it may be” all the way through the 6 table “join” that I have in fact just specified implicitly – this would save me some valuable time for every query - (Y/N) Slide 7

Here we move on to the Semantic Web “benefits” of refining your Model, Inferencing and so on – Would you tell us what amount of /or any relevance this has for you by clicking through the next 3 slides and casting an opinion? : 

Here we move on to the Semantic Web “benefits” of refining your Model, Inferencing and so on – Would you tell us what amount of /or any relevance this has for you by clicking through the next 3 slides and casting an opinion? These next 3 slides carry on from the previous model based on a large corporate set of Oracle Databases connected via our Semantic Query product - SDS. The first slide shows that you can tailor your own Desktop Model by adding your own idea of sub-concepts and so on – the thinking is that it makes asking questions more expressive. The next two slides from Prof Alan Rector is an introduction to Inferencing (posh word for re-classifying) which the SDS ‘machine’ can do in the background to generate some non-obvious new enhancements to your models. (There is more detail on this in the “Cookbook” on our website) Slide 8

These two concepts - GENE and TRANSCRIPT – are ‘Virtual ‘ i.e. they don’t ‘exist’ in the underlying databases – but they may help you Model your thinking. These “Semantic” concepts are subclasses of Molecular Target (which does exist as a database table). They may help me query the data in my model with more expressivity. : 

These two concepts - GENE and TRANSCRIPT – are ‘Virtual ‘ i.e. they don’t ‘exist’ in the underlying databases – but they may help you Model your thinking. These “Semantic” concepts are subclasses of Molecular Target (which does exist as a database table). They may help me query the data in my model with more expressivity. Slide 9

The last slide showed a simple means of adding virtual concepts to your model. There are many more sophisticated things that can be done along these lines – [QUESTION 3] are they of value to you? (Y/N) : 

The last slide showed a simple means of adding virtual concepts to your model. There are many more sophisticated things that can be done along these lines – [QUESTION 3] are they of value to you? (Y/N) Prof Alan Rector is an acknowledged expert in Medical ontologies – some tremendous examples exist here: http://protege.stanford.edu/conference/2005/slides/5.2_Rector_Why_classify_Protege_workshop_2005-v2.pdf Here (next slide) he uses the example of Pericarditis: Slide 10

Slide 12: 

The idea is that we can use a classifier in SDS to unify the two different views of the world (anatomist versus clinician) to provide not-immediately-obvious-conclusions: Slide 11

Now we move on to the ‘dirty data problem’ - ‘this GENE is known as abc here and xyz there – how can I ever reconcile them.’ Semantic Web/SDS Controlled Vocabularies and Ontologies *may* be of great benefit – but in what way? : 

Now we move on to the ‘dirty data problem’ - ‘this GENE is known as abc here and xyz there – how can I ever reconcile them.’ Semantic Web/SDS Controlled Vocabularies and Ontologies *may* be of great benefit – but in what way? We are all very familiar with the difficulties outlined above – the ambiguities and inconsistencies riddle our databases and its seems to be an intractable problem. [QUESTION 4] If this problem was solved for you (leaving aside for a moment the magical “how”) would the benefit be seen for you in a better and more accurate means of ‘searching’ +/- asking questions such as the ones we have just seen? (Y/N) [QUESTION 5] Or in what way(s) would this provide benefit. Slide 12

[QUESTION 6] Finally – some Semantic Web Arcana – is the ability to Semantically Search internal Corporate Oracle ‘as if’ one was using Google – of value? (Y/N) [QUESTION 7] What about the Holy Grail of expressing a query that will find any relationships that exist between any two items of data as below? : 

[QUESTION 6] Finally – some Semantic Web Arcana – is the ability to Semantically Search internal Corporate Oracle ‘as if’ one was using Google – of value? (Y/N) [QUESTION 7] What about the Holy Grail of expressing a query that will find any relationships that exist between any two items of data as below? “I don’t know if or how these two items of data may be connected but if they are, I want to find out about the paths by which they are connected” This extract - courtesy Prof Amit Sheth & Kemafor Anyanwu - gives an overview of the upcoming Semantic Connect functionality: “As a motivation for the domains used in our dataset consider the following example. A fraud investigator with the Securities and Exchange Commission (SEC) receives the following piece of information about a week after the stock prices for EntertainmentCompany_9982 plummet. Actor_5567 sold 70% of his shares of EntertainmentCompany_9982 one week after Capt_8262 sold all of his shares in the same company. Both transactions took place two weeks before the prices plummeted. The example sub graph shown in Figure 6.1 might help an investigator visualize the connections between the resources Actor_5567 and Captain_8262.” Slide 13

Could you please give us feedback on these summary headings of what you have seen? : 

Could you please give us feedback on these summary headings of what you have seen? [QUESTION 8] Semantic Web/SDS as a ‘better mousetrap for asking queries (Useful/Not)? [QUESTION 9] Semantic Web/SDS letting you create more elegant/focused/personalised models for asking queries (Useful/Not)? [QUESTION 10] Semantic Web/SDS letting you use other peoples Ontologies (e.g. Snomed/OWL) +/-inferencing/reasoning to enhance your models, for asking queries (Useful/Not)? [QUESTION 11] Semantic Web/SDS letting you discover non-obvious hidden connections between data – “has this compound ever had any adverse effect on these new pathways we are starting to study”? (Useful/Not)? [QUESTION 12] Did we miss anything? Thanks so much! Slide 14