Session 2.3 semantics for safeguarding & security – a police story

Editor's Notes

• #3: Load Data As Is In reality, there’s two kinds of ETL: The kind that focuses on transformation, cleansing, and quality because there’s a business process that requires standardization because they have to count things, do math, or disambiguate. The kind that has to be done merely to allow RDBMS to function because of their dependency on harmonized, normalized, and deorthoganalized data.  We eliminate #2 completely, and enable a more iterative approach to #1. How does ML deliver this integration? Ingest data as is. No upfront data modeling required. ETL tool not required. Structured and unstructured data. This includes scalar, text, geospatial, binary, semantic triples. Data and meta data. Schemas accepted, but not required. Data Formats (XML, JSON, binary etc.) with efficient tokenized storage Methods Content Pump - high speed data loading, serial writes REST APIs Java Client API Node.js Client API Java / .NET XCC Competitive Advantage? Data modeling and transformation is not a mandatory pre-requisite to loading data.  
• #4: Quote is from the NPCC’s Policing Vision 2025 http://www.npcc.police.uk/NPCCBusinessAreas/ReformandTransformation/PolicingVision2025.aspx
• #6: Need information quickly, need confidence that it is correct, and need it to be complete, To know that they have a complete picture of a suspect/perpetrator Need to be able to see a full picture of a suspect in a live intelligence situation To have all of the right information to hand during a live intelligence situation
• #7: Major problem is their silos are blocking them Lots of small steps that taken individually do not seem insurmountable but taken together cause paralysis And the status quo ends often ends up winning in this situation because if you can’t figure out where to start then you start anywhere One PCC of a force shared that the highly place ministers are telling the police forces that they need to get on with their transformation
• #8: Story: Imagine you’re a police analyst and you need to provide a full history of a individual to allow officers to proceed with a terrorism investigation Currently the analysts need to log into several different databases. At some forces there are more than 10 different systems holding important data None of these systems is useful for combining and sharing the results of the analysis So in order to share they then copy and paste the data in a WORD or Excel spreadsheet This means that important intelligence reports are then further cut off from their source data and are not reproduceable
• #10: Google like search appliance Lists of documents - Keyword Searches No context False positives– for example Nigel Brown will bring back tons of false positives How do you see the person in their context if you can only search for their name and not their relationships?
• #11: The Promise - Access all data from a single user interface Migrate all of the data to a new relational database and use a search engine
• #12: Schema design for this number and complexity of systems is time consuming Losing the unstructured data e.g. the police intelligence report, the storm call transcipt e.g. the valuable data that gives meaning and context
• #13: Does more than just puts the data in one place Allows you to discover how that data is related by building connections between the information It’s the powerful combination of unstructured data and the semantic triples that help surface the relationships between people, events, locations and objects Able to exploit all of the value in the unstructured data that has heretofore been considered dark matter and therefore difficult to exploit
• #14: An example of the IDEAL SOLUTION Create a model for the delivery of shared services for future application development and for rolling out as good practice to other forces Increase effectiveness and outcomes of analysts and police officers Prove that data driven application design can produce results quickly
• #15: Single record for a person - not a list of documents Name misspellings Network diagram showing the relationships that have been identified by the system Information is organised by it's context e.g. person rather than by it's format e.g. storm call record
• #16: Combine external data sources with your rich linked data set Dictionaries of words/phrases associated with grooming highlighted in search results
• #17: Geospatial search making use of cleansed address data to show crime patterns over time and in geographic location
• #18: When there is confidence that a complete 360 view of an individual can be identified you can start to use statistical analysis to identify behavioral patterns that indicate vulnerability
• #19: Hand over to JW Higher contrast colour of purple. SB: - Explain up-front why you are using phonetic names – to deal with misspelling etc. - Same for dates – so I can type in a date in any format.
• #22: SB: - Out of UPRN can get lat, long -
• #23: Give a clearer description of a dimension combination - A combination of fields that uniquely identify the entity. SB: - Calculating hashes – so every unique entity can be uniquely identified.
• #25: Explanation of transitive
• #28: Loosely Integrated Multi System Stack SB: - We have demonstrate the approach of a multi-model data platform with semantics. - Other vendors try to replicate this using multiple technologies – but this approach is different.
• #29: Architects paid by the box. Doing joins is incredibly expensive. Brittle Question to Jen – still thinking of a good quesiton
• #30: Multi model is a good thing Multi-model in once place is even better. Simpler, transactional, all backed up together. Query across three models – all in once.