UPRN address matching algorithm - Revision history

DavidStables: /* Qualified matching */

2021-04-06T16:17:20Z

Qualified matching

← Older revision		Revision as of 16:17, 6 April 2021
Line 228:		Line 228:
	# Parent. The Discovery address represents a supra-property (parent) of the UPRN property. For example the Discovery address ‘1 Angel Lane’ does not have an equivalent in DPA or LPI but ‘flat 11, 1 Angel Lane’ does exist so it matches to a more detailed identifier qualified as a supra-property		# Parent. The Discovery address represents a supra-property (parent) of the UPRN property. For example the Discovery address ‘1 Angel Lane’ does not have an equivalent in DPA or LPI but ‘flat 11, 1 Angel Lane’ does exist so it matches to a more detailed identifier qualified as a supra-property
	# Sibling. The Discovery address may represent a sibling of the UPRN property. For example the Discovery address ‘flat 12 , 1 Angel Lane’ may not exist in either ABP file but ‘flat 11, 1 Angel Lane’ does		# Sibling. The Discovery address may represent a sibling of the UPRN property. For example the Discovery address ‘flat 12 , 1 Angel Lane’ may not exist in either ABP file but ‘flat 11, 1 Angel Lane’ does
			#Best match. This means that the algorithm thinks that it has found an entry as being the best match and the correct location. This does not mean it is an exact match, only that it thinks that the user 'candidate address' is the same location as the one that is listed below and thinks it is a better match than others. It may not be the case that it is the best match. Algorithms explain the "best fit" approach which differentiates what the machine thinks is the best match from what a human might think
			#Best (residential) match. indicates that the user has attempted to match only on residential properties or those that may be residential or dual use.
			#Best (+commercial) match Indicates that the user has included commercial properties in the match algorithm

	Qualifiers are assigned in relation to the final post manipulated address match.		Qualifiers are assigned in relation to the final post manipulated address match.

DavidStables: /* Match patterns */

2021-04-06T12:59:51Z

Match patterns

← Older revision		Revision as of 12:59, 6 April 2021
Line 244:		Line 244:
	There are around 12 match terms with around 50 or so theoretical combinations of those terms. For example a candidate field may be dropped to match, and matched as a sibling (ds). Applying these to 5 fields results in the potential of 300 million or so different combinations.		There are around 12 match terms with around 50 or so theoretical combinations of those terms. For example a candidate field may be dropped to match, and matched as a sibling (ds). Applying these to 5 fields results in the potential of 300 million or so different combinations.

	However, with the algorithms being determined by plausibility, not mathematics, only a number end up being used, usually around 200 or so across 100,000 addresses. further restrictions on the combination occur due to implausibility of some field swaps. For example, post codes are never swapped with streets. Streets are not moved to numbers (as this would have occurred during the initial address formatting algorithm)		However, with the algorithms being determined by plausibility, not mathematics, only a number end up being used, usually around 200 or so across 100,000 addresses. further restrictions on the combination occur due to implausibility of some field swaps. For example, post codes are never swapped with streets. Streets are not moved to numbers (as this would have occurred during the initial address formatting algorithm).

			The following table lists the match pattern
	{\| class="wikitable"		{\| class="wikitable"
	\|+		\|+

DavidStables: /* Match patterns */

2021-04-06T12:59:29Z

Match patterns

@@ Line 250: / Line 250: @@
 !Term
 !Description
 |-
 |&

DavidStables: /* Qualified matching */

2021-04-06T12:57:25Z

Qualified matching

Show changes

DavidStables: /* Qualified matching */

2020-08-10T14:01:46Z

Qualified matching

← Older revision		Revision as of 14:01, 10 August 2020
Line 225:		Line 225:
	As well as matching to a UPRN there is also requirement to qualify the relationship between a Discovery address and the UPRN. We refer to these qualifiers as “approximation qualifiers” as they mean that the UPRN is geographically close to the property. Each of the main 5 address fields is assigned a qualifier. The five qualifiers are:		As well as matching to a UPRN there is also requirement to qualify the relationship between a Discovery address and the UPRN. We refer to these qualifiers as “approximation qualifiers” as they mean that the UPRN is geographically close to the property. Each of the main 5 address fields is assigned a qualifier. The five qualifiers are:
	#Equivalent match i.e. we believe the property ''is'' the property that the UPRN refers to		#Equivalent match i.e. we believe the property ''is'' the property that the UPRN refers to
	#Near match. We believe the property is the property that the UPRN refers to but the flat or property may be split. This only applies to matches where the flat or number suffix in either the Discovery or the ABP addresses may be ignored or dropped. This is in effect equivalent.
	#Child - Sub- property. The Discovery address represents a sub-property of the UPRN property. For example the Discovery address “flat 1a Eagle house” may only match to the higher level DPA or LPI entry of “Eagle house”		#Child - Sub- property. The Discovery address represents a sub-property of the UPRN property. For example the Discovery address “flat 1a Eagle house” may only match to the higher level DPA or LPI entry of “Eagle house”
	# Parent. The Discovery address represents a supra-property (parent) of the UPRN property. For example the Discovery address ‘1 Angel Lane’ does not have an equivalent in DPA or LPI but ‘flat 11, 1 Angel Lane’ does exist so it matches to a more detailed identifier qualified as a supra-property		# Parent. The Discovery address represents a supra-property (parent) of the UPRN property. For example the Discovery address ‘1 Angel Lane’ does not have an equivalent in DPA or LPI but ‘flat 11, 1 Angel Lane’ does exist so it matches to a more detailed identifier qualified as a supra-property

Gill Harper: /* Overview - Best fit method */

2020-08-07T13:01:31Z

Overview - Best fit method

Show changes

DavidStables: /* The Standard address database */

2020-06-28T15:22:06Z

The Standard address database

← Older revision		Revision as of 15:22, 28 June 2020
Line 152:		Line 152:

	There can be more than one version of the LPI geographic address for a UPRN, capturing approved, alternative, provisional and historical (inactive) versions of the DSI exclude status 8.		There can be more than one version of the LPI geographic address for a UPRN, capturing approved, alternative, provisional and historical (inactive) versions of the DSI exclude status 8.
	All are included as potential matches		All are included as potential matches.

	~~address. All are available for matching~~.

	These files are considered as the standard address details for each UPRN. Whilst each file represents addresses in a slightly different way, they are aligned with each other. The LPI file contains entries that are often more granular than the DPA file. Some UPRNs only exist in the LPI file and not in the DPA file and vice versa. Thus the LPI file can usually be considered as the more definitive of the two but a match to either can be considered correct.		These files are considered as the standard address details for each UPRN. Whilst each file represents addresses in a slightly different way, they are aligned with each other. The LPI file contains entries that are often more granular than the DPA file. Some UPRNs only exist in the LPI file and not in the DPA file and vice versa. Thus the LPI file can usually be considered as the more definitive of the two but a match to either can be considered correct.

DavidStables: /* Overview - Best fit method */

2020-06-28T15:13:02Z

Overview - Best fit method

← Older revision		Revision as of 15:13, 28 June 2020
Line 4:		Line 4:
	The algorithms use human semantic pattern recognition, applying rankings of matching judgements following rules that manipulate the text, supported by a few machine based algorithms such as the [[wikipedia:Levenshtein_distance\|Levenshtein distance algorithm]].		The algorithms use human semantic pattern recognition, applying rankings of matching judgements following rules that manipulate the text, supported by a few machine based algorithms such as the [[wikipedia:Levenshtein_distance\|Levenshtein distance algorithm]].

	The rankings, which can be considered as a set of numbers, could be described as a plausibility measure, as opposed to a probability measure or deterministic measure.		The rankings, which can be considered as a set of numbers, 1-n, could be described as a plausibility measure, as opposed to a probability measure or deterministic measure.

	Firstly, some definitions of terms used:		Firstly, some definitions of terms used:

DavidStables: /* Overview - Best fit method */

2020-06-28T15:07:57Z

Overview - Best fit method

← Older revision		Revision as of 15:07, 28 June 2020
Line 3:		Line 3:

	The algorithms use human semantic pattern recognition, applying rankings of matching judgements following rules that manipulate the text, supported by a few machine based algorithms such as the [[wikipedia:Levenshtein_distance\|Levenshtein distance algorithm]].		The algorithms use human semantic pattern recognition, applying rankings of matching judgements following rules that manipulate the text, supported by a few machine based algorithms such as the [[wikipedia:Levenshtein_distance\|Levenshtein distance algorithm]].

			The rankings, which can be considered as a set of numbers, could be described as a plausibility measure, as opposed to a probability measure or deterministic measure.

	Firstly, some definitions of terms used:		Firstly, some definitions of terms used:

DavidStables: /* Overview - Best fit method */

2020-06-28T15:03:14Z

Overview - Best fit method

← Older revision		Revision as of 15:03, 28 June 2020
Line 1:		Line 1:
	== Overview - Best fit method ==		== Overview - Best fit method ==
	The address matching ~~algorithm uses~~ a human mediated best fit method to match a candidate address to one address from the set of all available 'standard' addresses.		The address matching algorithms use a human mediated best fit method to match a candidate address to one address from the set of all available 'standard' addresses.

	~~Address matching algorithm uses~~ human semantic ~~based~~ pattern recognition, ~~manipulation and matching judgement as it's~~ matching ~~technique~~, supported by a few machine based algorithms such as the [[wikipedia:Levenshtein_distance\|Levenshtein distance algorithm]].		The algorithms use human semantic pattern recognition, applying rankings of matching judgements following rules that manipulate the text, supported by a few machine based algorithms such as the [[wikipedia:Levenshtein_distance\|Levenshtein distance algorithm]].

	Firstly, some definitions of terms used:		Firstly, some definitions of terms used: