Title: | Beta Record Linkage |
Version: | 0.1.0 |
Description: | Implementation of the record linkage methodology proposed by Sadinle (2017) <doi:10.1080/01621459.2016.1148612>. It handles the bipartite record linkage problem, where two duplicate-free datafiles are to be merged. |
Depends: | R (≥ 3.5.0) |
License: | GPL-3 |
Encoding: | UTF-8 |
LazyData: | true |
Imports: | utils |
RoxygenNote: | 7.0.2 |
URL: | https://github.com/msadinle/BRL |
BugReports: | https://github.com/msadinle/BRL/issues |
NeedsCompilation: | yes |
Packaged: | 2020-01-11 01:12:16 UTC; Mauricio Sadinle |
Author: | Mauricio Sadinle |
Maintainer: | Mauricio Sadinle <msadinle@uw.edu> |
Repository: | CRAN |
Date/Publication: | 2020-01-13 16:50:06 UTC |
Beta Record Linkage
Description
Beta record linkage methodology for probabilistic bipartite record linkage: the task of merging two duplicate-free
datafiles that lack unique identifiers.
This function runs all the steps of beta record linkage: creates comparisons of the records, runs Gibbs sampler on bipartite matchings,
and derives point estimate of bipartite matching (this determines the final linkage). The parameters of BRL
consist of
all the parameters needed to run compareRecords
, bipartiteGibbs
and linkRecords
,
except for intermediate input/output, and in addition to a parameter burn
for the burn-in period of the Gibbs sampler.
Usage
BRL(
df1,
df2,
flds = NULL,
flds1 = NULL,
flds2 = NULL,
types = NULL,
breaks = c(0, 0.25, 0.5),
nIter = 1000,
burn = round(nIter * 0.1),
a = 1,
b = 1,
aBM = 1,
bBM = 1,
seed = 0,
lFNM = 1,
lFM1 = 1,
lFM2 = 2,
lR = Inf
)
Arguments
df1 , df2 |
two datasets to be linked, of class |
flds |
a vector indicating the fields to be used in the linkage. Either a |
flds1 , flds2 |
vectors indicating the fields of |
types |
a vector of characters indicating the comparison type per comparison field. The options
are: |
breaks |
break points for the comparisons to obtain levels of disagreement.
It can be a list of length equal to the number of comparison fields, containing one numeric vector with the break
points for each comparison field, where entries corresponding to comparison type |
nIter |
number of iterations of Gibbs sampler. |
burn |
number of iterations to discard as part of the burn-in period. |
a , b |
hyper-parameters of the Dirichlet priors for the |
aBM , bBM |
hyper-parameters of beta prior on bipartite matchings. Default is |
seed |
seed to be used for pseudo-random number generation. By default it sets |
lFNM |
individual loss of a false non-match in the loss functions of Sadinle (2017), default |
lFM1 |
individual loss of a false match of type 1 in the loss functions of Sadinle (2017), default |
lFM2 |
individual loss of a false match of type 2 in the loss functions of Sadinle (2017), default |
lR |
individual loss of 'rejecting' to make a decision in the loss functions of Sadinle (2017), default |
Details
Beta record linkage (BRL, Sadinle, 2017) is a methodology for probabilistic bipartite record linkage, that is, the task of merging two duplicate-free datafiles that lack unique identifiers. This is accomplished by using the common partially identifying information of the entities contained in the datafiles. The duplicate-free requirement means that we expect each entity to be represented maximum once in each datafile. This methodology should not be used with datafiles that contain duplicates nor should it be used for deduplicating a single datafile.
The first step of BRL, accomplished by the function compareRecords
, consists of constructing comparison vectors for each pair of records from the two datafiles.
The current implementation allows binary comparisons (agree/disagree), numerical comparisons based on the absolute difference,
and Levenshtein-based comparisons.
This can be easily extended to other comparison types, so a resourceful user should be able to construct an object that recreates
the output of compareRecords
for other types of comparisons (so long as they get transformed to levels of disagreement), and still be able to run the next step outside
the function BRL
.
The second step of BRL, accomplished by the function bipartiteGibbs
, consists of running a Gibbs sampler that explores the space of bipartite matchings
representing the plausible ways of linking the datafiles. This sampler is derived from a model for the comparison data and a beta prior
distribution on the space of bipartite matchings. See Sadinle (2017) for details.
The third step of BRL, accomplished by the function linkRecords
, consists of deriving a point estimate of the bipartite matching
(which gives us the optimal way of linking the datafiles)
by minimizing the expected value of
a loss function that uses different penalties for different types of linkage errors. The current implementation only supports the
Bayes point estimates of bipartite matchings that can be obtained in closed form according to Theorems 1, 2 and 3 of Sadinle (2017).
The losses have to be positive numbers and satisfy one of three conditions:
Conditions of Theorem 1 of Sadinle (2017):
(lR == Inf) & (lFNM <= lFM1) & (lFNM + lFM1 <= lFM2)
Conditions of Theorem 2 of Sadinle (2017):
((lFM2 >= lFM1) & (lFM1 >= 2*lR)) | ((lFM1 >= lFNM) & (lFM2 >= lFM1 + lFNM))
Conditions of Theorem 3 of Sadinle (2017):
(lFM2 >= lFM1) & (lFM1 >= 2*lR) & (lFNM >= 2*lR)
If one of the last two conditions is satisfied, the point estimate might be partial, meaning that there might be some records in datafile 2 for which the point estimate does not include a linkage decision. For combinations of losses not supported here, the linear sum assignment problem outlined by Sadinle (2017) needs to be solved.
Value
A vector containing the point estimate of the bipartite matching, as in the output of linkRecords
. If lR != Inf
the output might be a partial estimate.
A number smaller or equal to n1
in entry j
indicates the record in datafile 1 to which record j
in datafile 2
gets linked, a number n1+j
indicates that record j
does not get linked to any record in datafile 1, and the value -1
indicates a 'rejection' to link, meaning that the correct linkage decision is not clear.
References
Mauricio Sadinle (2017). Bayesian Estimation of Bipartite Matchings for Record Linkage. Journal of the American Statistical Association 112(518), 600-612. [Published] [arXiv]
See Also
compareRecords
for examples on how to work with different types of comparison data,
bipartiteGibbs
for Gibbs sampler on bipartite matchings, and linkRecords
for examples
on full and partial point estimates of the true bipartite matching that indicates which records to link.
Examples
data(twoFiles)
(Zhat <- BRL(df1, df2, flds=c("gname", "fname", "age", "occup"),
types=c("lv","lv","bi","bi")))
n1 <- nrow(df1)
Ztrue <- df2ID
## number of links (estimated matches)
nLinks <- sum(Zhat <= n1)
## number of actual matches according to the ground truth
nMatches <- sum(Ztrue <= n1)
## number of links that are actual matches
nCorrectLinks <- sum(Zhat[Zhat<=n1]==Ztrue[Zhat<=n1])
## compute measures of performance
## precision
nCorrectLinks/nLinks
## recall
nCorrectLinks/nMatches
## the linked record pairs
cbind( df1[Zhat[Zhat<=n1],], df2[Zhat<=n1,] )
## finally, note that we could run BRL step by step as follows
## create comparison data
myCompData <- compareRecords(df1, df2,
flds=c("gname", "fname", "age", "occup"),
types=c("lv","lv","bi","bi"))
## Gibbs sampling from posterior of bipartite matchings
chain <- bipartiteGibbs(myCompData)
## bipartite matching Bayes estimate derived from the loss functions of Sadinle (2017)
Zhat2 <- linkRecords(chain$Z, n1=n1)
identical(Zhat, Zhat2)
Gibbs Sampler Used for Beta Record Linkage
Description
Run a Gibbs sampler to explore the posterior distribution of bipartite matchings that represent the linkage of the datafiles in beta record linkage.
Usage
bipartiteGibbs(cd, nIter = 1000, a = 1, b = 1, aBM = 1, bBM = 1, seed = 0)
Arguments
cd |
a list with the same structure as the output of the function
|
nIter |
number of iterations of Gibbs sampler. |
a , b |
hyper-parameters of the Dirichlet priors for the |
aBM , bBM |
hyper-parameters of beta prior on bipartite matchings. Default is |
seed |
seed to be used for pseudo-random number generation. By default it sets |
Value
a list containing:
Z
matrix with
n2
rows andnIter
columns containing the chain of bipartite matchings. A number smaller or equal ton1
in rowj
indicates the record in datafile 1 to which recordj
in datafile 2 is linked at that iteration, otherwisen1+j
.m,u
chain of
m
andu
parameters in the model for the comparison data among matches and non-matches, respectively.
References
Mauricio Sadinle (2017). Bayesian Estimation of Bipartite Matchings for Record Linkage. Journal of the American Statistical Association 112(518), 600-612. [Published] [arXiv]
Examples
data(twoFiles)
myCompData <- compareRecords(df1, df2, flds=c("gname", "fname", "age", "occup"),
types=c("lv","lv","bi","bi"))
chain <- bipartiteGibbs(myCompData)
Creation of Comparison Data
Description
Create comparison vectors for all pairs of records coming from two datafiles to be linked.
Usage
compareRecords(
df1,
df2,
flds = NULL,
flds1 = NULL,
flds2 = NULL,
types = NULL,
breaks = c(0, 0.25, 0.5)
)
Arguments
df1 , df2 |
two datasets to be linked, of class |
flds |
a vector indicating the fields to be used in the linkage. Either a |
flds1 , flds2 |
vectors indicating the fields of |
types |
a vector of characters indicating the comparison type per comparison field. The options
are: |
breaks |
break points for the comparisons to obtain levels of disagreement.
It can be a list of length equal to the number of comparison fields, containing one numeric vector with the break
points for each comparison field, where entries corresponding to comparison type |
Value
a list containing:
comparisons
-
matrix with
n1*n2
rows, where the comparison pattern for record pair(i,j)
appears in row(j-1)*n1+i
, fori
in{1,\dots,n1}
, andj
in{1,\dots,n2}
. A comparison field withL+1
levels of disagreement, is represented byL+1
columns of TRUE/FALSE indicators. Missing comparisons are coded as FALSE, which is justified under an assumption of ignorability of the missing comparisons, see Sadinle (2017). n1,n2
the datafile sizes,
n1 = nrow(df1)
andn2 = nrow(df2)
.nDisagLevs
a vector containing the number of levels of disagreement per comparison field.
compFields
a data frame containing the names of the fields in the datafiles used in the comparisons and the types of comparison.
References
Mauricio Sadinle (2017). Bayesian Estimation of Bipartite Matchings for Record Linkage. Journal of the American Statistical Association 112(518), 600-612. [Published] [arXiv]
Examples
data(twoFiles)
myCompData <- compareRecords(df1, df2,
flds=c("gname", "fname", "age", "occup"),
types=c("lv","lv","bi","bi"),
breaks=c(0,.25,.5))
## same as
myCompData <- compareRecords(df1, df2, types=c("lv","lv","bi","bi"))
## let's transform 'occup' to numeric to illustrate how to obtain numeric comparisons
df1$occup <- as.numeric(df1$occup)
df2$occup <- as.numeric(df2$occup)
## using different break points for 'lv' and 'nu' comparisons
myCompData1 <- compareRecords(df1, df2,
flds=c("gname", "fname", "age", "occup"),
types=c("lv","lv","bi","nu"),
breaks=list(lv=c(0,.25,.5), nu=0:3))
## using different break points for each comparison field
myCompData2 <- compareRecords(df1, df2,
flds=c("gname", "fname", "age", "occup"),
types=c("lv","lv","bi","nu"),
breaks=list(c(0,.25,.5), c(0,.2,.4,.6), NULL, 0:3))
Bayes Estimates of Bipartite Matchings
Description
Bayes point estimates of bipartite matchings that can be obtained in closed form according to Theorems 1, 2 and 3 of Sadinle (2017).
Usage
linkRecords(Zchain, n1, lFNM = 1, lFM1 = 1, lFM2 = 2, lR = Inf)
Arguments
Zchain |
matrix as the output |
n1 |
number of records in datafile 1. |
lFNM |
individual loss of a false non-match in the loss functions of Sadinle (2017), default |
lFM1 |
individual loss of a false match of type 1 in the loss functions of Sadinle (2017), default |
lFM2 |
individual loss of a false match of type 2 in the loss functions of Sadinle (2017), default |
lR |
individual loss of 'rejecting' to make a decision in the loss functions of Sadinle (2017), default |
Details
Not all combinations of losses lFNM, lFM1, lFM2, lR
are supported. The losses have to be positive numbers and satisfy one of three conditions:
Conditions of Theorem 1 of Sadinle (2017):
(lR == Inf) & (lFNM <= lFM1) & (lFNM + lFM1 <= lFM2)
Conditions of Theorem 2 of Sadinle (2017):
((lFM2 >= lFM1) & (lFM1 >= 2*lR)) | ((lFM1 >= lFNM) & (lFM2 >= lFM1 + lFNM))
Conditions of Theorem 3 of Sadinle (2017):
(lFM2 >= lFM1) & (lFM1 >= 2*lR) & (lFNM >= 2*lR)
If one of the last two conditions is satisfied, the point estimate might be partial, meaning that there might be some records in datafile 2 for which the point estimate does not include a linkage decision. For combinations of losses not supported here, the linear sum assignment problem outlined by Sadinle (2017) needs to be solved.
Value
A vector containing the point estimate of the bipartite matching. If lR != Inf
the output might be a partial estimate.
A number smaller or equal to n1
in entry j
indicates the record in datafile 1 to which record j
in datafile 2
gets linked, a number n1+j
indicates that record j
does not get linked to any record in datafile 1, and the value -1
indicates a 'rejection' to link, meaning that the correct linkage decision is not clear.
References
Mauricio Sadinle (2017). Bayesian Estimation of Bipartite Matchings for Record Linkage. Journal of the American Statistical Association 112(518), 600-612. [Published] [arXiv]
Examples
data(twoFiles)
myCompData <- compareRecords(df1, df2, flds=c("gname", "fname", "age", "occup"),
types=c("lv","lv","bi","bi"))
chain <- bipartiteGibbs(myCompData)
## discard first 100 iterations of Gibbs sampler
## full estimate of bipartite matching (full linkage)
fullZhat <- linkRecords(chain$Z[,-c(1:100)], n1=nrow(df1), lFNM=1, lFM1=1, lFM2=2, lR=Inf)
## partial estimate of bipartite matching (partial linkage), where
## lR=0.5, lFNM=1, lFM1=1 mean that we consider not making a decision for
## a record as being half as bad as a false non-match or a false match of type 1
partialZhat <- linkRecords(chain$Z[,-c(1:100)], n1=nrow(df1), lFNM=1, lFM1=1, lFM2=2, lR=.5)
## for which records the decision is not clear according to this set-up of the losses?
undecided <- which(partialZhat == -1)
df2[undecided,]
## compute frequencies of link options observed in the chain
linkOptions <- apply(chain$Z[undecided, -c(1:100)], 1, table)
linkOptions <- lapply(linkOptions, sort, decreasing=TRUE)
linkOptionsInds <- lapply(linkOptions, names)
linkOptionsInds <- lapply(linkOptionsInds, as.numeric)
linkOptionsFreqs <- lapply(linkOptions, function(x) as.numeric(x)/sum(x))
## first record without decision
df2[undecided[1],]
## options for this record; row of NAs indicates possibility that record has no match in df1
cbind(df1[linkOptionsInds[[1]],], prob = round(linkOptionsFreqs[[1]],3) )
Two Datasets for Record Linkage
Description
Two data frames, df1
and df2
, containing 300 and 150 records of artificially created
individuals, where 50 of them are included in both datafiles. In addition, the vector df2ID
contains one entry per record in df2
indicating the true matching between the datafiles, codified as follows:
a number smaller or equal to n1=300
in entry j
indicates the record in df1
to which record j
in df2
truly matches, and a number n1+j
indicates that record j
in df2
does not match any record in df1
.
Usage
data(twoFiles)
Source
Extracted from the datafiles used in the simulation studies of Sadinle (2017). The datafiles were originally generated using code provided by Peter Christen (https://users.cecs.anu.edu.au/~Peter.Christen/).
References
Mauricio Sadinle (2017). Bayesian Estimation of Bipartite Matchings for Record Linkage. Journal of the American Statistical Association 112(518), 600-612. [Published] [arXiv]
Examples
data(twoFiles)
n1 <- nrow(df1)
## the true matches
cbind( df1[df2ID[df2ID<=n1],], df2[df2ID<=n1,] )
## alternatively
df1$ID <- 1:n1
df2$ID <- df2ID
merge(df1, df2, by="ID")
## all the records in a merged file
merge(df1, df2, by="ID", all=TRUE)