Patent Application
20240220547
In certain genealogical or family history databases, ancestor data is stored in trees and records which contain one or more persons or individuals. Genealogical data has various types. Genealogical data includes historical and genealogical records such as vital, census, voter, military, immigration, and church records. Genealogical data also includes user generated information. With all genealogical records and user generated data alternate spellings, typos, missing data, and bad data are common. Genealogical data may also include intra-tree or record relationships which indicate the relationships between the various individuals within the database. In many cases, persons in one data collection may correspond to persons in another data collection, as many genealogical records and user data have references to multiple individuals within the same collection. Also, user generated data may contain references to relatives within the same user tree.
One challenge in maintaining genealogical databases has been entity resolution, which refers to the problem of identifying and linking different manifestations of the same real world object. For example, many manifestations of the same person may appear across multiple trees. This problem arises due to discrepancies between different historical records, discrepancies between historical records and human accounts, and discrepancies between different human accounts. For example, different users having a common ancestor may have different opinions as to the name, date of birth, and place of birth of that ancestor. The problem becomes particularly prevalent when large amounts of historical documents are difficult to read, causing a wide range of possible ancestor data. Accordingly, there is a need for improved techniques in the area.
Examples given below provide a summary of the present invention. As used below, any reference to a series of examples is to be understood as a reference to each of those examples disjunctively (e.g., “Examples 1-4” is to be understood as “Examples 1, 2, 3, or 4”).
Example 1 is a computer-implemented method comprising: obtaining first tree data and second tree data, wherein the first tree data corresponds to a first tree person from a first genealogical tree and the second tree data corresponds to a second tree person from a second genealogical tree, and wherein each of the first genealogical tree and the second genealogical tree comprise a plurality of interconnected tree persons corresponding to individuals that are related to each other; extracting a set of features from the first tree data and the second tree data; generating, using an individual-level machine learning (ML) model, an individual-level similarity score for each possible pairing of tree persons from the first genealogical tree and the second genealogical tree based on the set of features; identifying, based on the individual-level similarity score for each possible pairing of tree persons, a set of most-similar tree persons from the first genealogical tree and the second genealogical tree; providing a plurality of individual-level similarity vectors for the set of most-similar tree persons as input to a family-level ML model; and determining, using the family-level ML model, that the first tree person and the second tree person correspond to a same individual based on the plurality of individual-level similarity vectors for the set of most-similar tree persons.
Example 2 is the computer-implemented method of example(s) 1, wherein at least one of the individual-level ML model or the family-level ML model is a neural network.
Example 3 is the computer-implemented method of example(s) 1, further comprising: prior to generating the individual-level similarity score for each possible pairing of tree persons, training the individual-level ML model using training data and user-prepared labels.
Example 4 is the computer-implemented method of example(s) 1, further comprising: generating, based on the set of features, an individual-level similarity vector for each possible pairing of tree persons.
Example 5 is the computer-implemented method of example(s) 4, wherein the individual-level similarity score for each possible pairing of tree persons is generated based on the individual-level similarity vector for each possible pairing of tree persons.
Example 6 is the computer-implemented method of example(s) 5, wherein providing the plurality of individual-level similarity vectors includes: selecting the plurality of individual-level similarity vectors for the set of most-similar tree persons from the individual-level similarity vector for each possible pairing of tree persons; or generating the plurality of individual-level similarity vectors for the set of most-similar tree persons.
Example 7 is the computer-implemented method of example(s) 5, wherein the family-level ML model includes a first stage and a second stage, wherein the first stage determines whether the first tree person and the second tree person do not correspond to the same individual, and wherein the second stage determines whether the first tree person and the second tree person correspond to the same individual.
Example 8 is a non-transitory computer-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising: obtaining first tree data and second tree data, wherein the first tree data corresponds to a first tree person from a first genealogical tree and the second tree data corresponds to a second tree person from a second genealogical tree, and wherein each of the first genealogical tree and the second genealogical tree comprise a plurality of interconnected tree persons corresponding to individuals that are related to each other; extracting a set of features from the first tree data and the second tree data; generating, using an individual-level machine learning (ML) model, an individual-level similarity score for each possible pairing of tree persons from the first genealogical tree and the second genealogical tree based on the set of features; identifying, based on the individual-level similarity score for each possible pairing of tree persons, a set of most-similar tree persons from the first genealogical tree and the second genealogical tree; providing a plurality of individual-level similarity vectors for the set of most-similar tree persons as input to a family-level ML model; and determining, using the family-level ML model, that the first tree person and the second tree person correspond to a same individual based on the plurality of individual-level similarity vectors for the set of most-similar tree persons.
Example 9 is the non-transitory computer-readable medium of example(s) 8, wherein at least one of the individual-level ML model or the family-level ML model is a neural network.
Example 10 is the non-transitory computer-readable medium of example(s) 8, wherein the operations further comprise: prior to generating the individual-level similarity score for each possible pairing of tree persons, training the individual-level ML model using training data and user-prepared labels.
Example 11 is the non-transitory computer-readable medium of example(s) 8, wherein the operations further comprise: generating, based on the set of features, an individual-level similarity vector for each possible pairing of tree persons.
Example 12 is the non-transitory computer-readable medium of example(s) 11, wherein the individual-level similarity score for each possible pairing of tree persons is generated based on the individual-level similarity vector for each possible pairing of tree persons.
Example 13 is the non-transitory computer-readable medium of example(s) 12, wherein providing the plurality of individual-level similarity vectors includes: selecting the plurality of individual-level similarity vectors for the set of most-similar tree persons from the individual-level similarity vector for each possible pairing of tree persons; or generating the plurality of individual-level similarity vectors for the set of most-similar tree persons.
Example 14 is the non-transitory computer-readable medium of example(s) 12, wherein the family-level ML model includes a first stage and a second stage, wherein the first stage determines whether the first tree person and the second tree person do not correspond to the same individual, and wherein the second stage determines whether the first tree person and the second tree person correspond to the same individual.
Example 15 is a system comprising: one or more processors; and a computer-readable medium comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising: obtaining first tree data and second tree data, wherein the first tree data corresponds to a first tree person from a first genealogical tree and the second tree data corresponds to a second tree person from a second genealogical tree, and wherein each of the first genealogical tree and the second genealogical tree comprise a plurality of interconnected tree persons corresponding to individuals that are related to each other; extracting a set of features from the first tree data and the second tree data; generating, using an individual-level machine learning (ML) model, an individual-level similarity score for each possible pairing of tree persons from the first genealogical tree and the second genealogical tree based on the set of features; identifying, based on the individual-level similarity score for each possible pairing of tree persons, a set of most-similar tree persons from the first genealogical tree and the second genealogical tree; providing a plurality of individual-level similarity vectors for the set of most-similar tree persons as input to a family-level ML model; and determining, using the family-level ML model, that the first tree person and the second tree person correspond to a same individual based on the plurality of individual-level similarity vectors for the set of most-similar tree persons.
Example 16 is the system of example(s) 15, wherein at least one of the individual-level ML model or the family-level ML model is a neural network.
Example 17 is the system of example(s) 15, wherein the operations further comprise: prior to generating the individual-level similarity score for each possible pairing of tree persons, training the individual-level ML model using training data and user-prepared labels.
Example 18 is the system of example(s) 15, wherein the operations further comprise: generating, based on the set of features, an individual-level similarity vector for each possible pairing of tree persons.
Example 19 is the system of example(s) 18, wherein the individual-level similarity score for each possible pairing of tree persons is generated based on the individual-level similarity vector for each possible pairing of tree persons.
Example 20 is the system of example(s) 19, wherein providing the plurality of individual-level similarity vectors includes: selecting the plurality of individual-level similarity vectors for the set of most-similar tree persons from the individual-level similarity vector for each possible pairing of tree persons; or generating the plurality of individual-level similarity vectors for the set of most-similar tree persons.
The accompanying drawings, which are included to provide a further understanding of the invention, are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the detailed description serve to explain the principles of the invention. No attempt is made to show structural details of the invention in more detail than may be necessary for a fundamental understanding of the invention and various ways in which it may be practiced.
In the appended figures, similar components and/or features may have the same numerical reference label. Further, various components of the same type may be distinguished by following the reference label with a letter or by following the reference label with a dash followed by a second numerical reference label that distinguishes among the similar components and/or features. If only the first numerical reference label is used in the specification, the description is applicable to any one of the similar components and/or features having the same first numerical reference label irrespective of the suffix.
Entity resolution is a long studied problem that still poses some interesting challenges. The problem is defined as identifying and linking different manifestations of the same real world object. Many organizations collect data on entities with potential duplicate data. For certain genealogical databases, each group of user nodes or records may be considered in the format of a genealogical tree or records connected by biological and other family relationships which will be designated herein by the general term “tree data”. Tree data nodes often have overlap with other tree data nodes which creates duplicate entities in a genealogical database. One way to track known duplicates is to resolve multiple entities into one cluster. This provides immense value to users by allowing them to receive more targeted hints, understand how they are related to people in records outside their personal tree, and have a more precise searching capabilities.
Genealogical data may be received from a source selected from the group consisting of the Ancestry World Tree system, a Social Security Death Index database, the World Family Tree system, a birth certificate database, a death certificate database, a marriage certificate database, an adoption database, a draft registration database, a veterans database, a military database, a property records database, a census database, a voter registration database, a phone database, an address database, a newspaper database, an immigration database, a family history records database, a local history records database, a business registration database, and a motor vehicle database. Additionally, genealogical data can be user-generated. Genealogical data may also include data from a clustered database derived from records and user data.
Some embodiments of the present invention relate to modifying a cluster database when it is determined that two persons in different trees data collections correspond to the same individual. In some instances, a node connection may be generated between two nodes corresponding to the two persons in different trees. This technique may be used incrementally during the generation of the cluster database. Thus, as a user creates a new node or edits an existing node, the disclosed technique may be used to check whether that node resolves to a current entity cluster or whether it should generate its own cluster.
Another advantage of identifying these duplicate persons is that one user or record collection may have information for their ancestor that another user does not. As more duplicate persons are identified, rather than list every duplicate person for the user to browse, it may be more helpful to show a user an aggregation of all the duplicates merged together in a single, concise view of the cluster (i.e., a grouping of duplicate persons). Another advantage of identifying duplicate persons is that if users A and B have a duplicate person in their family trees, then it can be determined that users A and B are related to each other at least via the duplicate person. Furthermore, once the duplicate person is identified, user A may supplement his/her family tree with information from user B's family tree, and vice-versa.
In order to resolve which records and tree nodes refer to which real world entities, some embodiments described herein reduce the problem to a pairwise classification problem where the classes are “Match”, “Non-Match” or a “Plausible” match. There may be many criteria to consider when establishing what constitutes a match. Some embodiments described herein attempt to replicate the matching criteria of a genealogist by using a machine learning (ML) algorithm to combine all these criteria into a prediction score. Some embodiments described herein employ a hierarchical classification algorithm which leverages the familial structure of genealogical records. To learn such a model, an algorithm is trained on thousands of pairwise entity comparisons labeled by professional genealogists to determine whether two records or tree nodes refer to the same entity. This approach was determined to provide significant improvements over previous rule-based approaches used for entity recognition.
With billions of tree nodes and records, genealogy companies have a wealth of data on which to perform entity resolution. The data also comes with a few challenges. First, the tree nodes are user generated data. This means that typos, missing data, and bad data are possible. To handle some of these issues, genealogy companies have systems which normalize information like places and dates to a common format and appropriately clean and tokenize fields like names. Otherwise, the data is treated as entered by the user. Second, records span many decades as well as formats and are digitized from records of various qualities.
In some instances, in order to compare two entities, genealogists may have access to a limited amount of external source information about these persons or records provided by a tool. In some instances, active learning is used to better guide the labeling process of the genealogists. The main idea behind active learning is that a model can learn with fewer data if it can help choose the most effective data to be labeled. This assumes that not every piece of labeled data provides the same value to the model. For example, if a significant amount of census records had been labeled, but not very many death records, the model may learn more from an additional death record compare as opposed to a census record compare. Once a large enough sample of randomly selected labeled data with which to build a baseline model is obtained, the labeling efforts may be focused on compares for which the model was unsure or disagreed with predictions from previous rule-based models. Also, since deciding whether two entities match can be a subjective task, at least two professional genealogists may label each compare to help reduce the noise.
In some embodiments, extracted features include information on names, births, deaths, marriages, and other information from the compare. Thus, it can be determined how similar the relative information matches for each relative type. To limit the number of features, when a compare has more than three children, the three children that have the highest similarity score may be used. Similarly when a compare has more than one spouse, the one with the highest matching similarity score can be used. In addition to these features, family level features which encompass data from all members of the family can be extracted, such as how well do the locations and times of events match across both families.
Some embodiments of the present invention include a two-step learning process. First, a model is trained using only features from the person of interest, excluding family relations. Once trained, this model allows the strength of comparisons between family members to be evaluated. For example, given a comparison between two persons, this model can be used to determine how well their mothers match. This model can be referred to as the individual-level model. Once this model has extracted the most similar relations for mothers, fathers, spouses, and three children, features can be extracted from all these relations. This can result in almost 400 features, including the actual probabilities returned from the first model on the closeness of the family relations.
Next, a second model is trained using this extended feature set. This model can be referred to as the family-level model because it leverages information from the family relations of the entities. In some embodiments, the family-level model consists of two sub-models. The first sub-model is trained by treating matches and plausible matches as one class and non-matches as another class to first determine the probability that an entity compare is a match or plausible match. This first sub-model serves two purposes. First, it provides a similarity score between two entities. Second, this model learns how to separate clearly between non-matches and everything else without the burden of separating out plausible matches which can often look similar to matches. The second sub-model is trained only on the matches and the plausible matches. This sub-model focuses on separating out the clear matches from those which are more borderline. This distinction can be beneficial as it allows genealogy companies to provide different weights to plausible matches when providing recommendations to users.
As used herein, the terms “tree”, “family tree”, “genealogical tree”, and “tree data” may be used interchangeably and may refer to data from user data and historical records in multiple databases involving the grouping of a finite number of related persons that are interconnected according to their relationships. Two persons that are directly connected in a tree or record data may be in a parent-child relationship, in a sibling relationship, or in some other relationship. Tree data may be displayed as various points connected by lines. The base or bottom of the tree may comprise a single individual, which may or may not be a user of the tree.
As used herein, the terms “tree person”, “person”, “individual”, and “node” may be used interchangeably and may refer to a representation in a tree or historical record of a living or deceased real-life individual.
As used herein, the term “user” may refer to an owner or creator of a tree, or may refer to any entity, human or non-human, that is using a tree or genealogical database in some manner.
As used herein, the term “cluster” may refer to a grouping of tree persons, each from different trees and each determined to correspond to the same real-life individual. Although clusters are designed to group only tree persons that correspond to the same real-life individual, this is not always possible, and often clusters are either overinclusive or underinclusive based on the similarity threshold that is employed.
One method for determining whether Tree Persons A15, B13, and C5 correspond to the same real-life individual is a rule-based algorithm in which a human expert looks at different pairs of persons and creates rules. For example, consider that two persons are named “Jack Smith” but one is born in Mar. 1, 1981 and the other is born in Mar. 1, 1932. A rule-based algorithm may generate four separate scores, one for a comparison of the names (a high score in this example), one for a comparison of the month of birth (a high score in this example), one for a comparison of the day of birth (a high score in this example), and one for the year of birth (a low score in this example). The four separate scores are added together to generate a final similarity score. The higher the similarity score, the higher the probability that the two tree persons correspond to the same real-life individual.
There are several disadvantages to rule-based algorithms. First, they are subjective. When scores are combined into a final similarity score, they may be weighted such that the final similarity score is overly sensitive to the chosen weighting, which may be arbitrary. Second, rule-based algorithms become extremely complicated as they must account for several special cases, such as popular names. Third, rule-based algorithms are difficult to update and maintain. Over time, there may be hundreds of rules to generate a single final similarity score. If new special cases arise, a human expert has to verify whether all the previously generated rules will apply to the new case or not. If a particular rule does not apply, then a change may be needed.
In some embodiments, genealogical entity resolution system 200 includes a tree person selector for selecting tree data corresponding to two tree persons, referred to as TP1 and TP2, from tree database 212. Tree persons TP1 and TP2 are generally selected from two different family trees, although in some embodiments they may be selected from the same family tree. In some embodiments, genealogical entity resolution system 200 includes a feature extractor for extracting features from the tree data and likewise tree persons TP1 and TP2, and from additional tree persons that are selected from tree database 212.
In some embodiments, genealogical entity resolution system 200 includes a feature comparator for generating various similarity scores and vectors based on the features extracted by the feature extractor. In some instances, the first feature vector f1,i and the second feature vector f2,i are used to generate a metric function s(fi), where fi is a combined feature vector. If the number of features is equal to N, then the feature comparator may calculate N different metrics between the first feature vector f1,i and the second feature vector f2,i. For example, if n=2 and f1,i=(“John”, 1956) and f2,i=(“John”, 1958), then a first metric may be calculated based on the similarity between the two strings (“John” and “John”) to be equal to 1, and a second metric may be calculated based on the difference between the two integers (1956 and 1958) to be equal to 0.83. The resulting metric function s(fi) in this example would be equal to (1, 0.83). Metrics may yield values between 0 and 1, where 0 indicates low similarity and 1 indicates high similarity. In some embodiments, metrics are not restricted to yield values between 0 and 1.
In some embodiments, the Jaro-Winkler distance is used as the metric between two strings and cosine-similarity is used as the metric between any two non-strings, such as integers and vectors of normalized dates or normalized place identifiers. Other possible metrics that may be employed within the feature comparator include but are not limited to: edit distance, affine gap distance, Smith-Waterman distance, and Q-gram distance. When comparing features, it is clear to a person of ordinary skill in the art that similarity, dissimilarity, difference, or distance can be used to calculate how similar or different the features are.
In some embodiments, the feature comparator may generate a similarity score between tree persons (e.g., TP1 and TP2) using the following equation:
where N is the number of features fi and wi is the feature weight for the i-th feature fi. The similarity score t is generated based on a weighted sum of metric function s(fi) being weighted by feature weights wi.
In some embodiments, genealogical entity resolution system 200 may modify cluster database 214 based on whether it is determined that tree persons TP1 and TP2 correspond to the same individual. In some instances, when it is determined that tree persons TP1 and TP2 correspond to the same individual, a new node connection is generated in cluster database 214 between a first node corresponding tree person TP1 and a second node corresponding to tree person TP2. In contrast, when it is determined that tree persons TP1 and TP2 do not correspond to the same individual, a node connection may be removed in cluster database 214 between a first node corresponding tree person TP1 and a second node corresponding to tree person TP2. In some embodiments, generating a new node connection or removing a previous node connection may include closing or opening an electrical switch in an electrical circuit within cluster database 214.
In some embodiments, first tree data 302-1 and second tree data 302-2 are gathered from different genealogical trees in a tree database. For example, first tree data 302-1 may include all the data contained in the tree database corresponding to a first tree person (e.g., names, birth dates and locations, marriage dates and locations, death dates and locations for the first tree person and the first tree person's parents, spouse, siblings, and children) and second tree data 302-2 may include all the data contained in the tree database corresponding to a second tree person (e.g., names, birth dates and locations, marriage dates and locations, death dates and locations for the first tree person and the first tree person's parents, spouse, siblings, and children).
In some embodiments, each of first tree data 302-1 and second tree data 302-2 include data for additional tree persons that are from different family trees but belong to the same cluster as a particular tree person in a particular familial category. As an example, first tree data 302-1 may include data for the first tree person's father as well as data for two additional tree persons that belong to the same cluster as the first tree person's father. As another example, second tree data 302-2 may include data for the second tree person's spouse as well as data for one additional tree person that belongs to the same cluster as the second tree person's spouse.
In some embodiments, genealogical entity resolution system 300 includes a feature extractor 304 for extracting a first set of features 306-1 from first tree data 302-1 and a second set of features 306-2 from second tree data 302-2. For example, feature extractor 304 may extract features for the first tree person fE1,i, features for the first tree person's father fE1,i, features for the first tree person's mother fM1,i, features for the first tree person's spouse fS1,i, features for the first tree person's children fC1,i, features for the second tree person fE2,i, features for the second tree person's father fF2,i, features for the second tree person's mother fM2,i, features for the second tree person's spouse fS2,i, and/or features for the second tree person's children fC2,i. The number of features extracted for each tree person may be the same and may be indexed by i. In some examples, i may take on values between 1 and N, where N is the number of features extracted.
In some embodiments, feature extractor 304 may extract features for the additional tree persons that are from different family trees but belong to the same cluster as a particular tree person in a particular familial category. For example, if first tree data 302-1 includes data for the first tree person's father as well as data for two additional tree persons that belong to the same cluster as the first tree person's father, feature extractor 304 may extract three different versions of features for the first tree person's father fF1,i(3). As another example, if second tree data 302-2 includes data for the second tree person's spouse as well as data for one additional tree person that belongs to the same cluster as the second tree person's spouse, feature extractor 304 may extract two different versions of features for the second tree person's spouse fS2,i(2).
In some embodiments, genealogical entity resolution system 300 includes a feature comparator 308 for generating various similarity scores based on set of features 306. Feature comparator 308 may generate a similarity vector for a pair of features 306. Or, for example, feature comparator 308 may generate an individual-level similarity vector 310 for each possible pairing of tree persons from the first genealogical tree (as well as additional tree persons that belong to the same cluster as tree persons from the first genealogical tree) and tree persons from the second genealogical tree (as well as additional tree persons that belong to the same cluster as tree persons from the second genealogical tree).
For example, feature comparator 308 may generate an individual-level similarity vector sE,i(fE1,ifE2,i) for the pairing of the first tree person and the second tree person (and additional pluralities of individual-level similarity scores for pairings of multiple versions of either), an individual-level similarity vector sF,i(fF1,i,fF2,i) for the pairing of the first tree person's father and the second tree person's father (and additional pluralities of individual-level similarity scores for pairings of multiple versions of either), an individual-level similarity vector sM,i(fM1,i,fM2,i) for the pairing of the first tree person's mother and the second tree person's mother (and additional pluralities of individual-level similarity scores for pairings of multiple versions of either), an individual-level similarity vector sS,i(fS1,i,fS2,i) for the pairing of the first tree person's spouse and the second tree person's spouse (and additional pluralities of individual-level similarity scores for pairings of multiple versions of either), and/or an individual-level similarity vector sC,i(fC1,i,fC2,i) for the pairing of the first tree person's children and the second tree person's children (and additional pluralities of individual-level similarity scores for pairings of multiple versions of either). The size (e.g., the number of values) of the individual-level similarity vectors generated for each possible pairing of tree persons may be the same and may be indexed by i, where i may take on values between 1 and N.
In some embodiments, genealogical entity resolution system 300 includes an individual-level ML model 312 for generating an individual-level similarity score 314 for each possible pairing of tree persons. In some instances, individual-level ML model 312 combines the N individual-level similarity scores for each of the individual-level similarity vectors 310 into a single, combined similarity score. For example, individual-level ML model 312 may generate an individual-level similarity score 314 for each possible pairing of tree persons from the first genealogical tree (as well as additional tree persons that belong to the same cluster as tree persons from the first genealogical tree) and tree persons from the second genealogical tree (as well as additional tree persons that belong to the same cluster as tree persons from the second genealogical tree).
For example, individual-level ML model 312 may generate an individual-level similarity score tE(sE,i) for the pairing of the first tree person and the second tree person (and additional individual-level similarity scores for pairings of multiple versions of either), an individual-level similarity score tF(sF,i) for the pairing of the first tree person's father and the second tree person's father (and additional individual-level similarity scores for pairings of multiple versions of either), an individual-level similarity score tM(sM,i) for the pairing of the first tree person's mother and the second tree person's mother (and additional individual-level similarity scores for pairings of multiple versions of either), an individual-level similarity score tS(sS,i) for the pairing of the first tree person's spouse and the second tree person's spouse (and additional individual-level similarity scores for pairings of multiple versions of either), and/or an individual-level similarity score tC(sC,i) for the pairing of the first tree person's children and the second tree person's children (and additional individual-level similarity scores for pairings of multiple versions of either).
In some embodiments, genealogical entity resolution system 300 includes a most-similar relationships identifier 316 for identifying a set of most-similar tree persons 318 from among the possible pairings of tree persons based on individual-level similarity scores 314. In some embodiments, most-similar relationships identifier 316 may identify the maximum individual-level similarity scores 320 of individual-level similarity scores 314 to identify the set of most-similar tree persons 318.
For example, most-similar relationships identifier 316 may identify the largest individual-level similarity score tE(sE,i)(1) for the pairing of the first tree person and the second tree person, the largest individual-level similarity score tF(sF,i)(1) for the pairing of the first tree person's father and the second tree person's father, the largest individual-level similarity score tM(sM,i)(1) for the pairing of the first tree person's mother and the second tree person's mother, the largest individual-level similarity score tS(sS,i)(1) for the pairing of the first tree person's spouse and the second tree person's spouse, and/or the top three largest individual-level similarity scores tM(sM,i)(3) for the pairing of the first tree person's children and the second tree person's children.
In some embodiments, genealogical entity resolution system 300 includes a selector 322 for selecting a plurality of individual-level similarity vectors 324 for the set of most-similar tree persons 318. In some embodiments, the plurality of individual-level similarity vectors 324 for the set of most-similar tree persons 318 are selected from the individual-level similarity vectors 310. In some embodiments, feature comparator 308 generates the plurality of individual-level similarity vectors 324 for the set of most-similar tree persons 318.
In some embodiments, feature comparator 308 generates a plurality of family-level similarity scores 326 for the set of most-similar tree persons 318 based on set of features 306. For example, feature comparator 308 may generate a plurality of family-level similarity scores sj(fE1,i,fE2,ifF1,i,fF2,i,fM1,i,fM2,i,fS1,i,fS2,i,fC1,i(3),fC2,i(3)) which considers the similarity of features across the various familial relationships. The number of the plurality of family-level similarity scores generated for the set of most-similar tree persons 318 may be indexed by j, where j may take on values between 1 and M.
In some embodiments, genealogical entity resolution system 300 includes a family-level feature aggregator 328 which provides an aggregate input 330 to a family-level ML model 332. Aggregate input 330 may include the maximum individual-level similarity scores 320, the plurality of individual-level similarity vectors 324 for the set of most-similar tree persons 318, and/or the plurality of family-level similarity scores 326 for the set of most-similar tree persons 318.
In some embodiments, genealogical entity resolution system 300 includes a family-level ML model 332 that determines whether the first tree person and the second tree person correspond to the same individual based on aggregate input 330. In some embodiments, family-level ML model 332 includes a first stage for determining whether the first tree person and the second tree person are a non-match (do not correspond to the same individual) or a possible match (possibly correspond to the same individual) and a second stage for determining whether the first tree person are a match (correspond to the same individual) or a plausible match (maybe correspond to the same individual).
In some embodiments, the first stage is implemented using a family match model 334, which may be a ML model such as a neural network. In some embodiments, the second stage is implemented using a family plausible model 336, which may also be a ML model such as a neural network. While family match model 334 operates on aggregate input 330, family plausible model 336 operates on a subset input 338 that includes aggregate input 330 only when a possible match is determined. As such, family plausible model 336 can be fine-tuned during training to decide between match and plausible match cases.
In reference to
After familial categories 402 have been identified, a set of features 306 may be extracted from each of the tree persons shown in
For example, in reference to
In reference to
In reference to
In reference to
In reference to
In reference to
In some instances, a user may examine training tree data 502 and create a label 550A through a computer interface indicating an individual-level similarity score based on the user's examination of training tree data 502. Label 550A is compared to output 548A to generate an error 552A, which may represent the difference between output 548A and label 550A. In some instances, individual-level ML model 312 is then modified by a modifier 554A based on error 552A. Modifier 554A may change weights associated with individual-level ML model 312 such that the predicted individual-level similarity score better approximates label 550A during subsequent inferences. This process is then repeated for multiple training tree data 502 and user-provided labels.
In some instances, a user may examine training tree data 502 and create a label 550B through a computer interface indicating whether the first tree person and the second tree person are a non-match or a possible match. Label 550B is compared to output 548B to generate an error 552B, which may represent the difference between output 548B and label 550B. In some instances, family match model 334 is then modified by a modifier 554B based on error 552B. Modifier 554B may change weights associated with family match model 334 such that predictions better approximate label 550B during subsequent inferences. This process is then repeated for multiple training tree data 502 and user-provided labels.
In some instances, a user may examine training tree data 502 and create a label 550C through a computer interface indicating whether the first tree person and the second tree person are a match or a plausible match. Label 550C is compared to output 548C to generate an error 552C, which may represent the difference between output 548C and label 550C. In some instances, family plausible model 336 is then modified by a modifier 554C based on error 552C. Modifier 554C may change weights associated with family plausible model 336 such that predictions better approximate label 550C during subsequent inferences. This process is then repeated for multiple training tree data 502 and user-provided labels.
At step 602, first tree data and second tree data are obtained. In some embodiments, the first tree data and the second tree data are obtained from a genealogical database. In some embodiments, the first tree data and the second tree data are obtained from a variety of genealogical records, user-generated content, or clustered data. In some embodiments, the first tree data corresponds to a first tree person from a first genealogical tree and the second tree data corresponds to a second tree person from a second genealogical tree.
At step 604, a set of features are extracted from the first tree data and the second tree data.
At step 606, an individual-level similarity vector is generated for each possible pairing of tree persons from the first genealogical tree and the second genealogical tree.
At step 608, a (combined) individual-level similarity score is generated for each possible pairing of tree persons in each familial category based on the individual-level similarity vector for each possible pairing of tree persons. In some embodiments, the individual-level similarity score is generated by an individual-level ML model.
At step 610, a set of most-similar tree persons from the first genealogical tree and the second genealogical tree for each familial category are identified based on the individual-level similarity score.
At step 612, a plurality of individual-level similarity vectors for the set of most-similar tree persons are provided as input to a family-level ML model.
At step 614, it is determined, using the family-level ML model, that the first tree person and the second tree person correspond to a non-match, match, or plausible match based on the plurality of individual-level similarity vectors for the set of most-similar tree persons.
At step 702, an individual-level ML model is trained. In some embodiments, the individual-level ML model is trained using user-prepared labels and/or training tree data.
At step 704, a family-level ML model is trained. In some embodiments, the family-level ML model is trained using user-prepared labels and/or training tree data.
At step 706, a first stage of the family-level ML model is trained. In some embodiments, the first stage of the family-level ML model is trained using user-prepared labels and/or training tree data.
At step 708, a second stage of the family-level ML model is trained. In some embodiments, the second stage of the family-level ML model is trained using user-prepared labels and/or training tree data.
The computer system 800 may further include and/or be in communication with one or more non-transitory storage devices 825, which can comprise, without limitation, local and/or network accessible storage, and/or can include, without limitation, a disk drive, a drive array, an optical storage device, a solid-state storage device, such as a random access memory (“RAM”), and/or a read-only memory (“ROM”), which can be programmable, flash-updateable, and/or the like. Such storage devices may be configured to implement any appropriate data stores, including without limitation, various file systems, database structures, and/or the like.
The computer system 800 might also include a communications subsystem 830, which can include without limitation a modem, a network card (wireless or wired), an infrared communication device, a wireless communication device, and/or a chipset such as a Bluetooth™ device, an 802.11 device, a WiFi device, a WiMax device, cellular communication facilities, etc., and/or the like. The communications subsystem 830 may include one or more input and/or output communication interfaces to permit data to be exchanged with a network such as the network described below to name one example, other computer systems, television, and/or any other devices described herein. Depending on the desired functionality and/or other implementation concerns, a portable electronic device or similar device may communicate image and/or other information via the communications subsystem 830. In other embodiments, a portable electronic device, e.g. the first electronic device, may be incorporated into the computer system 800, e.g., an electronic device as an input device 815. In some embodiments, the computer system 800 will further comprise a working memory 835, which can include a RAM or ROM device, as described above.
The computer system 800 also can include software elements, shown as being currently located within the working memory 835, including an operating system 840, device drivers, executable libraries, and/or other code, such as one or more application programs 845, which may comprise computer programs provided by various embodiments, and/or may be designed to implement methods, and/or configure systems, provided by other embodiments, as described herein. Merely by way of example, one or more procedures described with respect to the methods discussed above, such as those described in relation to
A set of these instructions and/or code may be stored on a non-transitory computer-readable storage medium, such as the storage device(s) 825 described above. In some cases, the storage medium might be incorporated within a computer system, such as computer system 800. In other embodiments, the storage medium might be separate from a computer system e.g., a removable medium, such as a compact disc, and/or provided in an installation package, such that the storage medium can be used to program, configure, and/or adapt a general purpose computer with the instructions/code stored thereon. These instructions might take the form of executable code, which is executable by the computer system 800 and/or might take the form of source and/or installable code, which, upon compilation and/or installation on the computer system 800 e.g., using any of a variety of generally available compilers, installation programs, compression/decompression utilities, etc., then takes the form of executable code.
It will be apparent to those skilled in the art that substantial variations may be made in accordance with specific requirements. For example, customized hardware might also be used, and/or particular elements might be implemented in hardware, software including portable software, such as applets, etc., or both. Further, connection to other computing devices such as network input/output devices may be employed.
As mentioned above, in one aspect, some embodiments may employ a computer system such as the computer system 800 to perform methods in accordance with various embodiments of the technology. According to a set of embodiments, some or all of the procedures of such methods are performed by the computer system 800 in response to processor 810 executing one or more sequences of one or more instructions, which might be incorporated into the operating system 840 and/or other code, such as an application program 845, contained in the working memory 835. Such instructions may be read into the working memory 835 from another computer-readable medium, such as one or more of the storage device(s) 825. Merely by way of example, execution of the sequences of instructions contained in the working memory 835 might cause the processor(s) 810 to perform one or more procedures of the methods described herein. Additionally or alternatively, portions of the methods described herein may be executed through specialized hardware.
The terms “machine-readable medium” and “computer-readable medium,” as used herein, refer to any medium that participates in providing data that causes a machine to operate in a specific fashion. In an embodiment implemented using the computer system 800, various computer-readable media might be involved in providing instructions/code to processor(s) 810 for execution and/or might be used to store and/or carry such instructions/code. In many implementations, a computer-readable medium is a physical and/or tangible storage medium. Such a medium may take the form of a non-volatile media or volatile media. Non-volatile media include, for example, optical and/or magnetic disks, such as the storage device(s) 825. Volatile media include, without limitation, dynamic memory, such as the working memory 835.
Common forms of physical and/or tangible computer-readable media include, for example, a floppy disk, a flexible disk, hard disk, magnetic tape, or any other magnetic medium, a CD-ROM, any other optical medium, punchcards, papertape, any other physical medium with patterns of holes, a RAM, a PROM, EPROM, a FLASH-EPROM, any other memory chip or cartridge, or any other medium from which a computer can read instructions and/or code.
Various forms of computer-readable media may be involved in carrying one or more sequences of one or more instructions to the processor(s) 810 for execution. Merely by way of example, the instructions may initially be carried on a magnetic disk and/or optical disc of a remote computer. A remote computer might load the instructions into its dynamic memory and send the instructions as signals over a transmission medium to be received and/or executed by the computer system 800.
The communications subsystem 830 and/or components thereof generally will receive signals, and the bus 805 then might carry the signals and/or the data, instructions, etc. carried by the signals to the working memory 835, from which the processor(s) 810 retrieves and executes the instructions. The instructions received by the working memory 835 may optionally be stored on a non-transitory storage device 825 either before or after execution by the processor(s) 810.
The methods, systems, and devices discussed above are examples. Various configurations may omit, substitute, or add various procedures or components as appropriate. For instance, in alternative configurations, the methods may be performed in an order different from that described, and/or various stages may be added, omitted, and/or combined. Also, features described with respect to certain configurations may be combined in various other configurations. Different aspects and elements of the configurations may be combined in a similar manner. Also, technology evolves and, thus, many of the elements are examples and do not limit the scope of the disclosure or claims.
Specific details are given in the description to provide a thorough understanding of exemplary configurations including implementations. However, configurations may be practiced without these specific details. For example, well-known circuits, processes, algorithms, structures, and techniques have been shown without unnecessary detail in order to avoid obscuring the configurations. This description provides example configurations only, and does not limit the scope, applicability, or configurations of the claims. Rather, the preceding description of the configurations will provide those skilled in the art with an enabling description for implementing described techniques. Various changes may be made in the function and arrangement of elements without departing from the spirit or scope of the disclosure.
Also, configurations may be described as a process which is depicted as a schematic flowchart or block diagram. Although each may describe the operations as a sequential process, many of the operations can be performed in parallel or concurrently. In addition, the order of the operations may be rearranged. A process may have additional steps not included in the figure. Furthermore, examples of the methods may be implemented by hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof. When implemented in software, firmware, middleware, or microcode, the program code or code segments to perform the necessary tasks may be stored in a non-transitory computer-readable medium such as a storage medium. Processors may perform the described tasks.
Having described several example configurations, various modifications, alternative constructions, and equivalents may be used without departing from the spirit of the disclosure. For example, the above elements may be components of a larger system, wherein other rules may take precedence over or otherwise modify the application of the technology. Also, a number of steps may be undertaken before, during, or after the above elements are considered. Accordingly, the above description does not bind the scope of the claims.
As used herein and in the appended claims, the singular forms “a”, “an”, and “the” include plural references unless the context clearly dictates otherwise. Thus, for example, reference to “a user” includes a plurality of such users, and reference to “the processor” includes reference to one or more processors and equivalents thereof known to those skilled in the art, and so forth.
Also, the words “comprise”, “comprising”, “contains”, “containing”, “include”, “including”, and “includes”, when used in this specification and in the following claims, are intended to specify the presence of stated features, integers, components, or steps, but they do not preclude the presence or addition of one or more other features, integers, components, steps, acts, or groups.
This application is a continuation of U.S. patent application Ser. No. 17/261,458, filed Jan. 19, 2021, which is a 371 National Stage filing of international application No. PCT/US2019/042807, filed Jul. 22, 2019, which claims benefit of U.S. Provisional Patent Application No. 62/701,322, filed Jul. 20, 2018. Each of the aforementioned applications are hereby incorporated by reference in their entirety.
| Number | Date | Country | |
|---|---|---|---|
| 62701322 | Jul 2018 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 17261458 | Jan 2021 | US |
| Child | 18603537 | US |
