Object Identification with Attribute-Mediated Dependences

Abstract

Abstract. Object identification is the problem of determining whether different observations correspond to the same object. It occurs in a wide variety of fields, including vision, natural language, citation matching, and information integration. Traditionally, the problem is solved separately for each pair of observations, followed by transitive closure. We propose solving it collectively, performing simultaneous inference for all candidate match pairs, and allowing information to propagate from one candidate match to another via the attributes they have in common. Our formulation is based on conditional random fields, and allows an optimal solution to be found in polynomial time using a graph cut algorithm. Parameters are learned using a voted perceptron algorithm. Experiments on real and synthetic datasets show that this approach outperforms the standard one.

Swoosh: a generic approach to entity resolution

“…If this is the case, the record will be merged with the one pointed by the corresponding entry in the P fi hash table. If not, the feature values of the record are compared to those of the records in I (lines [21][22][23][24][25][26][27][28][29][30][31][32][33][34], and if no match is found, the record is inserted in I . As for R-Swoosh, when a match is found, the old records buddy and currentRecord are purged, while the merged record is placed in I for processing.…”

“…Finally, there has also been a great amount of research on non-pairwise ER, including clustering techniques [27,3,12], classifiers such as Bayesian networks [37], decision trees, SVM's, or conditional random fields [33]. The parameters of these models are learned either from a (hopefully representatitive) set of labeled example, possibly with the help of a user [31], or in an unsupervised way [39,12].…”

Reasoning about record matching rules

“…Rules for matching are studied in [3,5,6,11,20,23,29,28,31]. A class of rules is introduced in [20], which can be expressed as relative keys of this paper; in particular, the key used in Example 1.1 is borrowed from [20].…”

“…A class of constant keys is studied in [23], to match records in a single relation. Recursive algorithms are developed in [6,29], to compute matches based on certain dependencies. The AJAX system [18] also advocates matching transformations specified in a declarative language.…”

Refining Instance Coreferencing Results Using Belief Propagation

“…This assumption does not hold for scenarios where relevant data is distributed between different instances, which are related to each other. Thus, approaches, which analyze relations between data instances of different classes, have received significant attention in recent years (e.g., [3], [7], [8], [9]). One algorithm focusing on exploiting links between data objects for personal information management was proposed in [3], where the similarities between interlinked entities are propagated using dependency graphs.…”