Collective entity resolution in relational data

TLDR: In this article, a relational clustering algorithm that uses both attribute and relational information for determining the underlying domain entities is proposed, which improves entity resolution performance over both attribute-based baselines and over algorithms that consider relational information but do not resolve entities collectively.

Abstract: Many databases contain uncertain and imprecise references to real-world entities. The absence of identifiers for the underlying entities often results in a database which contains multiple references to the same entity. This can lead not only to data redundancy, but also inaccuracies in query processing and knowledge extraction. These problems can be alleviated through the use of entity resolution. Entity resolution involves discovering the underlying entities and mapping each database reference to these entities. Traditionally, entities are resolved using pairwise similarity over the attributes of references. However, there is often additional relational information in the data. Specifically, references to different entities may cooccur. In these cases, collective entity resolution, in which entities for cooccurring references are determined jointly rather than independently, can improve entity resolution accuracy. We propose a novel relational clustering algorithm that uses both attribute and relational information for determining the underlying domain entities, and we give an efficient implementation. We investigate the impact that different relational similarity measures have on entity resolution quality. We evaluate our collective entity resolution algorithm on multiple real-world databases. We show that it improves entity resolution performance over both attribute-based baselines and over algorithms that consider relational information but do not resolve entities collectively. In addition, we perform detailed experiments on synthetically generated data to identify data characteristics that favor collective relational resolution over purely attribute-based algorithms.

Figures

Figure 2.6: Comparison of different relational similarity measures against Jaccard over only the affected instances in BioBase in each case

Table 4.3: Comparison between unconstrained and adaptive expansion for BioBase

Table 3.4: Comparison of LDA-ER with relational clustering(CR)

Table 3.2: Performance of LDA-ER, A and A* for CiteSeer and arXiv datasets. The standard deviation of the F1 is 3× 10−4 for CiteSeer and 1.7× 10−4 for arXiv.

Figure 4.1: Illustration of (a) identifying relation and (b) ambiguous relation using references from the running example. Dashed lines represent co-occurrence relations and dashed circles show δ-boundaries around references.

Figure 4.3: (a) Size of the relevant set for increasing expansion depth for sample queries in arXiv and BioBase (b) Execution time of RC-ER with increasing number of references

Full text

ABSTRACT
Title of Dissertation: COLLECTIVE ENTITY RESOLUTION
IN RELATIONAL DATA
Indrajit Bhattacharya, Doctor of Philosophy, 2006
Dissertation directed by: Dr. Lise Getoor
Department of Computer Science
Many databases contain imprecise references to real-world entities. For exam-
ple, a social-network database records names of people. But different people can go
by the same name and there may be different observed names referring to the same
person. The goal of entity resolution is to determine the mapping from database
references to discovered real-world entities.
Traditional entity resolution approaches consider approximate matches be-
tween attributes of individual references, but this does not always work well. In
many domains, such as social networks and academic circles, the underlying entities
exhibit strong ties to each other, and as a result, their references often co-occur in
the data. In this dissertation, I focus on the use of such co-occurrence relationships
for jointly resolving entities. I refer to this problem as ‘collective entity resolution’.
First, I propose a relational clustering algorithm for iteratively discovering entities
by clustering references taking into account the clusters of co-occurring references.
Next, I propose a probabilistic generative model for collective resolution that finds
hidden group structures among the entities and uses the latent groups as evidence

for entity resolution. One of my contributions is an efficient unsupervised infer-
ence algorithm for this model using Gibbs Sampling techniques that discovers the
most likely number of entities. Both of these approaches improve performance over
attribute-only baselines in multiple real world and synthetic datasets. I also perform
a theoretical analysis of how the structural properties of the data affect collective
entity resolution and verify the predicted trends experimentally. In addition, I mo-
tivate the problem of query-time entity resolution. I propose an adaptive algorithm
that uses collective resolution for answering queries by recursively exploring and
resolving related references. This enables resolution at query-time, while preserv-
ing the performance benefits of collective resolution. Finally, as an application of
entity resolution in the domain of natural language processing, I study the sense dis-
ambiguation problem and propose models for collective sense disambiguation using
multiple languages that outperform other unsupervised approaches.

COLLECTIVE ENTITY RESOLUTION IN RELATIONAL DATA
by
Indrajit Bhattacharya
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
2006
Advisory Committee:
Dr. Lise Getoor, Chair/Advisor
Dr. Carol Espy-Wilson, Dean’s Representative
Dr. Amol Deshpande
Dr. Philip Resnik
Dr. Marie desJardins

c
 Copyright by
Indrajit Bhattacharya
2006

Dedication
To my parents.
ii

Frequently asked questions

Q1. What are the contributions in this paper?

Traditional entity resolution approaches consider approximate matches between attributes of individual references, but this does not always work well. 

Q2. What are the future works in this paper?

Interesting directions of future research include exploring stronger coupling between the extraction and resolution phases of query processing and investigating localized resolution for offline data cleaning as well. In this chapter, I look at a potential application of entity resolution in the do- main of natural language processing and consider the related problem of word sense disambiguation. 

Q3. What is the third dataset used in the IBM KDD Challenge?

The third dataset, describing biology publications, is the Elsevier BioBase dataset2 which was used in a recent IBM KDD-Challenge competition. 

Q4. What is the secondary similarity measure for Soft TF-IDF?

Jaro-Winkler is reported to be the best secondary similarity measure for Soft TF-IDF, but for completeness, The authoralso experiment with the Jaro and the Scaled Levenstein measures. 

Q5. How many pairs are rejected by an O(n2) approach?

Apart from the scaling issue, most pairs checked by an O(n2) approach will be rejected since usually only about 1% of all pairs are true matches. 

Q6. What is the effect of naive relational approaches?

The naive relational approaches (NR and NR*) degrade in performance with higher neighborhood sizes, again highlighting the importance of resolving related references. 

Q7. How many merge operations are required to exhaust a queue that has q entries?

If the merge tree is perfectly balanced, then the size of each cluster is doubled by each merge operation and as few as O(log q) merges are required. 

Q8. What is the similarity measure for cluster pairs?

The similarity measure for cluster pairs accounts for relationships between different6references, and as a result of this, each merge operation affects similarities for related cluster pairs. 

Q9. What is the common way to resolve a sense?

As for entity resolution, The authorexplore the problem of collective sense disambiguation, where senses are resolved for multiple languages simultaneously. 

Q10. What is the effect of adding more relationships between entities?

As more relationships get added between entities, relationship patterns between entities are less informative, and may actually hurt performance. 

Q11. What is the relational40 component of the similarity between clusters?

As a result, the relational40component of the similarity between clusters would be zero and merges would occur based on attribute similarity alone. 

Q12. How do The authorcalculate the second order neighborhood Nbr2(c) for a cluster?

The authorcalculate the second order neighborhood Nbr2(c) for a cluster c by recursively taking the set union (alternatively, multi-set union) of the neighborhoods of all neighboring clusters: Nbr2(c) = ⋃c′∈Nbr(c) Nbr(c ′).