Learning expressive linkage rules from sparse data

Abstract

A central problem in the context of the Web of Data, as well as in data integration in general is to identify entities in different data sources that describe the same real-world object. There exists a large body of research on entity resolution. Interestingly, most of the existing research focuses on entity resolution on dense data, meaning data that does not contain too many missing values. This paper sets a different focus and explores learning expressive linkage rules from as well as applying these rules to sparse web data, i.e. data exhibiting a large amount of missing values. Such data is a common challenge in various application domains including e-commerce, online hotel booking, or online recruiting. We propose and compare three entity resolution methods that employ genetic programming to learn expressive linkage rules from sparse data. First, we introduce the GenLinkGL algorithm which learns groups of matching rules and applies specific rules out of these groups depending on which values are missing from a pair of records. Next, we propose GenLinkSA, which employs selective aggregation operators within rules. These operators exclude misleading similarity scores (which result from missing values) from the aggregations, but on the other hand also penalize the uncertainty that results from missing values. Finally, we introduce GenLinkComb, a method which combines the central ideas of the previous two into one integrated method. We evaluate all methods using six benchmark datasets: three of them are e-commerce product datasets, the other datasets describe restaurants, movies, and drugs. We show improvements of up to 16% F-measure compared to handwritten rules, on average 12% F-measure improvement compared to the original GenLink algorithm, 15% compared to EAGLE, 8% compared to FEBRL, and 5% compared to CoSum-P.

An Exploratory Study on Utilising the Web of Linked Data for Product Data Mining

“…Similar to product classification, typically product linking will make use of product names (e.g., Kannan et al (2011); Gopalakrishnan et al (2012);Vandic et al (2012);van Bezu et al (2015); Shah et al (2018);Tracz et al (2020); Li et al (2020)) and descriptions (e.g., Petrovski et al (2014); Ristoski et al (2018); Li et al (2020)). The difference however, is that the task also makes use of a diverse range of structured product attributes (e.g., van Bezu et al (2015); Shah et al (2018); Petrovski and Bizer (2020); Li et al (2020)), often defined as 'key-value' pairs such as those that can be extracted from product specifications (e.g., product ID, model, brand, manufacturer). Intuitively, offers that have the similar sets of key-value pairs are more likely to match.…”

“…Algorithms. Since the prediction of linking/matching of product offers depends on a notion of 'similarity', some methods will have an 'intermediary' step that converts product metadata features to similarity features (Vandic et al (2012); Li et al (2020); Petrovski and Bizer (2020)). This is typically done by applying similarity metrics -usually based on string form, or word/character distribution -to the textual feature representations of two offers.…”

Profiling Entity Matching Benchmark Tasks

“…The coverage of models learned with different combinations of attributes can significantly vary as not all attributes contribute equally to the solution of the matching task [21]. Discovering the set of attributes that encode the most-identifying information, is crucial for the extraction of more focused profiling meta-information.…”

“…Under this group fall the following tasks: phones, headphones, and tvs. The matching methods used for evaluating these tasks need to especially address the challenge of low data density [21]. Group 3: Small and Difficult.…”

BECKEY: Understanding, comparing and discovering keys of different semantics in knowledge bases

“…The data linking problem in data graphs has been the main focus of numerous studies (see [9,14] for survey), and applied in different research fields such as knowledge extraction [23,24], geospatial analysis [27], sentiment analysis [19,10], etc. Some of the existing approaches are based on expressive linking rules that can be learned from a set of existing reference links [18,16]. These rules consist of attribute-specific comparisons, aggregation functions along with different weights and thresholds.…”