A Graph-Based Approach to Learn Semantic Descriptions of Data Sources

Abstract

Abstract. Semantic models of data sources and services provide support to automate many tasks such as source discovery, data integration, and service composition, but writing these semantic descriptions by hand is a tedious and time-consuming task. Most of the related work focuses on automatic annotation with classes or properties of source attributes or input and output parameters. However, constructing a source model that includes the relationships between the attributes in addition to their semantic types remains a largely unsolved problem. In this paper, we present a graph-based approach to hypothesize a rich semantic description of a new target source from a set of known sources that have been modeled over the same domain ontology. We exploit the domain ontology and the known source models to build a graph that represents the space of plausible source descriptions. Then, we compute the top k candidates and suggest to the user a ranked list of the semantic models for the new source. The approach takes into account user corrections to learn more accurate semantic descriptions of future data sources. Our evaluation shows that our method produces models that are twice as accurate than the models produced using a state of the art system that does not learn from prior models.

A Scalable Approach to Learn Semantic Models of Structured Sources

“…We compared our new approach with the previous work [8] to show that the new approach scales better. In this experiment, we used k = 1 because the previous work was not able to take into account more than one semantic type for each source attribute.…”

“…This work has two major contributions compared to our previous work on learning semantic models of data sources [8]. First, we consider uncertainty in learning the semantic types: we assume that for each source attribute we are given a set of candidate semantic types along with their confidence values rather than a single, correct semantic type.…”

“…The central component of our method is a directed weighted graph G built on top of the known semantic models and expanded using the semantic types T and the domain ontology O. The algorithm we use to construct G is described in detail in our past work [8] and here we only provide a brief explanation.…”

“…Although Karma remembers new semantic type labels assigned by the user, it does not learn from the structure of previously modeled sources. This prompted our recent work [8] to exploit the known semantic models to learn a semantic model for a new unknown source. However, this approach assumed that each source attribute was labeled with the correct semantic type and failed to scale to sources with a large number of attributes.…”

“…In our previous work [8], we presented a novel approach to learn semantic models of data sources from known semantic models, semantic models of sources that have already been modeled. The work is inspired by the idea that different sources in the same domain often provide similar or overlapping data and have similar semantic models.…”