Generalized Monge-Elkan Method for Approximate Text String Comparison

Toponym matching, i.e. pairing strings that represent the same real-world location, is a fundamental problemfor several practical applications. The current state-of-the-art relies on string similarity metrics, either specifically developed for matching place names or integrated within methods that combine multiple metrics. However, these methods all rely on common sub-strings in order to establish similarity, and they do not effectively capture the character replacements involved in toponym changes due to transliterations or to changes in language and culture over time. In this article, we present a novel matching approach, leveraging a deep neural network to classify pairs of toponyms as either matching or nonmatching. The proposed network architecture uses recurrent nodes to build representations from the sequences of bytes that correspond to the strings that are to be matched. These representations are then combined and passed to feed-forward nodes, finally leading to a classification decision. We present the results of a wide-ranging evaluation on the performance of the proposed method, using a large dataset collected from the GeoNames gazetteer. These results show that the proposed method can significantly outperform individual similarity metrics from previous studies, as well as previous methods based on supervised machine learning for combining multiple metrics.

Several tasks related to geographical information retrieval and to the geographical information sciences involve toponym matching, that is, the problem of matching place names that share a common referent. In this article, we present the results of a wide-ranging evaluation on the performance of different string similarity metrics over the toponym matching task. We also report on experiments involving the usage of supervised machine learning for combining multiple similarity metrics, which has the natural advantage of avoiding the manual tuning of similarity thresholds. Experiments with a very large dataset show that the performance differences for the individual similarity metrics are relatively small, and that carefully tuning the similarity threshold is important for achieving good results. The methods based on supervised machine learning, particularly when considering ensembles of decision trees, can achieve good results on this task, significantly outperforming the individual similarity metrics.

Proceedings of the ICSOFT 2006 - International conference on Software and Data Technologies, pp 115-120, Setubal, Portugal, 11-14 Sept. 2006.

In order to deal with the problem of semantic and schematic heterogeneity in collaborative networks, matching components among database schemas need to be identified and heterogeneity needs to be resolved, by creating the corresponding mappings in a process called schema matching. One important step in this process is the identification of the syntactic and semantic similarity among elements from different schemas, usually referred to as Linguistic Matching. The Linguistic Matching component of a schema matching and integration system, called SASMINT, is the focus of this paper. Unlike other systems, which typically utilize only a limited number of similarity metrics, SASMINT makes an effective use of NLP techniques for the Linguistic Matching and proposes a weighted usage of several syntactic and semantic similarity metrics. Since it is not easy for the user to determine the weights, SASMINT provides a component called Sampler as another novelty, to support automatic generation of weights.

CoopIS 2006

In most suggested systems aiming to enable interoperability and collaboration among heterogeneous databases, schema matching and integration is performed manually. The SASMINT system introduced in this paper proposes a (semi-) automated approach to tackle the following: 1) identification of the syntactic/semantic/structural similarities between the donor and recipient schemas to resolve their heterogeneities, 2) suggestion of corresponding mappings among the pairs of matched components, 3) facilitation of user-interaction with the system, necessary for validation/enhancement of results, and 4) generation of a proposed integrated schema, and a set of derivation rules for each of its components to support query processing against integrated sources. Unlike other systems that typically apply one specific algorithm, SASMINT applies a hybrid approach for schema matching that combines a selection of algorithms from NLP and graph theory. Furthermore, SASMINT exploits the uservalidated schema matching results in its semi-automatic generation of the integrated schema and its necessary derivations.

Despite all existing methods to identify a person, such as fingerprint, iris, and facial recognition, the personal name is still one of the most common ways to identify an individual. In this paper, we propose two novel algorithms: The first one uses sound techniques to create a multi-dimensional vector representation of names to compute a degree of similarity. This algorithm compares names that are written in different languages (cross-language) and transliterated names in English. The second algorithm builds on the first one and measures the similarities between full names, taking into account the full name structure. We evaluate the algorithms for transliterated names and across languages using Arabic and English as an example. Significant results are achieved by both algorithms compared with other existing algorithms.

Instance-based matching is the process of finding the correspondence of schema elements by comparing the data from different data sources. It is used as an alternative option when the match between schema elements fails. Instance-based matching is applied in many application areas such as website creation and management, schema evolution and migration, data warehousing, database design and data integration. Sometimes the schema information such as (element name, description, data type, etc.) is unavailable or is unable to get the correct match especially when the element name is abbreviation, therefore, if the schema matching failed, the next step is to focus on values stored in the schemas. For these reasons, many recent approaches focus on instance-based matching. In this study, we propose an approach that combines the strength of pattern recognition utilizing regular expressions for numerical domain as well with WordNet for string domain by getting the similarity coefficient in the range of [0,1]. In previous approach, the regular expression is achieved with a good accuracy for numerical instances only and is not implemented on string instances because we need to know the meaning of string to decide if there is a match or not. The using of WordNet-based measures for string instances should guarantee to improve the effectiveness in terms of Precision (P), Recall (R) and F-measure (F). This approach is evaluated with real dataset and the results are found better than using just equality measure for string especially if the schemas are disjoint. The approach achieved 95.3% F-measure (F).

2011

PRO-VE 2006

Schematic and semantic heterogeneity are two important types of heterogeneities that need to be resolved in order to enable interoperability and exchange of data among distributed and heterogeneous databases in a collaborative network of biodiversity nodes. This paper describes the SASMINT system, which performs schema matching and integration among databases. SASMINT identifies syntactic/semantic/structural similarities between two schemas as automatically as possible, resolving their heterogeneity and creating mappings among the pairs of matched components. Unlike other systems that are typically limited to specific algorithms, SASMINT combines a number of algorithms from the NLP and graph theory domains. After obtaining the user-input on validation/enhancement of matching results, SASMINT exploits the results of schema matching to automatically generate an integrated schema.

Instance-based matching is the process of identifying the correspondences of schema elements by comparing the instances of different data sources. It is used as an alternative option when the schema-based matching fails. Instance-based matching is applied in many application areas such as website creation and management, data warehousing, database design, and data integration. Many recent approaches focus on instance-based matching. In this paper, we propose an approach that utilizes WordNet-based measure for string domain by getting the similarity coefficient in the range of [0..1]. In previous approach, the regular expression is achieved with a good accuracy for numerical instances only and is not implemented on string instances because we need to know the meaning of string to decide if there is a match or not. The using of WordNet-based measures for string instances should guarantee to improve the effectiveness in terms of Precision (P), Recall (R) and f-measure (F). In this paper we implemented Lin’s measure to find the similarity of two instances. This approach is evaluated with real dataset and the results are found better than using just equality measure for string especially if the schemas are disjoint. The approach achieved 91.8% f-measure (F).

Measuring string similarity is a frequently used technique in various Language Technology (LT) applications, such as: Spell checkers, Translation Memories, Example-Based Machine Translation (EBMT) etc. In this paper experimental results on string similarity measures are presented. The main goal of the experiments is to detect the most appropriate similarity measure which can be applied for retrieving candidate sentences for translation templates to be used in an EBMT system. The advantage of this approach is that it is based entirely on surface forms, therefore being independent from any linguistic resources. The results show that token-based measures are the most suitable for translation template extraction.