Sampling labelled profile data for identity resolution

Abstract

Identity resolution capability for social networking profiles is important for a range of purposes, from open-source intelligence applications to forming semantic web connections. Yet replication of research in this area is hampered by the lack of access to ground-truth data linking the identities of profiles from different networks. Almost all data sources previously used by researchers are no longer available, and historic datasets are both of decreasing relevance to the modern social networking landscape and ethically troublesome regarding the preservation and publication of personal data. We present and evaluate a method which provides researchers in identity resolution with easy access to a realistically-challenging labelled dataset of online profiles, drawing on four of the currently largest and most influential online social networks. We validate the comparability of samples drawn through this method and discuss the implications of this mechanism for researchers as well as potential alternatives and extensions.

Multi-Source Spatial Entity Linkage

“…However, finding an appropriate similarity function that combines the similarities of attributes and decides on whether to link or not the entities is often difficult. Several works use a training set to learn a classifier [7,8,17], others base the decision on a threshold derived through experiments [9,18]. Other approaches decide the include the uncertainty of a match into the decision [19].…”

“…The problem that we need to solve is which s i , s j pairs indicate a strong similarity to be considered for a match. The related work solutions propose using a classifier [7,8,33] or experimenting with different thresholds [9,18,33]. We propose a more relaxed technique that uses Pareto optimality [34] for filtering the positive class.…”

Identity Management Systems: Singular Identities and Multiple Moral Issues

“…In fact it is possible to identify a single trend largely definitional both for the technical developments in the area of IM systems and social and ethical concerns associated with this field-the identity resolution problem. This problem has emerged as a rather innocuous and purely technical issue in the data base management and statistics as a problem of classification task whereby two or more entities (collections of attributes)-often from different databases-are matched together based on the similarity of their features (Edwards et al, 2016). This problem has also motivated the development of novel identity resolution techniques and tools assisted by the advancements in artificial intelligence.…”

Multi-Source Spatial Entity Linkage

“…However, finding an appropriate similarity function that combines the similarities of attributes and decides on whether to link or not the entities is often difficult. Several works use a training set to learn a classifier [9][10][11], others base the decision on a threshold derived through experiments [12,13]. Other approaches that deal with uncertainty are described in the survey of Magnani and Montesi [7], including probabilistic, rule-based probabilistic, fuzzy, and preference-based relationships between records as well as aggregation of multi-matches.…”

“…This is an old problem in the entity linkage community. A classifier [9,11,31] can learn the behavior of the matches and detect the positive class. However, it is difficult to obtain labeled data, especially across different sources.…”