arXiv subjects

Mayank Kejriwal

Publications and source records attributed to Mayank Kejriwal.

62 records · Page 4Linked to original sources

arXiv2017

Using Contexts and Constraints for Improved Geotagging of Human Trafficking Webpages

Extracting geographical tags from webpages is a well-motivated application in many domains. In illicit domains with unusual language models, like human trafficking, extracting geotags with both high precision and recall is a challenging problem. In this paper, we describe a geotag extraction framework in which context, constraints and the openly available Geonames knowledge base work in tandem in an Integer Linear Programming (ILP) model to achieve good performance. In preliminary empirical investigations, the framework improves precision by 28.57% and F-measure by 36.9% on a difficult human trafficking geotagging task compared to a machine learning-based baseline. The method is already being integrated into an existing knowledge base construction system widely used by US law enforcement agencies to combat human trafficking.

cs.AI↗

arXiv2017

Supervised Typing of Big Graphs using Semantic Embeddings

We propose a supervised algorithm for generating type embeddings in the same semantic vector space as a given set of entity embeddings. The algorithm is agnostic to the derivation of the underlying entity embeddings. It does not require any manual feature engineering, generalizes well to hundreds of types and achieves near-linear scaling on Big Graphs containing many millions of triples and instances by virtue of an incremental execution. We demonstrate the utility of the embeddings on a type recommendation task, outperforming a non-parametric feature-agnostic baseline while achieving 15x speedup and near-constant memory usage on a full partition of DBpedia. Using state-of-the-art visualization, we illustrate the agreement of our extensionally derived DBpedia type embeddings with the manually curated domain ontology. Finally, we use the embeddings to probabilistically cluster about 4 million DBpedia instances into 415 types in the DBpedia ontology.

cs.CL↗

arXiv2017

Information Extraction in Illicit Domains

Extracting useful entities and attribute values from illicit domains such as human trafficking is a challenging problem with the potential for widespread social impact. Such domains employ atypical language models, have `long tails' and suffer from the problem of concept drift. In this paper, we propose a lightweight, feature-agnostic Information Extraction (IE) paradigm specifically designed for such domains. Our approach uses raw, unlabeled text from an initial corpus, and a few (12-120) seed annotations per domain-specific attribute, to learn robust IE models for unobserved pages and websites. Empirically, we demonstrate that our approach can outperform feature-centric Conditional Random Field baselines by over 18\% F-Measure on five annotated sets of real-world human trafficking datasets in both low-supervision and high-supervision settings. We also show that our approach is demonstrably robust to concept drift, and can be efficiently bootstrapped even in a serial computing environment.

cs.CL↗

arXiv2016

An Ensemble Blocking Scheme for Entity Resolution of Large and Sparse Datasets

Entity Resolution, also called record linkage or deduplication, refers to the process of identifying and merging duplicate versions of the same entity into a unified representation. The standard practice is to use a Rule based or Machine Learning based model that compares entity pairs and assigns a score to represent the pairs' Match/Non-Match status. However, performing an exhaustive pair-wise comparison on all pairs of records leads to quadratic matcher complexity and hence a Blocking step is performed before the Matching to group similar entities into smaller blocks that the matcher can then examine exhaustively. Several blocking schemes have been developed to efficiently and effectively block the input dataset into manageable groups. At CareerBuilder (CB), we perform deduplication on massive datasets of people profiles collected from disparate sources with varying informational content. We observed that, employing a single blocking technique did not cover the base for all possible scenarios due to the multi-faceted nature of our data sources. In this paper, we describe our ensemble approach to blocking that combines two different blocking techniques to leverage their respective strengths.

cs.AI↗

arXiv2016

Self-contained NoSQL Resources for Cross-Domain RDF

Cross-domain knowledge bases such as DBpedia, Freebase and YAGO have emerged as encyclopedic hubs in the Web of Linked Data. Despite enabling several practical applications in the Semantic Web, the large-scale, schema-free nature of such graphs often precludes research groups from employing them widely as evaluation test cases for entity resolution and instance-based ontology alignment applications. Although the ground-truth linkages between the three knowledge bases above are available, they are not amenable to resource-limited applications. One reason is that the ground-truth files are not self-contained, meaning that a researcher must usually perform a series of expensive joins (typically in MapReduce) to obtain usable information sets. In this paper, we upload several publicly licensed data resources to the public cloud and use simple Hadoop clusters to compile, and make accessible, three cross-domain self-contained test cases involving linked instances from DBpedia, Freebase and YAGO. Self-containment is enabled by virtue of a simple NoSQL JSON-like serialization format. Potential applications for these resources, particularly related to testing transfer learning research hypotheses, are also briefly described.

cs.DB↗

arXiv2016

Experience: Type alignment on DBpedia and Freebase

Linked Open Data exhibits growth in both volume and variety of published data. Due to this variety, instances of many different types (e.g. Person) can be found in published datasets. Type alignment is the problem of automatically matching types (in a possibly many-many fashion) between two such datasets. Type alignment is an important preprocessing step in instance matching. Instance matching concerns identifying pairs of instances referring to the same underlying entity. By performing type alignment a priori, only instances conforming to aligned types are processed together, leading to significant savings. This article describes a type alignment experience with two large-scale cross-domain RDF knowledge graphs, DBpedia and Freebase, that contain hundreds, or even thousands, of unique types. Specifically, we present a MapReduce-based type alignment algorithm and show that there are at least three reasonable ways of evaluating type alignment within the larger context of instance matching. We comment on the consistency of those results, and note some general observations for researchers evaluating similar algorithms on cross-domain graphs.

cs.DB↗

arXiv2015

A DNF Blocking Scheme Learner for Heterogeneous Datasets

Entity Resolution concerns identifying co-referent entity pairs across datasets. A typical workflow comprises two steps. In the first step, a blocking method uses a one-many function called a blocking scheme to map entities to blocks. In the second step, entities sharing a block are paired and compared. Current DNF blocking scheme learners (DNF-BSLs) apply only to structurally homogeneous tables. We present an unsupervised algorithmic pipeline for learning DNF blocking schemes on RDF graph datasets, as well as structurally heterogeneous tables. Previous DNF-BSLs are admitted as special cases. We evaluate the pipeline on six real-world dataset pairs. Unsupervised results are shown to be competitive with supervised and semi-supervised baselines. To the best of our knowledge, this is the first unsupervised DNF-BSL that admits RDF graphs and structurally heterogeneous tables as inputs.

cs.DB↗

arXiv2015

On the Complexity of Sorted Neighborhood

Record linkage concerns identifying semantically equivalent records in databases. Blocking methods are employed to avoid the cost of full pairwise similarity comparisons on $n$ records. In a seminal work, Hernandez and Stolfo proposed the Sorted Neighborhood blocking method. Several empirical variants have been proposed in recent years. In this paper, we investigate the complexity of the Sorted Neighborhood procedure on which the variants are built. We show that achieving maximum performance on the Sorted Neighborhood procedure entails solving a sub-problem, which is shown to be NP-complete by reducing from the Travelling Salesman Problem. We also show that the sub-problem can occur in the traditional blocking method. Finally, we draw on recent developments concerning approximate Travelling Salesman solutions to define and analyze three approximation algorithms.

cs.CC↗