Enabling intelligent search systems that can navigate and facet on entities, classes and relationships, rather than plain text, to answer questions in complex domains is a longstanding aspect of the Semantic Web vision. In this demo, we present an investigative search engine that meets some of these challenges, at scale, for a variety of complex queries in the human tra cking domain. The search engine has been rigorously prototyped as part of the DARPA MEMEX program and has been integrated into the latest version of the Domain-speci c Insight Graph (DIG) architecture, currently used by hundreds of US law enforcement agencies for investigating human tra cking. Over a hundred million ads have been indexed. We demonstrate the in-use version of DIG1, allowing a user to experience the system in real time.
1 The prototype described in the in-use paper titled `An investigative search engine for the human tra cking domain.' is not a solved problem in either the NLP or Semantic Web community; even on traditional datasets, consistently achieving F-scores above 70% for common attributes like name and location remains challenging. The second challenge is that of o ering an intuitive interface to law enforcement so that they are both comfortable using the system, and are also able to make full use of the facilities that such a system o ers. Investigative o cials, for example, are not adept even at simple query formulation in languages like SPARQL or SQL; even worse, human users are sometimes known to pose a query that is slightly di erent from what they intend. Finally, investigative o cials need to know that they can trust the system given the sensitive and resource-constrained nature of the human tra cking domain.
In our group, we have developed the Domain-speci c Insight Graph (DIG) system for addressing these challenges by semi-automatically constructing and completing, in an o ine phase, a knowledge graph containing entities, attributes, relationships and clusters over hundreds of millions of otherwise unconnected webpages. During the interactive search phase, DIG uses powerful query reformulation and ranking strategies to handle the noise in the knowledge graph and answer complex queries that are posed by the user by lling out an intuitive form on the GUI. The search engine in DIG has been rigorously evaluated against a set of competitive baselines from academia and industry by the DARPA MEMEX program, and has proven to be both competitive and e cient.
In this demo, we present an in-use prototype of DIG that is already being used by over 200 law enforcement agencies, and is currently being permanently transitioned to the o ce of the District Attorney of New York. The prototype can answer complex semantic queries over at least a hundred million HT webpages that have crawled and indexed over the last 2-3 years. DIG supports both faceted and entity-centric search, as Figures 1 and 2 illustrate.
Speci cally, we will showcase how DIG can be used to jump-start an investigation starting from a vague, under-speci ed search query. For example, a user searches for a hispanic escort in Chicago who is known to provide certain sex services in her ads. By exploring a ranked list of ads retrieved by the system, along with images, the user is able to drill down, in an entity-centric fashion, on important details like phone numbers and emails that are promising avenues for eld investigations. Acquiring such a list without the help of the system would ordinarily have taken months of eld level investigations and online searches.
Conclusion. Human tra cking is an egregious crime that has been helped, rather than hindered, by the growth of the Web. The DIG search engine attempts to use technologies from various communities, and especially the Semantic Web, in the hopes of signi cantly raising the barrier-of-entry for would-be tra ckers, by providing law enforcement with state-of-the-art tools for expediting prosecution and evidence gathering. DIG, in tandem with other tools developed under the DARPA MEMEX program, has already been used to bring several prosecutions to court in the US2 and is in the process of being permanently transitioned 2 A potent example is the recent case described in http://www.sfgate.com/crime/ article/Man-sentenced-to-97-years-in-human-trafficking-7294727.php A Semantic Search Engine For Investigating Human Tra cking to law enforcement. We will showcase the in-use version of DIG in this demo, with the ultimate hope that our system will serve as a case study of the real-world social impact that Semantic Web research can have.
Acknowledgements. We gratefully acknowledge our collaborators and all (former and current) members of our team who contributed their e orts and expertise to DIG, particularly during the dry and nal evaluation runs: Amandeep Singh, Linhong Zhu, Lingzhe Teng, Nimesh Jain, Rahul Kapoor, Muthu Rajendran R. Gurumoorthy, Sanjay Singh, Majid Ghasemi Gol, Brian Amanatullah, Craig Knoblock and Steve Minton. This research is supported by the Defense Advanced Research Projects Agency (DARPA) and the Air Force Research Laboratory (AFRL) under contract number FA8750- 14-C-0240. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the o cial policies or endorsements, either expressed or implied, of DARPA, AFRL, or the U.S. Government.