Entity Linking with Extensible Knowledge Graphs
October 11th 2017 13:30 - 13:45
Entity linking is the task of automatically identifying named entities such as persons, companies, or products mentioned in text. For instance, in the sentence "Jack founded Alibaba in Hangzhou with investments from SoftBank and Goldman," entity linking methods would understand that in this context Jack refers to Jack Ma, founder of Alibaba, and not for example to Jack Nicholson or Jack Kerouac. It understands by linking the name to a knowledge graph like Wikidata, DBpedia, or YAGO. This deep text understanding enables the development of a whole range of next-generation applications to better search, write, and analyze unstructured texts.
Today there are many services providing entity linking. One crucial ingredient that is missing are extensible knowledge graphs. Existing entity linking services almost exclusively use knowledge graphs derived from Wikipedia and are thus not able to understand people or companies that are not part of Wikipedia. This dramatically limits the applicability of entity linking for specific applications in a corporate environment, as the entities of interest, e.g. lists of clients, partners, or products, are not part of the knowledge graph. Our Ambiverse technology allows the integration of application- or customer-specific entities, finally enabling the use of entity linking for applications in a B2B environment.
Two use cases that are enabled by extensible knowledge graphs will be presented in more detail:
KYC (Know-Your-Customer): Banks and other creditors spent a large amount of time verifying the creditworthiness of potential debtors, especially if they are companies requiring a loan. Not only do they need to assess the balance sheets and other structured data, but also need to do search the Web and other document collections to uncover unwanted connections. These connections might be obvious, for example to individuals on governmental blacklists, but might be more subtle and require extensive research. With entity linking (and extensible knowledge graphs), this research can be sped up tremendously, as relevant news items and documents can be automatically compiled, with ambiguity already resolved. Just imagine how much time it takes to google for a creditor called "Orange" when you are not interested in the fruit or the television series "Black is the new Orange."
Sales: Trust sells. As a sales person, you can establish trust by knowing the background about a potential buyer, by showing that you care. However, acquiring this background knowledge is time-consuming and thus many sales calls are not properly prepared. With Ambiverse entity linking technology, your CRM becomes your extensible knowledge graph, and background information about your clients in the form of news snippets or parts of your own internal memos will be available right at your fingertips: displayed right next to your lead or contact in your CRM system. This is only possible because entity linking resolves ambiguous names to your person of interest, filtering unwanted results. Just imagine your customer is called John Smith -- good luck finding him on Bing!