Semantic search & matching
Entering documents manually into forms is laborious and costly and the process does not scale well. Automated document understanding removes this bottleneck, and to a large extent makes form filling a thing of the past.
But why do we want to have data in forms or structured databases in the first place? The main reason for this is that we want to have access to this information in an effective way. We want to be able to search, filter and analyse the data and not just on how many times a certain word occurs. Search has become main stream. But domain specific search depends on a domain specific semantic layer across documents in order to be able to express complex queries about the relationship between concepts. Exactly the type of searches which are common in HR.
Why is searching in better in Textkernel’s system? Because it is semantic search, and because we take the person doing the searching very seriously. The system judges the relevance of a document based on a sophisticated model of understanding what you mean, rather just what keywords you type as a query. But the technology would not be a useful tool without equal attention to the user experience. Transparent. Easy to use. Powerful at its core. And always fast and relevant.
By semantic matching we mean finding two objects together that make a good fit. For example a job and a person. The underlying technology for searching and matching is the same. The difference is that matching is less of an interactive process by the user, and more a component which performs searches fully automatically as part of some business process. The main advantage of our Match! solution is its ability to understand unstructured documents both at the query level as well as at the search index level. This means that it can take a very large collection of unstructured jobs and person profiles and produce shortlists of matches for each item in the set.
And the more criteria you can take into account, the more relevant the match results will be. If you are matching only on one or two keywords or criteria, you are likely to get a low satisfaction from the user. Formulating a query with many interacting criteria is not only time consuming it is also difficult for most non-technical users. A matching engine, in our view, is supposed to contain the knowledge that helps you with this. This involves parsing a job to understand the requirements, but also knowledge of synonyms, ontologies, logical career steps, travel distance, soft criteria and the importance of different criteria with respect to each other.
So semantic matching is search, plus document understanding, plus domain knowledge. It is automatic, but also transparent, and the user remains in control. Machine learning allows us to optimise matching systems by learning from feedback that users are generating in their workflow systems, by applying for jobs, rejecting or inviting candidates for interviews, and most importantly hiring the best ones.