Semantic search: What exactly is it?

Semantic search is an information retrieval method in which the meaning of the search query is taken into account. By incorporating domain knowledge in the form of synonyms, similarities and knowledge models ("ontologies and taxonomies"), the search query can deliver relevant "hits" in the text with greater precision.

In contrast to a character- or word-based search, in which only letters are checked for syntactic correspondence and corresponding hits are returned, a semantic search attempts to capture and understand the meaning or sense of the query and return "meaningful" results.

For semantic searches, content is first analyzed morphosyntactically and tagged. This enables, for example, the basic form "gehen" to be evaluated as a hit in a search for "gegangen".

Knowledge models, entities and relationships

In further analysis steps, the content is annotated with general and relevant entities of domain-specific knowledge, which can then be used to identify relationships during the search. General entities can be persons, organizations or places, for example. By using regular expressions and a rule-based scripting language, further entities can be recognized and annotated. Among other things, synonyms and hyperonyms similarities are also used in the semantic search, for example "car" and "automobile" have a 100% "synonym" match. In contrast, the hyperonym "vehicle" would have an 80% match with the question about a car. Semantic research thus leads to more comprehensive and precise results.

If the organizational structure and product structure are stored in a knowledge model, these references are also taken into account in a query. If the respective countries are also stored, the results - such as patent applications - can be assigned to a map. For example, the active ingredient acetylsalicylic acid is automatically taken into account in a query for aspirin. The effect of this substance, such as analgesic or anticoagulant, can also be included.

The correlations in the knowledge model provide support beyond the initial query. They are also used to navigate the content of the results. The respective area of interest is interactively refined through this navigation. The knowledge model is also suitable for grouping results in order to create coherent overviews. This makes you aware of connections that were not immediately apparent beforehand.

Why is this important? What is the point?

The underlying knowledge model operationalizes the expertise available in an organization. Especially for professional research, it is equally important to quickly find suitable results and to comprehensively penetrate a new topic.

By taking the knowledge model into account when querying and navigating, semantic searches achieve what would otherwise only be achieved by reading and receiving: content indexing.

Through this more precise and comprehensive research, information providers offer their users significant added value and thus set themselves apart from the competition. Information within an organization is easier to find, thus reducing duplication of work. The consideration of relationships in the knowledge model also solves another problem when searching an organization's internal databases. Web-based search methods use the linking of content to determine relevance. Pages to which many other pages point are more relevant than isolated content. These relationships at document level are usually missing within organizations and can be compensated for by relationships in the knowledge model.