Entity Recognition

"Entity" is a word for some kind of proper noun - a person, place, or thing. The "thing" could be an organization such as Lexalytics, a product, such as an iPhone, or any other type of thing you need to track. Recognizing that a word or phrase is a particular type of thing is called entity recognition. Entity recognition is different from using searches to find particular entities, because entity recognition finds them without you having to specify the particular ones you want. For instance, entity recognition allows you to ask questions like "What companies I am not currently tracking are being mentioned along with mine?"

Lexalytics provides two ways to recognize entities. We find them out of the box by ourselves, and you can also import your own dictionary of entities for us to find. Perhaps you have a list of SKUs you need to track, or you want to recognize hotel amenities (swimming pool, front desk and so on) as entities. You can import these, and you can use our query syntax (AND, OR, NOT, NEAR, WITH) to define them.

We also support normalization. This is something done to roll variations of entities together. For instance, companies are sometimes mentioned by their full legal name (Cisco Systems, Inc), sometimes by their bare name (Cisco) and sometimes by something else entirely, like their stock ticker symbol (CSCO). For reporting purposes, you want all of these to be treated as the same entity - its not really relevant if someone used the stock ticker vs the name.

Finally, you can label an entity as a different type if you so choose. Instead of using a type of "company" for your competitors, perhaps you want to label them all as "competitors" instead so you can easily find all competitors in your report.

Entity names are returned to you in the output, along with their type, their label, the normalized form, whether they were found by us or defined by you, their sentiment, themes, and the number of times they were mentioned. Sometimes multiple mentions of an entity are found and linked together. This happens a lot for people - the first mention might be President Barack Obama, but the second mention is just Obama, and the third is the pronoun "he." We link those together and report the longest mention - President Barack Obama - as the name of the entity, along with three mentions. By default we don't report each individual mention back to you in Semantria, but that can be enabled if desired.