 Before the break. So now, instead of doing this relation instruction complete by then, that is kind of hard, and the outcome is OK, but perhaps it's not solved the problem. So if you have a text, the exact knowledge base is not perfect, it's not really extracting all the possible relation that I have in text. So these graphs are perhaps not the better to be used in my lab. So we developed another approach, more simple, to exploit the semantic information. So the idea is to create a knowledge graph from a text document, and so this knowledge graph is based on many other things. And I define here the different kind of graphs, the knowledge-based graph, the graph of entities, and the semantic language graph. So I will explain now the difference between these graphs and then in the next slide, we will see how to apply this to my lab. So we now have a knowledge base. This is a knowledge base, like dvp, for example, and we do a query, and we get information, so we build a knowledge graph with the information in the knowledge base. So the knowledge base is also a graph, but we can take a subset. For example, all the music information that is dvp, related to music, and build this knowledge graph with the information there. So for example here, we call this entity dvpdia and some of the entities dvpdia, they have some information, like the band member, general, hometown, record label. So we can build a graph with that. We have the entities, we call some of the relations, names, the predicates of the relation, and the entities that they are related to. What is defined? We plot them in the form of a graph. So OK, this knowledge base can also be constructed with the process of relation structure that Luis was playing before, so we can have an knowledge base extracted from text and then build a graph from that. So OK, but if instead of doing this process, we can also just apply entity linking to the documents and then create what we call a graph of entities. So this is, for example, two documents, the biography of Hugo and the biography of Sombol. And we have these sentences from that. So the idea here is applying the dvpnking first. So this is then the entity find different entities and then build a graph with this. So first, we have the notes of the document subjects. So Hugo and Sombol, and then we have all the entities that were identified in the text and connect them to these notes. So the entities in Nullco are at the Latin rock, the Ferrer and the Tupelo, and Sombol is American, the Ferrer and the Latin country. So we build this graph. This is the entity graph of entities. So in this way, we can have a set of documents, just apply entity linking, and build a graph like this. So we have these two approaches, and we also can combine a graph of entities with information from the knowledge base, okay? What we call a semantically enriched graph. So this is the graph of entities, but as we have used entity linking, we have identified all these entities in a reference knowledge base. So we can get some information from this reference knowledge base to enrich the graph. For example, and the Tupelo and the Ferrer are entities that weren't detected in these documents, and they are in Wikipedia, and we have information from Wikipedia of these entities. So we can expand this graph and more labels and more notes with the information from Wikipedia. So in this way, we can enrich the semantic graph and build this semantic representation of a document. So this is what we are going to use for NIM. This is another example of a semantic language graph. We are in this case from Free Sound. Free Sound is a website where users can upload songs and add a description to the sounds. So what we need is we apply entity link to the description of the sounds, and then we got this is a sound and this is another sound. We have the two words that were identified as entities, the Wikipedia or the Wikipedia of songs, and then from Wikipedia, we extract more information, the categories of these entities, and the product categories. So in this way, from at the beginning, in the other way, we just have this. If you do this, at the end you have all this information described in the item.