Mapping knowledge how to do it, which is certainly not a few words were clear. First, be sure to think first based on their own business, organized here some of the main path of knowledge map construction.
Built logical thinking
-
1, carding business, build the body: the need for a knowledge map? How kind of cost, how to achieve results? You have the capacity to build knowledge map? Whether the data, and so the team can support? If necessary, how to comb a body frame according to the business?
-
2, the body editing, business knowledge representation given frame: bulk edit Protege may be utilized to obtain a representation of knowledge representation in OWL file.
-
3, added to the body data Example: find some sample data, facilitate understanding.
Constructed in different ways
-
Construction from a top-down manner: the first body and defines the data pattern, and then added to the knowledge base entity. Use of the existing structure of the knowledge base as a knowledge base.
-
Building up from the bottom by: extracting from a number of open data link entity, select a higher confidence level is added to the knowledge base, then the top body build mode.
The key technology in the process of building
-
Generally contains five areas: knowledge extraction, knowledge representation, knowledge integration, knowledge processing, knowledge assessments
-
Through knowledge extraction techniques, knowledge elements can be extracted entities, relationships, attributes, etc. from the data disclosed in some of the semi-structured, unstructured and structured third-party database.
-
Knowledge Representation through certain elements of an effective means for knowledge representation, easy to use for further processing. Distributed knowledge representation form of integrated vector construct knowledge base, reasoning, integration and applications are of great significance.
-
Then through knowledge integration can eliminate the entities, relationships, attributes, etc. between the alleged ambiguity of terms with the fact that an object, a high-quality knowledge base.
-
Knowledge processing is further tap the existing knowledge base on the basis of implicit knowledge, to build a new body to complement the relationship, to enrich and expand the knowledge base.
-
Knowledge assessment can be carried out to quantify the reliability of knowledge, to retain a high degree of confidence, give up less confidence, effectively ensure the quality of knowledge.
-
In addition, large-scale knowledge map construction, also need to support multiple technologies: distributed storage and computing, database diagram, Figure reasoning, memory databases.
Select the data store database
-
Knowledge map storage and query language has gone through the washing of history, from RDF to OWL and SPARQL queries are gradually because of the high cost and inconvenience of use, and abandoned by the mainstream industry. Map database gradually become the main storage of knowledge map.
-
Currently used widely in the map database includes Neo4j, graphsql, sparkgraphx (includes a calculation engine), based hbase of Titan, BlazeGraph etc., various store language and query language are not the same. Practical application scenarios, OrientDB and postgresql there are a lot of applications, mainly due to its relatively low implementation cost and performance advantages.
Application reasoning and self-learning knowledge
-
In the knowledge map construction process, there are many relationships completion problems. While a general knowledge of patterns that may exist millions of the fact that hundreds of millions of entities and relationships, but still far away from completion. Knowledge Mapping is a complement to predict the relationship between entities through existing knowledge map is an important complement to the relation extraction.
-
TransE TransH conventional method and by a translation relation from the entity A to the entity B and entity to establish relationships embedded, these models simply assume that only entities and relationships in the same semantic space. In fact, a complex entity is composed by a variety of attributes, different relationships focus on different attributes of the entity, so just is not enough for them to model in one space.