Joint Learning Of Named Entity Recognition And Relation Extraction Based On CRF

NowadayswebtechnologyisadvancingfromWeb2.0toWeb3.0. ComparedwithWeb2.0, the main progress of Web3.0 is that the information in internet does not onlyexist in a form of data, but also knowledge. Information extraction becomes a populartopic in such a background.Usually we use a pipelined model, that is to say, firstly named entities are rec-ognized, then relation extraction is applied based on the recognized named entities.However, there is a disadvantage of such an approach: if the named entity is not recog-nized correctly, the result of relation extraction cannot be correct. Therefore, the resultof relation extraction is limited by the result of named entity recognition.In many state of art reasearches, people combine two related tasks to set up ajoint learning approach. Especially if there exists strong correlation between the twotasks, the performance of both tasks benefits from joint approach.Named entities are important to extract relations. Accurate relation classifica-tion helps recognize named entities. Therefore, we present a joint approach of namedentity recognition and relation identification. The identified relation is utilized to im-prove named entity recognition. The method has been applied to identify the namesof persons and organizations and five relations between them. The result shows thatthe joint approach has improved the recall and F-measure of named entity and relationextraction without scarifying the precision.