Today, I started trial blog, I only heard that Blog is a notebook and diary's combination, I don't know other use, so I will use it to test my Oerm algorithm today. The problem is precise. The accuracy of Recall and F-Measure (= 1) is not increased, and the cause may be in Targeters. For a token sequence, when defining its evaluation type, you may need to contact before and after, I have been fully considering the relationship when I am doing the assessment, and now I need to define a new one if you need to contact it before and after. Ontology, like Rapier to record some rules? If I apply my results to RAPIER, I believe that the effect is similar to Rapier's effect, and the ability of the OWL record rule is very poor (but the rule description of RAPIER should be enough) Just look at the test process, card in the 99th Yes, really TMD is slow, hey, I think my dream is an SCI, but unfortunately, such a good idea, there is no condition. My solution is: 1. Try to use Rough Set's knowledge about technology to subtract (this estimate is going back) 2. After trying to get the Evaluation Type of each token in the Token sequence, further further for Evaluation Type Treatment, as if it is considered whether the knowledge collected in Targeters is inappropriate. Oh, this process is really painful. Originally wanted to write a program to automatically call, to alleviate Jena's memory leak, huh, TMD Solaries's environment variables don't know how to set up, and the command line does not allow Java so long ClassPath, depressed, but also to study, bitter There is no information in hand. Hehehe, I found a giant interesting problem, it was my program has a problem, huh, the results are as follows: Show Statistics Data: 08: 49: 27,058 Info Oermain: 635 - Correct Answers: 2008: 49: 27,058 Info Oermain: 636 - answers produced: 10608: 49: 27,059 INFO OerMain: 637 - total possible corrects: 4408: 49: 27,060 INFO OerMain: 638 - precise: 0.1886792508: 49: 27,060 INFO OerMain: 639 - recall: 0.4545454708: 49: 27,061 INFO OerMain : 640 - F-measuer: 0.2666666808: 49: 27,062 INFO OerMain: 644 - write test record to testrecord.dat08: 49: 27,196 INFO OerMain: 676 - the error results are: technicalneedslookingGroup, Inc.systemsneeded.A.Group, Inc. , PlacementTechnical Search
08: 49: 27, 199 Info Oermain: 154 - Test Spend Time: 1979s End From this result analysis, I have found 106 results, I know that it is in 10 documents. And there are 20 Correct Answers, with average of 2 per document, and the most intimate Possible Correct is 44, grandmother, horror is horrible, this is too horrible, take a look, hurry. I have to do two things now: 1. Cut the relationship between TrainProcess and TestProcess to cut 2. Why is the statistics check so much. 9:29 Discover the problem, huh, huh, low-level mistakes, it turned out to read the old record files, crazy, such a horror thing happened. Oh, be careful, write a new TestProcess_indep (), indicating that it is an independent test process, or so good. The test was completed, found that many of the cries than expected are to improve Precise, find a lot of meaningless things when calculating a token's evaluation value, cannot be determined according to the first-order correlation is not 0, but should According to all the content, the integrated assessment is determined according to the relationship, but this calculation amount will increase.
