Consider a quite new field of research such as mine- normalization of Tweet->(converting noisy text and slang to the standard English), Where the performance is measured by its accuracy namely BLEU score.
When I can concern a result for publishing?
1- The achieved accuracy score should be higher than all reported scored in previous publications. Even there is a report on 98% accuracy.
2- The achieved accuracy score should be higher than some of previous reported results.
3- The achieved accuracy score should be higher than some of recent reported results.
4. your system can work and the output is understandable- the system is functional.
5- The achieved accuracy is quite lesser than others but there are some strength in your system such as low computational usage.
6- Any other idea?
Thanks a lot Dr. Ian for the answer. Do you think it is not a good idea say a PhD student pick a topic, which has a supervised solution with 97% accuracy, and try to obtain an unsupervised/semi-supervised approach with about 80% accuracy?
If you just want the certificate, then 97% probability of passing is a good probability.
If you want to be attain self-sufficiency and later start up your own research group, then more independence is required, and a lightly supervised probability of passing of 80% is the way to go.
I would say when you believe that you have new results that contribute to the field of study and can withstand the rigour of peer-review. So before one publishes it is a good practice to bounce one's ideas or findings off a critical and knowledgeable reviewer to determine the weaknesses in one's determinations. Then answer them until the remaining questions have been answered.
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publications