PRINCIPLE-BASED LANGUAGE MODEL
- Wen-Lian Hsu
The principle-based approach (PBA) is an explainable machine learning model, which combines the advantages of rule-based approach and statistical machine learning. It can automatically cluster the training set instances and summarize a principle for each cluster. Input is aligned with relevant principles to identify the closest cluster. PBA adopts alignment matching, which is more flexible and robust. Hence, it is easier to apply PBA to other applications, especially those requiring natural language understanding. For instance, PBA can be used in do biological text mining, multiple round dialog, and intelligent tutoring agents. PBA is currently under patent review in the US.
Fields of Application：
- PBA can be used to learn knowledge involved in dialog systems that demand in-depth understanding, anaphora resolution, and etc.;
- It can be used to learn natural language script and how to execute them as a software agent;
- It can be used to develop artificial intelligent software in education, such as automatic problem solving, solution explanation, blind spot resolution.
Advantages when compared to the existing technologies：
Natural language processing basically has two models: the rule-based approach and machine learning approach. PBA tries to combine the advantages of both approaches so that it can learn the patterns automatically just like machine learning, and make inferences as easily as rule-base. PBA can automatically cluster the training set instances and summarize a principle for each cluster. Input is aligned with relevant principles to identify the closest cluster. Error analysis could help reveal either the lack of knowledge or incorrect clustering. In the latter case, PBA allows human to correct the clusters. This makes PBA a more explainable machine learning model.
Contact Person：Shih-Yi Shiwe / 886-2-2787-2551 / email@example.com