| 摘要 |
PURPOSE: An information recognition engine is provided to overcome the limitation of the information processing by making the engine autonomously learn a language, automatically classify the data, and adapt to the growth and the environment change. CONSTITUTION: An NN1(Neural Network 1)(4), an MNN(Middle Neural Network)(5), and an NN2(6) are a combined neural network using the algorithms of the ART(Adaptive Resonance Theory), the SOM(Self Organized Map), the LVQ(Learning Vector Quantization), and the BAM(Bidirectional Associative Memory). The NN1(4) fetches the words and the phrases by recognizing a syllabic pattern and relation of the collected data and learn the pattern and relation. The MNN(5) builds a dictionary and a structural thesaurus by repeatedly learning the synonyms and fetches a main theme from the data based on the dictionary and the thesaurus. The NN2(6) fetches a middle theme and a minor theme from the data while compensating the main theme and structurally classifies the data based on the themes.
|
