- 學位論文
使用簡化法和計算概念框架於中文數學文字題解題機器人
Incorporating Reduction-based Approach and Frame into Chinese Math Word Problem Solver
摘要
隨著人工智慧近年來的蓬勃發展，在教育方面也期以智慧型教學系統協助孩童的學習。本研究面對自然語言理解的挑戰，以設計中文數學文字題(Chinese Math Word Problem, CMWP)的智慧型解題機器人為例，提出具解釋性的新穎模型架構，包含兩種方法：「簡化法（Reduction-based Approach）」和「計算概念框架（Frame）」。 簡化法，在中文數學文字題中作為語言分析的模型，對文句進行一系列的學習機制，歸納出其重要概念以理解句意。計算概念框架，則是利用前者的結果，歸納整理成更高一層次的「句型」，以簡化各種複雜且多變的自然語言表達。此流程可以完整記錄題意理解所需要的計算概念框架元件，並可藉由這些元件，進行題目的分類、動詞蘊涵推論、解題乃至於計算過程的解釋。 本研究提出中文數學文字題的智慧型解題機器人所需具備的模組，對於類型多元的題目區分出兩大類別，擬定相對應的解題策略。有別於現今十分熱門的深度學習模型，更能解釋計算過程。
並列摘要
With the vigorous development of artificial intelligence in recent years, it is also expected that intelligent tutoring systems will be used to assist children's learning in education. Faced with the challenge of natural language understanding, this thesis proposes a novel and explanatory model architecture, including two methods: "Reduction-based Approach" and "Frame". And we apply them to build an Chinese Math Word Problem (CMWP) solver. Reduction-based Approach (RBA), used as a language analysis model in CMWP, can conduct a series of learning mechanisms for sentence, and summarizes its important concepts for sentence meaning understanding. Frame is based on the results of RBA and generalize into a higher-level "Sentence Type" to simplify various and complex natural language expressions. This process can completely record the frame components required for the semantic understanding of the problems. And we can use these components to execute the problem classification, verb inference, problem solving and explanation of the calculation process. This thesis proposes the modules required by the Chinese math word problem solver, distinguishes multiple types of problems to two categories with its corresponding problem-solving strategies. Different from the deep learning models, it can explain the calculation process better.