An automatic solver for algebra word problems based on deep reinforcement learning

Abstract

The invention belongs to the technical field of artificial intelligence, and discloses an automatic solver for algebraic problems based on deep enhanced learning. The automatic solver for algebraic problems based on deep enhanced learning adopts a digital strategy to identify relevant numbers as the optimal expression tree Bottom layer, by determining the operator node of the number node, the expression tree is built in a bottom-up manner; the three related numbers (13, 4, 9) are extracted as the bottom layer of the expression tree. The method of the present invention is more efficient and more accurate than current state-of-the-art methods. Compared with the comparison algorithms, the framework of the present invention only needs 10 times less running time to solve an application problem while improving the accuracy. More importantly, the accuracy is improved from 45.2% to 63.7% on the most challenging benchmark dataset, verifying the generality of the DQN framework.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, and in particular relates to an automatic solver for algebraic problems based on deep reinforcement learning. Background technique [0002] The research history of automatically solving Applied Mathematics Problems (MWP) can be traced back to the 1960s, and it still continues to attract researchers' attention in recent years. MWP first maps human-readable sentences into logical forms understandable by machines, and then makes inferences. This process cannot be solved simply by pattern matching or end-to-end classification techniques, therefore, designing automatic solvers for applied mathematics problems with semantic understanding and reasoning capabilities has become an indispensable step on the road to general artificial intelligence. ARIS was proposed in "Mohammad Javad Hosseini, Hannaneh Hajishirzi, Oren Etzioni, and Nate Kushman. Learning to solve arithmetic word problems w...

AI Technical Summary

Problems solved by technology

[0003] To sum up, the problems existing in the existing technology are: the current automatic solvers for algebraic problems have not enough training data sets, the robustness is not strong, and the evaluation effect is poor. The main reason is that most methods require more manual intervention. , the versatility is not strong, and the reasoning mechanism grows exponentially with the increase of data volume

Images