Research And Implementation Of Key Technologies In Mathematical Natural Language Processing

Natural language processing(NLP)is an intersection subject of computer science and linguistics,and an important branch in the field of artificial intelligence.With the development of computers in recent years,natural language processing has played an increasingly important role to help people solve practical problems.Chinese natural language processing is an important part of the NLP field.With the continuous enhancement of China's comprehensive national strength and the continuous improvement of its international status,As a popular language,Chinese has attracted more and more attention from the world.Chinese has some problems and difficulties like other languages,such as “word sense disambiguation” and “multi-part-of-speech”.At the same time,there are some unique problems such as “automatic word segmentation”in Chinese NLP.How to make computers solve these difficulties in Chinese natural language processing and accurately understand Chinese text requires continuous exploration by researchers.Elementary mathematics text is a special branch of general language text.On the one hand,it is a standardized language description,with the main sentence structure such as subject,predicate and object;on the other hand,it is a mixed language text,and mostly a combination of Chinese and English.Therefore,many NLP tasks in mathematics texts have not only the common parts of general texts,but also the characteristics of mathematics.The exploration of natural language understanding on elementary mathematics texts is a difficult but challenging task.It puts forward higher requirements for many NLP tasks.In this thesis,we discuss a best practice of natural language processing in elementary mathematics text comprehension.Starting from the existing problems of natural language understanding and related technologies,we decompose the natural language processing of mathematics into many small parts,and conduct research and implementation of the key parts.Based on the LTP model,we have made enhancements in the word segmentation task and part-of-speech tagging task in the field of mathematics.On the issue of named entity recognition,we propose a ”type upgrade” strategy based on first-order predicate logic,and design a Chinese entity boundary recognition model based on deep learning,which improves the effect of named entity recognition on mixed-language text.The hybrid model we finally implemented can not only meet the requirements of high recall rate,but also realize fast error correction for online negative examples,which can be used to solve natural language understanding problems in elementary mathematics texts.