Teaching Reform of Natural Language Processing Course Under New Engineering Background

With the vigorous development of technologies such as the Internet, big data, and artificial intelligence, the new scientific revolution poses new opportunities and challenges for current engineering education in higher education institutions^[1]. In 2017, the Ministry of Education actively promoted the construction of new engineering disciplines, striving to explore a Chinese model and experience that leads global engineering education, thus aiding the development of higher education and the construction of a strong nation. Against this backdrop, new generation information technologies such as big data and artificial intelligence have become new directions and fields for exploring the construction of new engineering disciplines^[2]. As the “crown jewel” of artificial intelligence, natural language processing has received significant attention from both academia and industry, and thus this course has gradually been introduced into the teaching curriculum of higher education institutions, becoming part of the talent cultivation system in colleges and universities related to artificial intelligence and computer science.

Natural language processing is one of the core professional courses in artificial intelligence and computer science and technology, belonging to a relatively new interdisciplinary field that encompasses knowledge from linguistics, probability statistics, machine learning, data mining, and other disciplines^[3]. The natural language processing course is highly theoretical and has a vast knowledge system. Its main content includes foundational knowledge and techniques covering language models, morphology, syntax, semantic analysis, text classification, sentiment analysis, etc. Moreover, natural language processing is also an applied technology, with research content including machine translation, information retrieval, automatic question answering, information recommendation, and other application areas^[4]. In the natural language processing course system for artificial intelligence majors, how to balance foundational theoretical knowledge and applied technology within limited class hours, and how to better enhance students’ initiative and agency in learning through teaching content and methods, are key points for professional instructors to actively discuss during teaching reform and practical exploration.

As a foundational professional course, it is quite challenging to enable students to systematically and deeply master numerous related foundational knowledge and applied technologies within a single semester of classroom teaching. If the teaching process overly emphasizes theoretical knowledge, students may find it tedious and fail to see practical application value; conversely, if only project practice is conducted without learning the foundational knowledge of natural language processing, students may face the dilemma of knowing the outcomes without understanding the reasons, failing to establish a systematic framework between the knowledge structure of natural language and the applied technology. Determining appropriate teaching content and methods, and meticulously designing the teaching model are key to ensuring effective course teaching.

The interactive practice-driven teaching model for the natural language processing course combines the teaching mode of theoretical knowledge with the practice-driven teaching philosophy, liberating students from the drudgery of theoretical course learning, achieving practice-driven theoretical knowledge learning, and guiding theoretical knowledge to provide experimental methods and technical routes for course practice, thus bridging the gap between isolated foundational knowledge of natural language processing and various practical applications. In the interactive practice-driven teaching model, students can systematically learn important theoretical foundational knowledge while also exercising their ability to solve and analyze practical problems in the field of natural language processing, truly achieving the interactive development of theoretical knowledge learning and practical training, and stimulating interest in learning this course. In specific reform measures, the interactive course system is combined with problem-oriented and comparative teaching methods to drive theoretical knowledge learning through practice, reforming aspects such as theoretical teaching design and methods, experimental teaching design, and course assessment to achieve better teaching results.

The natural language processing course system generally covers the foundational knowledge and applied technologies in natural language processing^[5]. Due to the vast knowledge system related to natural language processing, current textbooks for this course mainly consist of fragmented explanations of foundational knowledge, such as isolated discussions on Chinese word segmentation, language models, text vectorization, dimensionality reduction, classification algorithms, clustering algorithms, etc. There is no comprehensive organization of the research system of natural language processing, nor has a complete process and framework for natural language processing been formed, leading to difficulties for students in learning, failing to form a systematic and organized understanding of the knowledge in each chapter, and not knowing the role of these isolated pieces of knowledge in the natural language processing process, nor being able to establish a systematic framework between the knowledge structure of natural language and applied technology. To achieve the goal of interactive practice-driven teaching reform, the author sorts out the teaching objectives and course positioning, selects teaching content, and formulates the content and framework of theoretical and practical teaching (as shown in Figure 1).

In the course teaching system shown in Figure 1, for foundational knowledge such as linguistics, word segmentation, and named entity recognition, the main teaching methods are teacher lectures and classroom discussions; based on this, a combination of lectures and practice is adopted, allowing students to learn text vectorization representation methods and distributed representation methods, then comparing traditional text classification with CNN-based modeling methods, and showcasing and discussing the role and function of text representation in system classification modeling; finally, the teacher and students together explore the applied technologies of natural language processing, stimulating students’ interest in learning and enhancing their sense of achievement.

2.1 Interactive Course System

In the teaching design, driven by intelligent software system development, related theoretical knowledge learning and project practice are linked together based on project development practice, forming a course teaching system oriented towards project development and multi-knowledge point interaction, such as using “text classification” and “sentiment classification” system modeling as the main thread, integrating foundational knowledge such as word segmentation and part-of-speech tagging, feature selection, text vectorization, etc., with classification algorithms and training models. Through this course, students can establish a comprehensive systematic knowledge framework in natural language processing, mastering foundational knowledge while also training their ability to model intelligent software systems using machine learning methods, and cultivating teamwork awareness during the project development process.

2.2 Problem-Oriented Teaching

In the theoretical course teaching of natural language processing, a problem-oriented teaching method^[6] is adopted, using various problems in the field of natural language processing as starting points for learning, allowing students to actively seek solutions to these issues; during this process, the teacher acts as the question proposer, course designer, and result evaluator, aiming to stimulate students’ initiative in learning, thereby increasing their participation in the teaching process and igniting their thirst for knowledge. For example, in the explanation of the “Chinese word segmentation” content, the teacher uses a problem-guided teaching method to help students quickly understand what automatic word segmentation is, how to perform automatic segmentation of Chinese vocabulary, what resources are needed to support it, what algorithms can be used for segmentation, and what key issues to pay attention to during the design of segmentation algorithms. In the course teaching design for “text classification,” the teacher asks: How do humans recognize animals like birds, tigers, cats, and dogs in a photo library? How are library book materials encoded and classified for storage? How does an email system filter spam? By answering similar questions, students can quickly clarify how humans perform classification or recognition, actively summarizing that intelligent models classify based on the characteristics of objects, and then use the identified features for object classification. Through problem-oriented teaching, students can understand the concept of text classification and actively summarize that the most important processes and tasks in text automatic classification are feature extraction and selection, and subsequently use various machine learning algorithms to establish text classification models.

2.3 Comparative Teaching

Due to the development of machine learning technologies, especially the widespread application of deep learning technologies in various fields of artificial intelligence, different solutions have emerged for various problems in the field of natural language processing. The development history of natural language processing has roughly gone through rule-based methods and statistical machine learning modeling methods. In designing the natural language processing course, the author largely follows the trajectory of technological development for continuous in-depth exploration. For instance, when explaining classic “text classification,” different text classification modeling schemes are compared and analyzed in a connected manner. First, the methods and processes of text classification modeling based on traditional machine learning techniques (including feature extraction and selection methods, data dimensionality reduction techniques, etc.) are explained, and on this basis, traditional classification algorithms are used for modeling. As a method comparison, the end-to-end text classification methods based on deep learning are explained next, exploring currently popular distributed text representation methods (such as Word2Vec, Paragraph2Vec, etc.), and then using deep neural networks for data modeling; at the same time, when selecting deep neural networks, the advantages and disadvantages of different neural network architectures in text classification modeling are compared, such as how recurrent neural networks and convolutional neural networks uniquely handle text vectors compared to image and video input data. Students are guided to think about what the input of a CNN network in NLP could be; if the input is a word vector, how to address the issue of varying text lengths. Through teaching that compares different modeling methods and processes for the same natural language application issue, students can connect previously learned scattered knowledge points, forming a clear knowledge system in natural language processing.

2.4 Practice-Driven Teaching

During the practical course teaching process, several practical projects can be designed to connect the foundational knowledge of natural language processing with development skills^[7], allowing students to experience and deepen their understanding of the natural language processing technologies and algorithms learned in theoretical courses, such as in constructing a “text classification” project, driving students to comprehensively utilize text vectorization techniques, feature extraction and selection, data dimensionality reduction, and various machine learning algorithms to train classification models, enhancing students’ depth of understanding of relevant foundational knowledge in natural language processing and their application skill levels. Through project practice, students’ initiative, agency, and creativity can be fully utilized, training their ability to autonomously discover, analyze, and solve problems.

In summary, through the flexible application of the above teaching designs and methods, students are effectively guided from mere textbook knowledge learning to autonomous exploration and academic research directions, fostering preliminary scientific literacy in students committed to graduate studies, while also igniting scientific research interest and enthusiasm in some students, helping senior undergraduate students gradually understand that university course learning is not only about acquiring foundational knowledge from textbooks but also involves a certain degree of scientific exploration and innovation awareness, actively exploring and creating knowledge and technology in the unknown world, contributing their strengths to the technological development and advancement of the nation and humanity.

Project-driven practical teaching is a significant feature of this course system design. The natural language processing course has strong theoretical and practical components, placing high demands on students, requiring them not only to master solid foundational theoretical knowledge of natural language processing but also to possess practical capabilities for constructing intelligent information processing systems^[8]. Through the experimental components of this course, students can organically integrate isolated foundational theoretical knowledge with project practice skills, alleviating the tedium of learning theoretical knowledge. The experimental content and objectives designed for this course are shown in Table 1, with four experimental projects arranged, and the difficulty of each project progressively increased, setting three levels: beginner, intermediate, and advanced, ensuring that the vast majority of students can complete the beginner experimental content with relative ease, while providing higher challenge goals for students with surplus learning capacity, enhancing their sense of fulfillment and satisfaction in participating in experimental projects.

This course combines classroom discussions, course experiments, and course papers for assessment.

(1) Classroom discussions: Student interaction, content presentation, and project demonstrations in class are used as scoring criteria.

(2) Course experiments: This course includes four experiments: Chinese word segmentation experiment, text vectorization experiment, text classification experiment based on traditional machine learning methods, and sentiment analysis experiment based on convolutional neural networks or recurrent neural networks. Each experimental project requires students to write detailed experimental reports, enhancing their ability to solve natural language problems using the knowledge acquired.

(3) Course papers: Based on the content of classroom teaching, several topics for natural language processing research tasks are proposed, requiring students to select a topic and complete a course paper.

To verify the effectiveness of this course assessment, statistical analysis of students’ assessment results over the past three years has been conducted. For undergraduates, the proportion of students with final scores between 80 and 90 is 79.21%, and the proportion of students scoring above 90 reaches 10.15%; for master’s students, the proportion of students with final scores between 80 and 90 is 38.24%, and the proportion of students scoring above 90 reaches 61.76%. Statistical data from recent years indicate that this teaching and assessment scheme is suitable for the learning and assessment of natural language processing courses for undergraduate and graduate students in higher education institutions. The theoretical teaching and project practice are integrated throughout the teaching process, and both have been well linked, allowing students to not only master basic theoretical knowledge but also to develop practical capabilities for developing natural language systems.

The interactive practice-driven teaching method for the natural language processing course liberates students from the tedious learning of foundational knowledge, bridging the gap between isolated foundational knowledge of natural language processing and practical applications, effectively stimulating students’ enthusiasm and initiative to learn, allowing them to form a systematic knowledge framework in natural language processing. This teaching model excellently cultivates students’ ability to design and develop intelligent natural language systems, enhancing their professional competence and achieving good teaching results.

References:

Funded Project: Chongqing Higher Education Teaching Reform Research Project “Artificial Intelligence Major Characteristic Course ‘Natural Language Processing'” (193188); Chongqing Municipal Education Commission Key Project of Graduate Education Teaching Reform Research “Research on Quality Evaluation System for Graduate Students in Computer New Engineering Majors under the Background of Innovation-Driven Strategy” (yjg192035).

First Author Introduction: Zhang Yihao, male, Associate Professor at Chongqing University of Technology, research interests include recommendation systems and natural language processing, [email protected].