In the era of digital information, the Internet affects the behavior and patterns of thinking in our daily life. People are increasingly aware of the importance of language learning and gradually integrate it into daily life. Online learning is a new learning method that has been developing continuously in recent ten years. Almost everyone in the intelligent era will contact it, and the field is still expanding. It has become an indispensable part of the education system.

As a mobile learning method, the English word learning application is widely popular among college students because of its portability, complexity and fragmentation. This makes the technology, marketing and users of this category tend to be stable. The same type of products is emerging one after another in the market, and the homogenization phenomenon is becoming more and more obvious.

Therefore, how to seek a new direction under such market saturation is a problem to be worthy of consideration and research at present. This paper finds that the next growth point of this kind of application is to explore the new needs of users, further improve the application functions and attract more users. Only by grasping the direction of user demand in application market research can we increase user stickiness and usage. The perfection of application function is the key to realize human-computer interaction scientifically and enrich the media of human and information interaction.

Hence, after a rigorous and comprehensive study of the user needs of such applications, this paper analyzes the user needs of English word learning applications from the perspective of user comment text mining, and adds a theoretical perspective of demand analysis, which can grasp a fairly perfect direction. The combination of text mining technology and questionnaire survey makes the research develop in depth and breadth, so as to draw a scientific conclusion. This has practical significance for the function improvement of such applications and the further improvement of user satisfaction, by helping such apps to guide and plan the direction of the application development, and show its commerciality and purpose.

The main supporting technologies of text mining are natural language processing and machine learning. Foreign studies on text mining were carried out earlier. Hu Bing et al. [2] sorted out and found that H.P. Luhn made pioneering research in this field as early as the late 1950s and proposed the idea of word frequency statistics for automatic classification. Text mining in the modern sense was put forward by scholar Feldman in 1995.

Today, with the rapid development of Internet technology, the analysis of user needs in the form of the traditional questionnaire has fallen behind business needs. Therefore, text mining technology has been widely used in the field of user demand analysis. Kreutler [3] mainly studies the individuation of consumer needs from two aspects: the relationship between consumers and product development designers, and whether product development designers can find a design scheme that adapts to user needs. Chen Lianghsuan [4] studied the perspective of how to maximize user satisfaction, and believed that user demands and design requirements in the process of product transformation and innovation could be quantified through a nonlinear fuzzy number model. Jungmok M [5] proposed the technology of snowball trend mining, which can carry out a detailed and accurate design of the whole life cycle of the product and predict related risks before the actual production design of the product. This technology has been verified in practical operation and proved to be highly feasible.

Chen Jianlin [6] believes that with the rapid development of information technology, traditional foreign language teaching elements (textbooks, content, methods, etc.) are replaced by new teaching elements (computer, Internet, information resources, information technology methods, etc.). At present, the relevant research on the combination of English learning software and teaching mainly focuses on three aspects: 1) The promotion effect of mobile software on English teaching reform. Liu Xiaoli et al. [7] discussed the prospect of the combination of the Internet, mobile devices and English teaching and its implications for college English teaching reform. 2) The influence of modern educational technology on the reform of teaching mode. Li Qianqu [8] proposed the combination of modern educational technology development and college English teaching reform to build a new model of college English teaching. 3) The educational security of learning software.

The research on text data mining is relatively late in China, but in recent years, text mining has been applied more and more widely in all walks of life. Liu Qingtang et al. [9] concluded that text mining generally includes data collection, text preprocessing (data selection and cleaning, document representation, feature selection, etc.), text mining (classification, clustering, association rule mining, etc.), and text post-processing (model evaluation and feedback, knowledge interpretation and visualization, etc.).

In terms of policy research, Yang Chengqi et al. [10] took government work reports as samples and conducted quantitative evaluation of energy and environmental protection policies based on the PMC-index model, establishing a set of scientific evaluation index system. In terms of public opinion analysis, Qiu Zeguo, He Baiyan [11] used text mining technology to analyze high-value public opinion topics in online public opinion, and used data visualization method to study the emotional tendency of netizens.

Yang Deqing [12] proposed a dynamic Kano Model construction method of user demand and built a fine-grained automatic marking model of product problems, which can accurately and effectively extract the problems existing in products. However, it is not enough to locate the charismatic needs of users, and the demand analysis is not comprehensive enough. Tan Qiang [13] took the H brand mobile phone as a case study and proposed a user demand identification method for online comment mining, which solves the problem that enterprises cannot effectively obtain user demand in the Internet era and enriches the sources for enterprises to obtain user demand information. Hu Yunbao [14] analyzed the data of users’ online comments, with the help of text mining technology and the Kano Model of analyzing users’ demands, so as to help enterprises accurately and effectively grasp users’ demands.

Jiang Ruyu [15] proposed that the interaction design principles of adult self-learning apps are: pertinence, education, consistency, interaction, emotion, motivation and individuation. Wang Xun et al. [16] proposed three functional requirements for the TCSL mobile learning platform: page design of learning resources, content design of learning resources, and APP function design, including 16 APP design principles such as the consistency principle, the principle of human-computer interaction and design of information management. Tang Shu [17] summarized the user experience design principles of knowledge payment APP from the perspective of micro-learning into five categories: matching user usage scenarios, efficiently accessing learning content, promoting multi-dimensional group co-creation, guiding the establishment of learning habits, and improving the perception of fragmented results.

After collecting and sorting out the literature, it is found that most of the current analysis of user needs is through questionnaires distribution, sentiment analysis, and demand classification of user comments. Only a few users analyze user needs by combining online review data with the questionnaire. Therefore, the combination of text mining and questionnaire survey is chosen to analyze the user needs of English word learning apps. At the same time, through the study of literature related to the design principles of online learning APP and the combination of specific principles proposed by scholars, the following demand categories are sorted out as the criteria for the classification of questionnaire indicators, as shown in Table 1.

The findings from the survey of users’ personal information are shown in Table 8. Among the 165 questionnaires collected, senior students have a larger proportion, reaching 50.3%. The proportion of freshmen is 10.91%, sophomores 15.76% and juniors 23.03%. Among the purposes of using English word learning apps, students preparing for CET-4 and CET-6 accounted for the highest proportion, reaching 76.96%.

Carry on cross-analysis. The result is shown in Figure 6.

College students use English word learning apps for the main purposes of CET-4 and CET-6 and postgraduate exams, and their emphasis gradually changes with the increase of grades. For the freshman, the importance of CET-4 and CET-6 is the highest. As the grade increases, the importance of CET-4 and CET-6 gradually decreases, but it is still the highest among the four tests listed. The trend of the degree of postgraduate entrance examination is opposite to that of CET-4 and CET-6. It becomes more and more important as the grade increases, but it still does not exceed the importance of CET-4 and CET-6.

It should be noted that after the statistics of the students who choose “other”, the purpose of using an English word learning APP for students who choose this option is just to learn and accumulate English words daily.

According to the Kano evaluation standard table, the number of users of A, O, M, I, R and Q are counted respectively, and the calculation formula is as follows:

BetterIndex = A + O A + O + M + I (1)

WorseIndex = O + M A + O + M + I × ( − 1 ) (2)

A Better index refers to the improvement in user satisfaction when a service or function is provided; A Worse index refers to the decline in user satisfaction when a product does not provide a service or feature. The calculation results of the Kano Model are shown in Table 9.

Fill the calculated Better index and Worse index into the Better-Worse index quadrant graph, as shown in Figure 7.

The first quadrant is the one-dimensional attribute. The better value is high, and the worse value is a high absolute value. The functions/services in this quadrant should be met first. Definition of words (5), example sentences (8), root affix association (9), forgetting curve (16), derivative (17), phrase (18) and word group (19), meaning discrimination (20), words downloaded offline (21), vocabulary words for free (23), search for new words (24), listening exercises(25), text translation(26), English articles reading(27), past exam papers (29), simple interface (30), word book (38), dictionary (39), these 18 services are expected attributes. Taking the index serial number as the object, the degree of satisfaction with the function and service is ranked as: 23 > 29 > 30 > 5 = 8 > 26 > 38 > 16 > 24 > 25 > 39 > 18 > 9 > 19 > 17 > 21 > 20 > 27. The more services provided by English word learning apps fall into this quadrant, the higher the user satisfaction will be. From the specific situation, most of the requirements in this quadrant belong to functional requirements. It can be seen that as an English word learning APP, function-related requirements are the requirements that users care most about. Free vocabulary, the display of real questions over the years, simple interface design, etc., will greatly improve user satisfaction and make users more loyal to the product.

The second quadrant is the attractive attribute. The better value is high, and the worse value is a low absolute value. The functions/services in this quadrant should be met first; Notes (7), import/export word book (35), these two services belong to the charm attribute. Taking the index serial number as the object, the order of function and service satisfaction is 7 > 35. There are 1 functional requirement and 1 personalized requirement in this quadrant. In addition to functional requirements, if an English word learning APP can better meet users’ personalized needs, users will be more satisfied with it. The note-taking function and word-book setting function greatly meet the needs of users. Users can customize according to different learning purposes. If word learning apps can make their services fall into this quadrant as much as possible, the appeal of the APP to users will be greatly enhanced. English listening (22) is between the attractive attribute and the indifferent attribute. This shows that users’ evaluation of this function is quite polarized. Some users believe that this function is the innovation of the APP Bubeidanci. The other part of users has no sense of this function. And whether this function exists or not, it will not change their satisfaction with the APP.

The third quadrant is the indifferent attribute. The better value is low, and the worse value is also low in absolute value. The indifferent attribute means that customer satisfaction does not change when the product provides or does not provide services falling into this quadrant. The 16 services include friends PK learning (1), checking-in learning (2), studying in teams (3), notes sharing (4), change of word meaning (6), picture aided memory (10), video aided memory (11), original movie sound (12), write words by heart (13), English learning videos (28), word lock screen wallpaper (31), APP background image (32), calendar reminders (33), gold/points rewards (34), payment for vocabulary (36), and membership system (37) are indifferent attribute. Taking the index serial number as the object, the order of function and service satisfaction is: 13 > 10 > 33 > 2 > 12 > 28 > 6 > 34 > 31 > 4 > 11 > 32 > 3 > 1 > 37 > 36. Requirements that fall into this quadrant are covered in all five types. It is worth noting that the interaction requirements all fall in this quadrant. It can be seen that when college students use English word learning apps to learn, the personal learning process is the most important, while whether they can study with friends is not important to most of them. Functional requirements in this quadrant are more “additional requirements”, that is, the need to decorate the process of memorizing words, such as image and video-assisted memory, original movie sound, etc. There is no difference for users in the services similar to the “decorating” word memory process. However, some common word memory apps often treat these services as gimmicks to gain more users. This takes time and effort, but has little impact. In terms of design requirements, the lock screen wallpaper function falls into this quadrant. In the incentive demand, the reward function of gold coins/points falls in this quadrant, indicating that it is necessary to pay attention to the method of motivating users, and not all incentives will play a positive role. What kind of incentive method should be adopted is still to be further studied by APP developers. In the personalized demand, the function of vocabulary payment and membership system falls in this quadrant, which is enough to see that freedom is the most popular. At this point, developers need to innovate their monetization models.

The fourth quadrant is the must-be attribute. The better value is low, and the worse value is the high absolute value. The must-be attribute refers to that when the product provides services falling into this quadrant, user satisfaction will not be improved; On the contrary, if the service is not provided, user satisfaction will be greatly reduced. The functions/services in this quadrant must be satisfied. The two services, British pronunciation (14) and American pronunciation (15) are the must-be attributes. Taking the index serial number as the object, the order of function and service satisfaction is 14 > 15. The two services in this quadrant are functional requirements related to word pronunciation. This shows that users attach great importance to standard English pronunciation. English word learning APP can tilt the service in this aspect as much as possible to obtain user satisfaction.

According to the general theory of Kano Model, the priority of function/service provision usually needs to be: must-be attribute > one-dimensional attribute > attractive attribute > indifferent attribute. By observing the overall results, it can be found that the correct pronunciation of words, the definition and category of words, the independent degree of reciting words and the functions derived from the whole aspect of English learning, such as listening, speaking, reading and writing, should be provided to users first. The “decoration” of the word memory process, such as picture/video-assisted memory, checking-in learning, gold rewards, etc., belongs to indifferent attribute. Users’ expectations are more rested on the process of “reciting words” service. English word learning apps should return the focus of providing services to word memory itself.

Demand attributes and demand categories of indicators are summarized, as shown in Table 10.

Better values of the LDA Model were calculated and the average value was 35.47%. The Better value of high-frequency words is calculated and the average value is 29.95%. It can be seen that the quality of the index obtained by using the LDA theme model to cluster is higher than that obtained by simple word frequency ranking. It is feasible to apply text mining technology to the design of a questionnaire index. At the same time, the indexes obtained by using text mining technology are obtained from scientific analysis of user comment information, which are the functions and services that users care most, and require APP developers to pay more attention to these functions and services.