Optimisation and Feasibility Study of APP Service Design Based on User Experience Psychology

Kano et al. extended the two-factor theory by adopting a two-dimensional quality perception model of subjective user perception and objective performance of APP services, and proposed the well-known Kano model, as shown in Figure 1.

Figure 1.

Kano model

Based on the relationship between different types of APP quality characteristics and customer satisfaction, the model categorizes CR into five different Kano categories:

1) Irrelevant-type quality (I): customer satisfaction is not affected whether this quality attribute is satisfied or not.

The Kano model categorizes functional attributes that have a greater impact on user satisfaction by identifying functional attributes of applications. The research methodology includes the following five steps: (1) Identify the functional attributes that users use in the process of using the APP through various user research methods. (2) Design the corresponding Kano questionnaire based on the functional attributes identified in the previous step. (3) Distribute and collect Kano questionnaires. (4) Statistically analyze the results of the questionnaire and categorize the attributes according to Table 1. “R”: reverse attribute, “Q”: there is a problem with this response, “I”: irrelevant attribute, “A “: charming attribute, “O”: expected attribute; “M”: required attribute. (5) After categorizing the APP attributes, optimize the APP by considering its own APP positioning, current market conditions and the categorization of each attribute.

Kano modeling, as a structural approach, allows for the study of the extent to which different functional attributes have an impact on user satisfaction. There are a wide range of applications available for managing user requirements categorization. However, the model is a qualitative research method. When determining which category an attribute belongs to by taking the maximum value, it is impossible to determine the nature of the attribute if more than one maximum value appears at the same time. In order to better categorize user needs, scholars in various countries have improved the Kano model, and Berger proposed a way to calculate the coefficient of relative user satisfaction, and use the coefficient of satisfaction to determine the categorization of functional attributes.

When the product provides this function, the calculation of the better coefficient is shown in equation (1): (1) Better(Si)=(A+O)/(A+O+M+I)$Better\left( {{S}_{i}} \right)=(A+O)/(A+O+M+I)$

When the product does not provide this feature, the Worse factor is calculated as shown in equation (2): (2) Worse(Di)=(M+O)/(A+O+M+I)$Worse\left( {{D}_{i}} \right)=(M+O)/(A+O+M+I)$

A,O,M,I represent the quantities attributed to category A,O,M,I, respectively.

In earlier sections, we showed how to use the Kano model to verify the attributes of a requirement item, which include charismatic attributes, desired attributes, required attributes, undifferentiated attributes, and reverse attributes. This leads to the question of how to confirm the order of importance between multiple functions when multiple requirement items are attributed to the same attribute. In actual design and development, we should prioritize the requirements that can efficiently improve user satisfaction.

In the Kano model, the relationship between user satisfaction and the degree of fulfillment of requirements can be expressed as equation (3): (3) s=cpk\[s=c{{p}^{k}}\]

s denotes user satisfaction, p denotes the degree of satisfaction of the requirement, and c is a constant: k is the adjustment coefficient corresponding to the different Kano attributes: required attribute, 0 < k 1, desired attribute, k = 1, charisma attribute, k > 1, non-differentiated attribute, and k = 0, reverse attribute k < 0. Although the value of k is an adjustment coefficient, based on previous empirical studies, the k values of required attributes, desired attributes, charismatic attributes, undifferentiated attributes, and reversed attributes are taken as 0.5, 1, 2, 0, and -1 in that order.The weights of the ith demand item are calculated in Equation (4): (4) wl′=(wiki)/(∑wiki)(i=1,2,⋯,n)\[{{w}_{l}}\prime =\left( {{w}_{i}}{{k}_{i}} \right)/\left( \sum\limits_{{{w}_{i}}}{{{k}_{i}}} \right)(i=1,2,\cdots ,n)\]

w_i is the initial weight of the ind software functional requirement, and k_i is the corresponding adjustment factor. The initial weights of w_i are converted from the answers to the question items in the Kano questionnaire.

To date, for various document processing tasks, TF-IDF is the most commonly used method for document analysis. It provides weights for words in a document based on two criteria: (1) how often the word occurs in the given document (TF) and (2) how rarely it is used in other documents in the corpus (IDF). The definition is as follows:

For word i in text j, the value of TF–IDF_i,j can be calculated as: (5) TF−IDFi,j=tfi,j⋅log(Ndfi+1)\[TF-ID{{F}_{i,j}}=t{{f}_{i,j}}\cdot \log \left( \frac{N}{d{{f}_{i}}+1} \right)\]

Where tf_i,j denotes the number of times word i occurs in text j, df_i contains the number of texts for word i and the total number of texts in the corpus is N. To avoid taking logarithms of zero, it will be possible IDFi,j=log(Ndfi+1)$ID{{F}_{i,j}}=\log \left( \frac{N}{d{{f}_{i}}+1} \right)$.

From Equation (5), it can be found that the TF-IDF algorithm assumes that if a word is important to a text, then it should be repeated in that text and occur rarely in other texts.

The user’s usage evaluation text can directly reflect the user’s topics of concern and interest preferences, so the user’s usage evaluation is subjected to LDA topic mining, and the user’s interest preferences and characteristics can be intuitively understood through visualization techniques.The LDA topic model can be evaluated by the degree of perplexity, and the lower the degree of perplexity is, the better the effect is [23]. The number of appropriate topics can be determined by the calculation of the degree of confusion, which is calculated as shown in Equation (6) and Equation (7), where D_test represents the test dataset with M documents, w_d represents the words appearing in document d, p(w_d) represents the probability of the occurrence of word w_d in each document, and N_d represents the number of words in document d: (6) Perplexity(Dtes)=exp{ −∑d=1Mlog(p(wd))∑d=1MNd }\[Perplexity\left( {{D}_{tes}} \right)=\exp \left\{ \frac{-\sum\limits_{d=1}^{M}{\log }\left( p\left( {{w}_{d}} \right) \right)}{\sum\limits_{d=1}^{M}{{{N}_{d}}}} \right\}\] (7) p(wd)=∑zp(z|d)p(wd|z)\[p\left( {{w}_{d}} \right)=\sum\nolimits_{z}{p}(z|d)p\left( {{w}_{d}}|z \right)\]

Sentiment analysis (SA) is the computational study of people’s opinions, attitudes, and emotions toward an entity. Entities can represent individuals, events, or themes. These themes are most likely to be covered by comments. Sentiment analysis identifies the emotions expressed in a text and then analyzes them. Thus, the goal of SA is to find opinions, identify the sentiments they express, and then categorize their polarity as shown in Figure 2.

Figure 2.

Emotional analysis process

Adverbs of degree and negative adverbs appear in different orders and express different emotions. When the combination relation is degree adverb + negative adverb + polarity word, it is expressed in letters as Int+Neg+PW, for example: especially bad, where Int means degree adverb, Neg means negative adverb, and PW means polarity word. Then its sentiment intensity is calculated as shown in equation (8), where P(Int) is the modification intensity of the degree adverb, P(Neg) is the modification intensity of the negative adverb, and G (PW) denotes the polarity coefficient of the sentiment word. For example, the intensity value of the sentiment “especially bad” is calculated as -1.7: (8) G(Int+Neg+PW)=(1+P(Int))×P(Neg)×G(PW)\[G(Int+Neg+PW)=(1+P(Int))\times P(Neg)\times G(PW)\]

When the combination relation is Negative Adverb + Negative Adverb + Polarizer, notated as Neg + Neg + PW, i.e., the Negative Adverb modifies the Negative Adverb to indicate a partial weakening of the degree of affirmation, rather than a double negation to indicate affirmation. Then its affective strength is calculated as equation (9). For example, the intensity value of the sentiment “not bad” is calculated as -0.8: (9) G(Neg+Neg+PW)=−0.8×P(Neg)×G(PW)\[G(Neg+Neg+PW)=-0.8\times P(Neg)\times G(PW)\]

When the combination is Negative Adverb + Degree Adverb + Polarity Word, it is written as Neg + Int + PW, that is, the Negative Adverb appears in front of the Degree Adverb for modification, thus reducing the degree of negation. The intensity of the sentiment is calculated as in equation (10), for example, the intensity value of the sentiment “not particularly good” is calculated as -0.3: (10) G(Neg+Int+PW)=P(Neg)×(1−P(Int))×G(PW)\[G(Neg+Int+PW)=P(Neg)\times (1-P(Int))\times G(PW)\]

According to the user’s emotional characteristics of different attribute facets of APP, the emotional portrait of the user’s attribute facets of APP is constructed, if a user’s comments on an attribute facet in the comment text are more frequent, it indicates that the user is more concerned about the attribute facet, and when the user’s emotional evaluation of an attribute facet is lower, it indicates that the user’s requirements for the attribute facet are higher and more picky. The attribute word-emotion score of the same attribute surface in all the comments of each user is summed up and divided by the total number of comments, so as to obtain the score of the user’s emotion on the APP attribute surface, such as formula (11), where g(d_i) indicates the emotion score of the ird attribute word in the dnd attribute surface of the user’s APP comments, and N indicates the number of comments of the same user. Then based on the score to determine the user’s emotional attitude towards different attribute facets, the emotional attitude is divided into positive and negative, where the emotional score is greater than 0 that is positive, the score is less than 0 that is negative, the attribute facets that are not mentioned in the comments emotional score is recorded as 0, expressed in Null: (11) F(d)=∑i=1ng(di)N\[F(d)=\frac{\sum\limits_{i=1}^{n}{g}\left( {{d}_{i}} \right)}{N}\]

The review data of an APP is analyzed by calculating the sentiment scores of different attribute feature words of the same APP, and then calculating the mean value of the sentiment scores of the attribute facet features of the same APP to indicate the scores of the attribute facet features, as shown in Equation (12), W¯(d)$\bar{W}(d)$ denotes the mean score of the APP’s sentiments in the dnd attribute facet, f(d_j) denotes the sentiment score of the jth attribute word in the dth attribute facet of all the reviews of the same APP score, M denotes the number of reviews of the same APP: (12) W(d)¯=∑j=1mf(dj)M\[\overline{W(d)}=\frac{\sum\limits_{j=1}^{m}{f}\left( {{d}_{j}} \right)}{M}\]

According to the statistical survey on China’s Internet development released by China Internet Information Center in 2019, as of June 2019, the scale of China’s Internet users reached 854 million, and the number of cell phone Internet users reached 846 million, with urban Internet users accounting for 73.7% of all Internet users, while rural Internet users accounted for only 26.3%.

In addition to the above conditions, the target users also need to meet the following requirements:

1) Users need to have the ability to take care of themselves and the ability to acquire materials, i.e., users have a certain economic basis to access the mobile Internet and use smartphones, and have a certain degree of cell phone operation ability.

In order to avoid cognitive overload during the process of using APP, it is necessary to deeply analyze the psychology of users during the process of using APP.The research on user psychology mainly relies on the two-sided nature of everything. Ancient classics have repeatedly recorded the theory of yin and yang of all things, which is born from the two sides of all things. “Lao Tzu” once said: “All things are negative yin and embrace yang, and they think they are harmonious”, yin and yang are the basic attributes of all things. Yin and Yang are both opposites and unified, and it was once said in Yi Chuan that “one yin and one yang is the way”, that is, yin and yang are one, transforming and balancing each other. The yin and yang thought of ancient Taoism emphasizes that only when things are balanced can things develop normally, which has profound guiding significance for the study of user psychology. As a user psychology, it should also be analyzed from two levels, that is, the “good” and “evil” psychology of users. The “good” and “evil” mentioned here do not refer to the fault caused, but simply based on the user’s behavior and habits, which aims to analyze the feasibility of improving the user experience by studying the user’s psychological form. User psychology is the premise of studying user behavior, and the research on user psychology is mainly discussed from two aspects, namely the “good” psychology of users and the “evil” psychology of users.

According to Table 1, Table 2, and the Better-Worse coefficient formula, the 17 elemental items of the Better and Worse indexes are calculated, and the Better-Worse coefficient graph shown in Figure 5 is obtained.

Figure 5.

Coefficient of Better-Worse

Different APP service demand types have different performance characteristics, by analyzing the differences in user demand for each index reflected in this survey, so as to propose and improve the strategy for improving APP service user satisfaction in a targeted manner. By constructing the satisfaction analysis based on the Better-Worse coefficient diagram, the following conclusions are obtained:

Based on the results of the APP service KANO questionnaire, the Better-Worse index and the service demand indicators of each APP were analyzed above, and the Better-Worse coefficient diagram was constructed into four quadrants in order to deeply analyze the impact of various online and offline APP demand indicators on user satisfaction. The Better Index (SI) is the abscissa, the Worse Index (DSI) absolute value is the ordinate, and the average value of the two indices calculated above is the origin of the coordinate system (0.581, 0.446) to construct the coordinate system, which is divided into four quadrant diagrams.

App service feasibility analysis uses subjective perception measures to gauge how feasible the app service is. In the survey, the question “How would you rate the quality of the services provided on the App (including the effectiveness, stability, controllability, satisfaction, visual design, and information framework of the services)?” (A. very poor feasibility B. relatively poor feasibility C. average D. relatively good feasibility E. very good feasibility) was measured. The value of “very poor feasibility” is assigned as “1”, and the value of “very good feasibility” is assigned as “5”, the higher the score, the more feasible the government app service is. The higher the score, the higher the feasibility of the government app service.

Figure 7 shows the distribution of the feasibility of an App service in City B. In general, the feasibility of the App service in 2022 is good. Among them, the overall feasibility of the App service is above “average” in 77.65%, 97%, 91.87%, 93.6%, 88.73% and 90.5% respectively. However, it should be noted that there is a possibility of overestimating with this estimation strategy. Given that only the ratings of all respondents who answered the question were considered here, and that there were multiple respondents who did not answer the question in any way, the subsequent approach of filling in the missing values provides an alternative answer to the question.

Figure 7.

The feasibility distribution of an app service in B city

Table 3 is a descriptive statistical analysis of the variables, and in order to facilitate the respondents’ answers and reduce sensitivity, the age band is set to the age group similar to Figure 3 as the dumb variable. The academic qualifications are divided into four groups, including “junior high school and below”, “high school or technical secondary school”, “undergraduate or junior college”, and “graduate degree and above”, which are set as dumb variants. The level of income is measured in the form of monthly income, which is also desensitized, and is divided into “3,000 yuan and below”, “3,000~6,000 yuan”, “6,000~10,000 yuan”, “10,000~20,000 yuan” and “more than 20,000 yuan”, which are set as dumb variables.

The description of the above dependent variables shows that the difference between the service feasibility of the optimized APP is not obvious between the missing value and the unfilled missing value, and the satisfaction rate is 3.645 (value range 1~5) before filling in the missing value, and it becomes 3.561 after filling in the missing value. Even so, in general, the app’s service satisfaction in 2022 is between 3 and 4, that is, between “average” and “relatively satisfied”, which shows that the optimization scheme is more feasible to improve the service quality of the app.