Study on Digitalization and Communication of Brand Image of Guangdong, Hong Kong and Macao Greater Bay Area

The world’s three first-class Bay Area economies have become important vehicles for promoting global economic growth and regional economic liberalization. New York Bay Area is located in the northeastern part of the United States, not only is the “economic center” of the United States, but also in the global commercial and financial fields play a huge influence, Manhattan is known as the world’s “financial heart” [1-3]. San Francisco Bay Area is an important center of science and technology innovation and science and culture in the United States and the world, gathering global high-quality talent, capital, technology and other resources, leading the direction of global development [4-5]. Tokyo Bay Area is located in the east-central part of Japan, is the world’s strong economic strength of the Bay Area and the world’s important financial, shipping, manufacturing center, while the cultural heritage of the Tokyo Bay Area is also a world-class treasure [6-7]. As the three Bay Areas compete for the world’s high-quality resources and form a new form of economy, Guangdong, Hong Kong and Macao need to continuously develop their own resources, give full play to their regional characteristics, and actively participate in the international arena [8-10].

The Bay Area is also the business card of a country, and building an effective brand image is not only conducive to the local people’s clear perception of the existence of the Guangdong-Hong Kong-Macao Greater Bay Area, but also helps to improve the international competitiveness of the Guangdong-Hong Kong-Macao Greater Bay Area and the country [11-13].

Digital transformation brings extensive opportunities for brand image building in the Guangdong-Hong Kong-Macao Greater Bay Area. By enhancing brand awareness, optimizing brand communication channels, strengthening brand competitiveness, improving brand service quality, and promoting brand internationalization, digital transformation can comprehensively improve the overall image and market competitiveness of Guangdong, Hong Kong, and Macao Greater Bay Area brands, and provide a strong impetus for achieving high-quality economic development [14-16]. The Guangdong-Hong Kong-Macao Greater Bay Area should seize the opportunity of digital transformation, actively promote the digitalization process in various fields, and build an internationally competitive brand image [17-18].

China’s research on urban image began after the reform and opening up, and the research path can be divided into, from the study of building and creating the objective form and environment of the city, to the publicity and promotion of the city to create a city brand to form a communication effect, and then to the formation of the digital era, the participation of all people in high-frequency, interactive, diversified and other characteristics, to a kind of “Netflix mode” of the city image. The construction of the city image of “Netflix mode”, the current stage of the relevant research also shows a sharp increase in the trend [19-20].

In this paper, we use crawler software to capture travelogues about Guangdong, Hong Kong and Macao Greater Bay Area as research samples, and weight the selected text data features by TF-IDF method. The brand personality feature words of Guangdong, Hong Kong and Macao Greater Bay Area are obtained by using statistical-based feature extraction method. The brand personality feature words are divided into dimensions, and the prominent brand personality of Guangdong, Hong Kong and Macao Greater Bay Area is determined by relying on the word frequency analysis method. The semantic network is constructed, and the Stanford syntactic analyzer is used to conduct tourists’ sentiment analysis based on EDT. Combining the brand personality and tourists’ emotions, the digitalization strategy for shaping the brand image of Guangdong, Hong Kong and Macao Greater Bay Area is proposed.

2

Overview of key technologies

2.1

Feature selection and weighting of text data

2.1.1

Document frequency

Document frequency (DF) statistical analysis of document sample data, through the calculation of the number of documents containing a certain feature item and the total number of document samples to get the ratio of the relationship between the number of documents and the total number of document samples, and in this way to determine whether the feature item can effectively distinguish a class of document data from other categories of document data. The formula is shown in equation (1).

(1) $D F (t_{i}, c) = \frac{\lg (x (t_{i}))}{1 - (X_{t})}$ where $x (t_{i})$ is the number of documents containing feature term $t_{i}$ in the data with document sample category c, and $X C$ is the total number of document samples with category c. The pre-set threshold value is used to distinguish the document containing the feature item from other categories of document data in the whole document sample. If the document frequency value of the requested feature item is higher than the pre-set threshold and has a high DF value only in one category, it means that the feature item has appeared many times in the documents of this category, which can distinguish different document categories well.

Document frequency based on the number of times the feature term appears in different document categories simplifies the process of calculating statistics, and has outstanding advantages in coping with the huge size of the original data. However, there is still room for improvement in counting the importance of the feature terms themselves. For example, a feature term is important for the whole document sample, but it is not used many times in the whole document sample or in a single document, and the result of the document frequency calculation is lower than the pre-set threshold, which makes the DF method mistakenly think that the feature term is not a differentiated term and filter it out, which leads to the missing of important text data.

2.1.2

Mutual information

The mutual information (MI) method is a measure calculation to compare the degree of correlation or association between two random variables. The formula is shown in (2).

(2) $I (t, c) \approx \log \frac{A N}{(A + C) (A + B)}$

Where $A$ is the number of documents in category $c$ where feature word $t$ occurs, $B$ is the number of documents in other categories except category 5 $c$ where feature word $t$ occurs, $C$ is the number of documents in category $C$ where feature word $t$ does not occur, and $N$ is the sum of the number of documents in all categories.

Although the mutual information method can calculate the correlation between two random variables very well, there are some special cases as follows.

When the text contains feature words whose word frequency is not very high, but they themselves are very critical in the text, then we will find that the value of the results calculated using the mutual information method will be high. In this case, we adopt the method of sorting the word frequencies of all the feature words in the text according to a predetermined order, and then carry out the calculation of the mutual information method through the sorted data, which can alleviate the calculation error caused by the above problem to a certain extent.

2.1.3

Feature weighting

The TF-IDF weighting method, which uses this method to determine whether certain feature items in a document are important, is that if a word or word appears very often in one document category and less often in other document categories, then the word or word has some ability to distinguish between different types of documents in the text set, and should have a higher weight than other words in the vector space model. In short, the importance of a word or words in a document largely depends on its frequency of occurrence in the text set. The calculation formula is shown in (3).

(3) $w_{i j} = t f_{i j} \times i d f_{i j} = t f_{i j} \times \log_{2} \frac{N}{n_{i}}$

Where $t f_{i j}$ represents the frequency of feature item $t_{i}$ in text $j$ , $n_{i}$ represents the number of texts containing feature item $t_{i}$ , $i d f_{i j}$ represents the frequency of inverse documents of feature item $t_{i}$ , and $N$ represents the number of texts in the whole corpus.

The overall idea of TF-IDF weighting algorithm is simple and easy to use, and the calculation method is relatively simple, but there is a defect in this approach, which is based on the frequency of a feature item in the text to determine the importance of the feature item, and does not take into account the position of the feature item located in the text, and the loss of the contextual semantics of the sentence in which the feature item is located.

2.2

Statistics-based text feature extraction

Early textual feature extraction methods were similar to early textual representation models in that they were statistically based. One of the more commonly used mutual information methods comes from information theory, in the field of information theory mutual information refers to the relationship between information, and in feature extraction refers to the statistically independent relationship between words and categories, and the calculation of mutual information $M I (t_{i}, C_{j})$ between feature $t_{i}$ and category $C_{j}$ is shown in equation (4).

(4) $M I (t_{i}, C_{j}) = \log \frac{P (t_{i} | C_{j})}{P (t_{i})} = \log \frac{P (t_{i}, C_{j})}{P (t_{j}) \cdot P (C_{j})}$

In Equation (4), $P (t_{i})$ is the probability of feature $t_{i}$ appearing in the dataset, $P (C_{j})$ is the probability of category $C_{j}$ appearing, and $P (t_{i}, C_{j})$ is the probability of feature $t_{i}$ appearing belonging to category $C_{j}$ . When the frequency of features appearing only in a certain category is high, then this feature has a close relationship with that category and the mutual information is high, this idea is similar to the inverse document frequency. However, when calculating the mutual information, the low-frequency features of a category in the corpus have higher mutual information values than the high-frequency features, and the high-frequency key features may be ignored. Moreover, when the distribution of corpus categories is unbalanced, a certain feature has higher mutual information value in small categories, and such an extraction method will affect the feature’s ability to characterize the categories.

The information gain method, also from information theory, refers to how much useful information the features can bring to the category, and the more useful information contained in the features, the greater the information gain. The amount of information can be understood as the entropy of the category, and the information gain is the difference between the information entropy when the feature exists and when it does not exist. Assuming that corpus $C$ contains $n$ categories, respectively $C_{1}, C_{2}, \dots C_{n}$ , for feature $t$ , the information set of the corpus is computed by considering $t$ and not considering $t$ , and the information set of the corpus is the sum of information considering feature $t$ , which is computed as shown in Equation (5).

(5) $H (C) = \sum_{i = 1}^{n} P (C_{i}) \log_{2} P (C_{i})$

In Equation (5), $P (C_{i})$ is the probability of occurrence of category $C_{i}$ . The conditional entropy for distinguishing different categories within the condition that the feature is fixed to $T$ is the case where feature $t$ is not considered, as shown in Equation (6).

(6) $H (C | T) = P (t) H (C | t) + P (\bar{t}) H (C | \bar{t})$

In Eq. (6), $t$ means the feature appears and $t$ means the feature does not appear. The information gain of this feature is the difference between the two entropies in equations (5) and (6) above, as shown in equation (7).

(7) $I G (T) = H (C) - H (C | T)$

According to the size of feature information gain, the key features of the text can be extracted. Information gain methods are often applied to the task of classification of corpus as a whole, Liu et al. proposed a semantic feature extraction method based on information gain in order to achieve problem classification, text semantic features are generated by the improved information gain and sequence pattern mining methods, and then the selected features are input into a classifier to classify the problem, and experiments have proved that the problem classification algorithms based on information gain have achieved advanced results.

Also, information gain is the main principle behind the operation of certain decision trees, Singer et al. proposed an ordered decision tree model which uses a new weighted information gain ratio to select categorical attributes in the tree. The model can be used as part of an expert system for ordered classification applications such as health status monitoring, portfolio classification and service system performance evaluation. It has been demonstrated that information gain methods perform well for textual semantic feature extraction, but when the distribution of positive and negative samples in the corpus is unbalanced, too many negative samples result in the vast majority of the samples not containing features, and the results of the classification are determined by the features that do not appear in the majority of the samples, and the modeling work is much less effective.

Statistical-based feature extraction methods also include word frequency methods, expected cross entropy methods, and genetic algorithms. Word frequency methods reduce the feature dimension by removing low-frequency words, assuming that low-frequency words have less impact on the semantics of the text, but in many cases this assumption does not hold true, so the use of wordon methods is limited. Expectation cross entropy method calculates the probability distribution distance of the topic class in the presence and absence of a certain vocabulary, and calculates the influence of vocabulary on the topic. Genetic algorithms regard text feature extraction as a superior evolutionary process of population reproduction, and finally obtain the optimal feature expression.

The above traditional methods usually obtain the relationship between features and categories through probability, and use the calculated feature weights to extract key features. This type of method requires a huge and comprehensive dataset to obtain all the useful features, which is more difficult to realize in practical applications. The features extracted from the limited dataset may not be the key features necessary for classification, and the key features are ignored due to their low frequency of occurrence. Therefore, statistical-based text semantic feature extraction methods cannot realize the complete extraction of key features, and focusing on text key semantics is the focus of feature learning methods.

2.3

EDT-based text sentiment analysis

When conducting sentiment analysis, bag-of-words models ignore the relevant associations between words often resulting in poor accuracy of sentiment analysis. And the direct analysis of syntactic dependencies without differentiation in some methods is easy to introduce the interference of thematically irrelevant sentiment. In order to overcome these shortcomings, this paper draws on the idea that dependency grammar realizes sentence frames with verbs as the architecture to analyze the structure of sentence sentiment expression, and defines the basic structure of such sentiment expression as the sentiment dependency tuple (EDT). Taking the real thematic feature word contained in the sentence as the center word of EDT, the other components of EDT are extracted through predefined rules to construct a complete emotion discrimination model based on emotion dependency tuple. Such an analysis process can well avoid the influence of the above two problems, closely follow the theme for sentiment analysis, accurate calculation, and overall can realize chapter-level text sentiment tendency discrimination.

The process of sentence sentiment discrimination based on sentiment dependency tuple is mainly divided into three steps: (1) syntactic analysis of the sentence to generate syntactic tree and dependency relationship; (2) extracting sentiment dependency tuple from the dependency tree and dependency relationship generated by the sentence; (3) establishing a comprehensive sentiment analysis model for the sentiment dependency tuple contained in the whole sentence, and carrying out the sentence sentiment tendency discrimination.

2.3.1

EDT extraction

The dependency relations and syntactic analysis tree generated by syntactic analysis are a kind of structured data, on which the information extraction can obtain knowledge more accurately and improve the performance of the information extraction system. This paper adopts Stanford syntactic analyzer, syntactic analysis before the use of the Chinese Academy of Sciences participle (NLPIR) first for the participle, in order to ensure the accuracy of the syntactic analysis, the user lexicon has been expanded, and did not deactivate the word filtering. Take the sentence “The reporter also found that many stockholders with a better mindset are very optimistic.” As an example, the result of participle and lexical annotation is: “The reporter/NN also/AD found/VV a lot/CD mentality/NN better/JJ’s/DEG stockholders/NN very much/AD optimistic/VA”.

The basis and key to establish a sentiment analysis model is to accurately and comprehensively extract sentiment tuples, and the following is the detailed extraction process: 1)

Constructing the center word set $T$ : The center word of the emotion-dependent tuple should be able to elicit an emotion verb, or carry an emotion noun or pronoun, and adjectives can also be used as the center word in some sentences where the subject is omitted; if the center word is derived from the topic feature set, only the emotion-dependent tuple related to the topic can be extracted, which is helpful to exclude the interference of other emotion factors. In this paper, we directly extract the words that meet the lexical requirements from the sentence as the center word, i.e., T={reporter, discovery, mindset, stockholder}.

2)

Purify the set of center words: for each word in $T$ , if the word is a leaf node of the syntactic analysis tree then remove the word from $T$ , because these words in the leaf nodes cannot exist alone, there must be words that dominate them, i.e., they have belonged to the other sentiment-dependent tuples. After the traversal T = {found, stockholder}.

3)

Extracting the modifier components of the center word: This paper summarizes the matching framework of real words as center words in the Real Word Collocation Dictionary of Modern Chinese and the extraction rules statistically from the experimental data. The specific method is to extract the modifier components of the center word by matching according to the rules in the sibling node where the center word is located and all the subtrees of the sibling node. For example, three pairs of modifiers like (center word, modifier) can be extracted from the center word “stockholders”: (stockholders, mentality), (stockholders, better), and (stockholders, optimistic).

4)

Extracting degree and negative dependency: Extract the negative dependency and degree dependency of the center word and modifier from the dependency relations of the sentence, and extract the dependency relations advmod (optimistic-10, very-9) and nummod (mentality-5, many-4).

According to the above steps, we complete the extraction of the sentiment dependency tuple of a sentence, in which the tuple with the center word “stockholder” can be represented as [[many [mind] [optimistic] stockholder [very [optimistic]]] according to the matching model.

2.3.2

EDT-based Sentiment Discriminant Modeling

The sentiment polarity and intensity of a sentence are determined by the number and polarity of the sentiment dependent tuples contained in the sentence, and we establish a sentiment analysis model to discriminate the sentiment tendency of a sentence, and the specific algorithm steps are designed as follows: 1)

For each sentiment dependency tuple, set the original polarity $P r i o r P o l a r i t y (C w) = 1$ of the center word and perform the following operations.

2)

Query the sentiment dictionary to obtain and set the sentiment polarity $P r i o r P o l a r i t y (C w)$ of each center word CW, with positive polarity 1, negative polarity -1, and elemental polarity is used when there is no polarity.

3)

Obtain for each central word its modifier EW, set its primitive polarity $P r i o r P o l a r i t y (E W) = 0$ if the modifier exists, and obtain for each modifier its sentiment polarity from the sentiment lexicon; if the central word does not dominate any modifier, perform 5).

4)

Initialize each degree and negation modification of a modifier to 1, i.e., $M o d i f i e d P o l a r i t y (E W) = 1$ , and then obtain degree dependencies and negation dependencies from the dependency set of the sentence. For each acquired degree dependency, multiply $M o d i f i e d P o l a r i t y (E W)$ by the degree coefficient (first build the degree word list). For each negative dependency, make $M o d i f i e d P o l a r i t y (E W) = - M o d i f i e d P o l a r i t y (E W)$ . After analyzing each dependency in turn, the degree of negation of each modifier is obtained.

5)

Calculate the degree of negativity ModifiedPolarity (CW) for each center word according to the method in 4).

6)

Calculate the sentiment polarityPolarity (EDT) of the whole sentiment dependency tuple, the polarity of both the center word and the modifier is superimposed by the original polarity and the modifier polarity, so the sentiment of the tuple is calculated by the formula: (8) $P (e d t) = P P (c w) * M P (c w) * [\prod_{i}^{n} P P (e w_{i}) * M P (e w_{i}) + 1]$ where $n$ is the number of modifiers of the center word, $P$ denotes Polarity, MP is Modified Polarity, and PP is Prior Polarity.To ensure that when the center word of the EDT has no modifiers or the modifiers are not tendentious, the polarity of the EDT is determined by the polarity of the center word and to make the formula general, we add 1 to Eq.

7)

The sentence sentiment value is derived by summing the sentiment values of each sentiment-dependent tuple of the sentence, which is calculated by the following formula: (9) $S c o r e (S e n) = \sum_{j = 1}^{n} P (e d t_{j})$ where $n$ represents the number of EDTs contained in sentence $s e n$ .

The sentiment analysis model based on sentiment dependency tuples integrally considers the center word with and without modifiers, considers negation and degree as a whole, and can superimpose the calculation of multi-layer negation and degree relations to ensure consistency with the actual sentiment value in terms of polarity and intensity.

3

Research design and data collection

3.1

Overview of the study area

The Guangdong-Hong Kong-Macao Greater Bay Area, covering 11 cities including Guangzhou, Shenzhen, Zhuhai, Foshan, Jiangmen, Zhongshan, Huizhou, Zhaoqing, Dongguan, Hong Kong, and Macao, is a frontier region for China to build a modernized socialist powerhouse and demonstrate the exuberant vitality of the market economy, and its tourism industry is perennially at a leading level in the country.

Although the cities in the Guangdong, Hong Kong and Macao Greater Bay Area differ greatly in their economic development methods and values, they are rich and colorful in culture, integrating Lingnan, Guangfu, Hakka, overseas Chinese, and Hong Kong and Macao extra-territorial cultures.In 2016, the tourism revenue of the Greater Bay Area reached one trillion yuan; in 2017, the Greater Bay Area Cities Tourism Alliance was established and adopted its constitution; in 2019, the Outline of the Plan for the Development of the Guangdong, Hong Kong and Macao Greater Bay Area was released, regional cooperation in tourism in the Greater Bay Area is in good shape.

3.2

Data collection and cleaning

In this study, two websites with a relatively high concentration of travelogue sharing, namely, GoWhere.com and Toutiao.com, were used as data sources, and crawler software was applied to crawl the travelogues of the tourist places in the Greater Bay Area from January 1, 2019, to December 31, 2023, and 963 travelogues of the Greater Bay Area were finally obtained after excluding the guide category, the pure picture category, and the same travelogues published by the same author. In order to ensure the accuracy of the final analysis results, the final obtained 963 travelogues were reviewed, and the text such as navigation, advertisement push, address, official website introduction, etc. were removed from the crawled webpages.

4

Data analysis

4.1

Comparison of brand personality dimensions

4.1.1

Brand Personality Characteristics of Major Cities

Using the statistical-based text feature extraction method, 14 brand personality feature words were extracted and categorized into four dimensions: internationalization, high speed, diversification, and livability. Five popular cities, Guangzhou, Shenzhen, Zhuhai, Hong Kong and Macau, were selected as the main research objects and analyzed in conjunction with the overall travelogues of Guangdong, Hong Kong and Macao Greater Bay Area.

The number of screened travelogues mentioning Guangzhou is 761, Shenzhen 513, Zhuhai 468, Hong Kong 793, and Macau 681. By checking these travelogues one by one, it was found that the number of travelogues containing brand personality words were 221, 183, 109, 215 and 195 respectively. The word frequency of the brand personality feature words contained in each city was statistically derived, and the statistical results of the brand personality dimensions of major cities and the Guangdong-Hong Kong-Macao Greater Bay Area are shown in Table 1.

Table 1.

Brand personality dimension statistics

Region	Internationalization		High speed		Diversification		Livable		Total
	n	%	n	%	n	%	n	%	n	%
Guangzhou	115	1.28	88	0.98	138	1.54	122	1.36	463	5.17
Shenzhen	156	1.74	163	1.81	77	0.86	70	0.78	466	5.20
Zhuhai	91	1.02	56	0.62	51	0.57	98	1.09	296	3.30
Hong Kong	164	1.83	155	1.73	79	0.88	60	0.67	458	5.11
Macau	143	1.60	132	1.47	81	0.90	53	0.59	409	4.57
Total	669	7.48	594	6.63	426	4.76	403	4.49	2092	23.35
Guangdong-Hong Kong-Macao Greater Bay Area	2576	28.76	2189	24.44	2261	25.24	1965	21.94	8958	100.00

Through statistics, it is found that: 1)

There are a total of 8,958 words used to describe the brand personality traits of the Guangdong-Hong Kong-Macao Greater Bay Area in the travelogues. According to the descriptions of the travelogues, the brand personality traits, in descending order of frequency of occurrence, are: internationalization (28.76%) > diversification (25.24%) > high speed (24.44%) > livability (21.94%). Therefore, this paper argues that the prominent brand personality of the Guangdong-Hong Kong-Macao Greater Bay Area can be identified as “internationalization”, “diversification”, “high-speed” and “livability”.

2)

For the five cities screened according to the official publicity channels, the word frequencies of their brand personality words that can be included in the Guangdong-Hong Kong-Macao Greater Bay Area Brand Personality Dimension Scale totaled 2,092, accounting for 23.35% of the total number of words captured. Among them, the contribution rate of “internationalization” was 7.48%, the contribution rate of “high-speed” was 6.63%, the contribution rate of “diversification” was 4.76%, and the contribution rate of “livability” was 4.49%. Guangzhou’s contribution to diversification is the highest among the five cities, followed by Macau. This can indicate that the perceived brand personality of travelers is more for the Guangdong-Hong Kong-Macao Greater Bay Area as a whole, and a few are for individual attractions.

3)

“Internationalization” accounted for the highest proportion in Hong Kong and Macao, “high-speed” accounted for the highest proportion in Shenzhen, “diversification” accounted for the highest proportion in Guangzhou, and “livability” accounted for the highest proportion in Zhuhai.

4)

10 of the 14 brand personality words are common to 5 cities, and the statistics of brand personality words in each city are shown in Table 2, that is, at least two cities share the characteristic words. There are 3 words that appear only in one city, accounting for 8.52%. This shows that the brand personality characteristics of these five cities in the Guangdong-Hong Kong-Macao Greater Bay Area have certain commonalities, but there are also great differences. The three unique characteristic words are “historical”, “artistic”, and “refreshing”. Among these three words, “refreshing” only appears 9 times in Zhuhai, and the other 2 words only appear in Guangzhou reviews, so it can be considered that the brand personality of Guangzhou in these five cities is more prominent than the other 4 cities, and it is different from the other 4 cities with a unique brand personality. The word “rustic” is not found in these five cities.

Table 2.

Statistics of brand personality words of each city

Dimensionality	Feature word	Guangzhou	Shenzhen	Zhuhai	Hong Kong	Macau
Internationalization	World-class	42	41	28	56	51
	World-leading	15	42	25	41	39
	Globalized	24	33	21	31	22
	Open	34	40	38	36	31
High speed	Efficient	21	53	22	49	58
	Energetic	22	61	18	51	31
	Modernize	45	49	16	55	43
Diversification	Inclusive	75	43	32	38	26
	Cross-cultural	15	34	19	41	55
	Historical	32	0	0	0	0
	Artistic	16	0	0	0	0
Livable	Leisurely	122	70	89	60	53
	Rustic	0	0	0	0	0
	Refreshing	0	0	9	0	0

4.1.2

Analysis of brand personality dynamics

In order to explore whether the brand personality of Guangdong, Hong Kong and Macao Greater Bay Area has changed over time, this paper counts the number of collated travelogue texts by year, and takes the words captured in each year as the base data, classifies them by year, and then conducts statistical analysis, and the results of the brand personality vocabulary statistics of Guangdong, Hong Kong and Macao Greater Bay Area in the past years are shown in Table 3.

Table 3.

Statistics of brand personality words over the years

Year	Internationalization		High speed		Diversification		Livable		Total
2019	281	26.02	155	14.25	352	32.60	292	27.04	1080
2020	392	27.41	264	18.46	489	34.20	285	19.93	1430
2021	588	33.75	367	21.07	301	17.28	486	27.90	1742
2022	650	27.25	684	28.68	562	23.56	489	20.50	2385
2023	665	28.25	719	30.54	557	23.66	413	17.54	2354

From Table 3, it can be seen that the brand personality dimensions of each year are ranked in order according to the word frequency statistics:

2019: diversification (32.60%) > livability (27.04%) > internationalization (26.02%) > speeding (14.25%)

2020: diversification (34.20%) > internationalization (27.41%) > livability (19.93%) > high speed (18.46%)

2021: Internationalization (33.75%) > Livability (27.90%) > High Speed (21.07%) > Diversification (17.28%)

2022: High speed (28.68%) > Internationalization (27.25%) > Diversification (23.56%) > Livability (20.50%)

2023: High speed (30.54%) > Internationalization (28.25%) > Diversification (23.66%) > Livability (17.54%)

4.2

Semantic network analysis

On the basis of word frequency analysis, a semantic network is constructed to show the relationship between high-frequency words. The semantic web analysis is shown in Figure 1. “Guangdong-Hong Kong-Macao Greater Bay Area”, as the main high-frequency word, has a relationship with other words. “Internationalization”, “high-speed”, “diversification” and “livability” have also become important nodes, which are the main concerns of most tourists, and are also important brand personalities in the Guangdong-Hong Kong-Macao Greater Bay Area.

Positive evaluation terms such as “global”, “inclusive” and “open” are highly correlated with the “Guangdong-Hong Kong-Macao Greater Bay Area”, indicating that tourists’ overall feelings towards the Guangdong-Hong Kong-Macao Greater Bay Area are positive.

4.3

Visitor sentiment analysis

Emotional image refers to the real psychological feelings and emotional reactions of tourists when visiting a particular tourist destination and after the visit, which is generally categorized into positive emotions, neutral emotions and negative emotions. Using the Stanford syntactic analyzer, based on EDT, the segmented network evaluation text was sentiment analyzed, and the statistical results of sentiment distribution were obtained as shown in Table 4. The results show that 78.98% of the positive emotions, 10.24% of the neutral emotions and 10.78% of the negative emotions of the tourists’ perception of the image of the Guangdong-Hong Kong-Macao Greater Bay Area are positive, among which the positive and neutral emotions account for as much as 89.22%, and as a whole, most of the tourists have a good travel experience in the Guangdong-Hong Kong-Macao Greater Bay Area, and the tourists have a better overall impression of the Guangdong-Hong Kong-Macao Greater Bay Area and a high overall degree of recognition. Comparatively speaking, the proportion of neutral and negative sentiments is small, which indicates that the brand image of the Greater Bay Area already has a certain degree of international influence and recognition.

Table 4.

Emotional distribution of tourists

Affective category	Account for%	Intensity	Account for%
Positive emotion	78.98%	Normal	42.64%
		Moderate	13.51%
		Altitude	22.83%
Neutral emotion	10.24%
Negative emotion	10.78%	Normal	8.24%
		Moderate	2.14%
		Altitude	0.40%
Total	100.00%

5

Conclusions and recommendations

5.1

Conclusion

This paper relies on the 963 online travelogues collected about the Guangdong-Hong Kong-Macao Greater Bay Area (GHMA) as the data to extract the brand personality words of the GHMA, and classify these words into dimensions to measure the brand personality of the GHMA.

A total of 8,958 words were used to describe the brand personality characteristics of the Guangdong-Hong Kong-Macao Greater Bay Area, and the brand personality characteristics were internationalized (28.76%)> diversified (25.24%)> high-speed (24.44%)> in order of frequencyLivability (21.94%). The prominent brand personalities of the Guangdong-Hong Kong-Macao Greater Bay Area are identified as “internationalization”, “diversification”, “high-speed” and “livability”.

Five popular cities, Guangzhou, Shenzhen, Zhuhai, Hong Kong and Macao, were selected as the main research objects, and 10 of the 14 brand personality words were common to 5 cities, and 3 words appeared in only one city, accounting for 8.52%. The three unique characteristic words are “historical”, “artistic”, and “refreshing”. Of these 3 words, “refreshing” only appears 9 times in Zhuhai, and the remaining 2 words only appear in Guangzhou reviews. The word “rustic” is not found in these five cities.

With the change of years, the number of occurrences of brand personality words in the Guangdong-Hong Kong-Macao Greater Bay Area has shown an overall upward trend. The language network analysis shows that positive evaluative words such as “global”, “inclusive” and “open” have a high correlation with the “Guangdong-Hong Kong-Macao Greater Bay Area”. The results of EDT were used to analyze the sentiment of tourists, and the results showed that 78.98% of tourists perceived the image of the Guangdong-Hong Kong-Macao Greater Bay Area with positive emotions, 10.24% with neutral emotions and 10.78% with negative emotions, of which 89.22% were positive emotions and neutral emotions.

5.2

Strategic Recommendations

1)

Strengthen digital content dissemination to enhance internationalization image

Utilize digital media platforms (e.g. social media, short video platforms, etc.) to disseminate the internationalization stories and success stories of the Greater Bay Area. At the same time, use VR and AR technologies to provide an immersive online travel experience for global audiences, showcasing the Greater Bay Area’s iconic attractions, modernized cityscape and thriving business atmosphere. In addition, establish an official multi-language digital platform for the Greater Bay Area, incorporating AI translation technology to provide real-time translation services in English, Cantonese, Mandarin and other major languages, so as to attract more international tourists and foreign-funded enterprises, and to further strengthen the internationalized image of the Guangdong-Hong Kong-Macao Greater Bay Area.

2)

Promoting Diversified Experiences and Cultural Diffusion

“Diversity” is one of the core competencies of the Guangdong-Hong Kong-Macao Greater Bay Area, where different cultures, languages, ethnic groups and lifestyles converge. Through the construction of virtual museums, digital cultural exhibitions and other platforms, the rich history, arts, cultural activities and cuisines of the Greater Bay Area will be presented, further highlighting its cultural diversity. Develop an online cultural exchange platform to allow users from different cultural backgrounds to share and interact with each other, and organize online art festivals, cultural lectures and other activities to enhance mutual understanding and tolerance among different cultures in the region, and to showcase the image of the Greater Bay Area as a multicultural exchange hub.

3)

Accelerating high-speed transformation and demonstrating innovation and vitality

“High-speed transformation” is an important feature of the Guangdong-Hong Kong-Macao Greater Bay Area, especially in terms of transportation network, information flow and technological innovation. Through the creation of a digital transportation platform, we will showcase the efficient transportation system of the Greater Bay Area, including various convenient modes of travel such as metro, light rail, and shared bicycles. With the help of big data, cloud computing and other technologies, it provides visitors with real-time information on traffic flow and travel recommendations, highlighting the achievements of the Greater Bay Area in the construction of a modern transportation network. Utilizing virtual technology to showcase the latest achievements and innovations of the Greater Bay Area in high-tech fields such as artificial intelligence, 5G communications and biomedicine. Establishing a digital expo on science, technology and innovation to showcase the new face of the Greater Bay Area as a global center of science, technology and innovation, and to attract the attention of scientific and technological talents and enterprises around the world.

4)

Optimizing livability and enhancing quality of life

“Livability” is an important factor in attracting talents and residents to the Guangdong-Hong Kong-Macao Greater Bay Area, especially in the areas of housing, education, healthcare and environment. Through the smart city digital platform, the Greater Bay Area’s efforts in environmental protection, green mobility and healthy living can be showcased. For example, information on air quality, distribution of public service facilities, and convenience of living in the cities of the Greater Bay Area can be displayed through a big data platform to help tourists and potential residents understand the superiority of living in the Greater Bay Area.

Lingua:: Inglese

Frequenza di pubblicazione:: 1 volte all'anno
Argomenti della rivista:: Scienze biologiche, Scienze della vita, altro, Matematica, Matematica applicata, Matematica generale, Fisica, Fisica, altro

Feed RSS della rivista

Study on Digitalization and Communication of Brand Image of Guangdong, Hong Kong and Macao Greater Bay Area

Xiaojian Chen

Jiaqi Yuan

Pubblicato online: 23 set 2025

Ricevuto: 21 dic 2024

Accettato: 20 apr 2025

DOI: https://doi.org/10.2478/amns-2025-0991

Parole chiaveGuangdong-Hong Kong-Macao Greater Bay Area, Brand image, Online text analysis, TF-IDF weighting method, EDT

© 2025 Xiaojian Chen and Jiaqi Yuan published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Parole chiave
Guangdong-Hong Kong-Macao Greater Bay Area, Brand image, Online text analysis, TF-IDF weighting method, EDT