Dataset of Tourism Geography Sentiment Evaluation Model Application in Cities of Greater Bay Area of China (2008-2021)

Liu, Y.^{1, 2  Chen}, H. L.¹  Xiao, W. J. ^{1, 3*}  Bao J. G.¹  Wu, X. H.¹  Xu, J. L.¹

1. School of Tourism Management, Sun Yat-sen University, Guangzhou 510275, China;

2. Key Laboratory of Intelligent Assessment Technology for Sustainable Tourism, Ministry of Culture and Tourism, Zhuhai 519080, China;

3. School of Tourism, Jishou University, Zhangjiajie 427000, China

Abstract: The evaluation of tourist sentiment can play a crucial role in informing decision-making processes for tourism destinations. This study introduces the Tourism Sentiment Evaluation (TSE) model and application platform, which was developed through the creation of a tourism-specific sentiment lexicon, the establishment of semantic rules, and the selection of a sentiment rectifier. This study introduces a novel methodology and a novel instrument for assessing the sentiment of tourist destination. To gather data for the TSE model, we employed online review data sourced from prominent tourism websites including Tripadvisor, Mafengwo, and Ctrip. In this study, we conducted a collection of online reviews pertaining to 11 cities within the Guangdong-Hong Kong-Macao Greater Bay. The dataset was gathered from the years 2008 to 2021, utilizing an application platform to obtain a comprehensive dataset for sentiment evaluation of these cities. The dataset consists of fifteen data files, encompassing various aspects such as the ranking of attention and reputation for eleven cities, the differences in attention and reputation rankings specifically for cities in the Greater Bay Area, overall sentiment analysis of the Greater Bay Area and sentiment analysis for individual cities including Hong Kong, Macao, Guangzhou, Shenzhen, Zhuhai, Foshan, Huizhou, Dongguan, Zhongshan, Jiangmen, Zhaoqing. The dataset is archived in .xlsx format with data size of 34 KB.

Keywords: sentiment evaluation; tourism destination; TSE model; reputation

DOI: https://doi.org/10.3974/geodp.2023.01.14

CSTR: https://cstr.escience.org.cn/CSTR:20146.14.2023.01.14

Dataset Availability Statement:

The dataset supporting this paper was published and is accessible through the Digital Journal of Global Change Data Repository at: https://doi.org/10.3974/geodb.2023.05.06.V1 or https://cstr.escience.org.cn/CSTR:20146.11.2023.05.06.V1.

1 Introduction

The evaluating of tourism destinations plays a crucial role in uncovering their level of development and competitive environment^[1]. Therefore, the evaluation results have garnered attention from various sectors^[2]. The utilization of questionnaires as the primary method in traditional evaluation models frequently leads to diminished efficiency^[3]. The proliferation of online reviews generated by mobile devices has emerged as a novel data source for assessing tourism destinations. Nevertheless, the extraction of preference and other pertinent information from this data source continues to present a significant obstacle^[4]. As a response, our research team has successfully developed the Tourism Sentiment Evaluation (TSE) model application platform. Additionally, we have created a dataset of sentiment evaluation specifically tailored to cities within Guangdong-Hong Kong-Macao Greater Bay Area. This development is based on principles of emotion classification theory and utilizes vocabulary matching technology. The model exhibits strong reliability^[4] and demonstrates a high level of accuracy^[5]. It has been successfully applied in the evaluation of tourism destination ^[4], the capture of urban tourism images^[6], and the measurement of spatial structure within tourism market^[7]. In addition, the TSE model application platform demonstrates suitability for various geographic application scenarios, including the analysis of tourism spatio-temporal behaviors and the investigation of mechanisms underlying human-land interaction.

2 Metadata of the Dataset

The metadata of the Dataset is summarized in Table 1.

3 TSE Model Application Platform

3.1 Overview of TSE Model Application Platform

The application platform utilizes the Tourism Sentiment Evaluation (TSE) model as its central framework and relies on large-scale reviews from big data as its primary data source. This model has the potential to be utilized in diverse geographical contexts, such as the assessment of tourism destinations and other similar scenarios.

3.2 The Building of the TSE model

The building of the TSE model comprises three distinct steps^[4].

(1) The creation of a sentiment lexicon tailored specifically for the tourism industry. We conducted a manual deep reading of travel logs and online reviews written by tourists. Through this process, we identified and extracted the most frequently used words that tourists employ to convey their sentiments. The newly introduced terms were subsequently incorporated into the sentiment lexicon of HowNet, which is a publication by the China National Knowledge Infrastructure (CNKI). The outcome of this procedure led to the creation of a sentiment lexicon specifically tailored for the domain of tourism. This lexicon consists of 3,507 positive words and 3,365 negative words.

(2) Establishing semantic rules. We have set 32 semantic rules by examining the roles of adverbs of degree, adverbs of denial, and adversative conjunctions in influencing sentiment tendencies, when combined in sentence patterns. The specific rules are outlined in the citation provided as reference^[4].

(3) Choosing a sentiment rectifier. The statistical data obtained from questionnaires administered by the World Tourism Organization (UNWTO) was employed to validate the robustness of our methodology. Based on our analysis, it has been determined that the ideal value for sentiment rectifier in the TSE model is 4. In essence, evaluations will be deemed favorable solely if the quantity of positive lexemes exceeds the quantity of negative lexemes by a factor of four or more.

Table 1  Metadata summary of the Dataset^[8]

3.3 The Primary Functionalities of the TSE Model Application Platform

The application platform primarily encompasses two fundamental functionalities, namely sentiment computation and co-occurrence analysis. The process of sentiment analysis can be accomplished using the semantic model. The algorithm can be described as follows^[5]:

Whenandis the firstly-appeared adversative conjunctions, then adopt Equation (1) to calculate sentiment score of the text.         

                                                                               (1)

Whenandis the secondly-appeared adversative conjunctions, then adopt Equation (2) to calculate sentiment score of the text.     

                                                                          (2)

When, then adopt Equation 3 to calculate sentiment score of the text.

                                                                                                          (3)

Among these,  is the sentiment score of the text, including 1 (positive), -1 (negative), 0 (neutral). is the number of adverbs of denial,is the number of adversative conjunctions,is the number of adverbs of degree before positive words,is the number of adverbs of degree before negative words, P is the number of positive words, N is the number of negative words, e is the sentiment rectifier.

The co-occurrence function is responsible for producing a co-occurrence matrix and an adjacency matrix of keywords, which is derived from the keywords inputted by users.

4 Data Development

4.1 Data Capture

Based on criteria such as popularity, the abundance of comments, user engagement, and the length of comments, the data sources selected for this study are Maotuying, Mafengwo and Ctrip. The objective is to gather tourism reviews from 11 cities covering the period from 2008 to 2021 by python. The collected data will include information such as the time of the review, the content of the review, the review score, and other relevant fields.

4.2 Attention and Reputation Calculation

Utilize sentiment analysis methodology to derive sentiment classification (negative, neutral or positive) for individual reviews across 11 cities spanning the time from 2008 to 2021. The quantity of comments is considered as a measure of attention, the ratio of positive comments is considered as an indicator of reputation, and the sentiment of tourism reviews is assessed on an annual basis. Based on the classification outcomes, the study has obtained data on the variations in attention, reputation and ranking among 11 cities.

4.3 Emotional Image

Based on the sentiment analysis outcomes, the positive and negative comments from 11 cities were utilized as the primary data source. The high-frequency word analysis function was employed to identify the most commonly occurring words associated with positive and negative sentiments. From this analysis, a keyword file was created, consisting of the top 200 high-frequency words. Subsequently, the co-occurrence analysis function was utilized to construct an adjacency matrix, representing the relationships between high-frequency words. Finally, Gephi was employed to visualize emotional image of the 11 cities for each year.

5 Data Results and Verification

5.1 Data Composition

The dataset comprises a total of 15 tables, including the rankings of the attention scores of 11 cities, the rankings of the reputation scores of 11 cities, the highlights variations in attention and reputation rankings across cities, a comprehensive emotional portrayal of the Greater Bay Area during this specified timeframe, and the emotional depiction of the cities such as Hong Kong, Guangzhou, Zhuhai, Foshan, Huizhou, Dongguan, Zhongshan, Jiangmen, Zhaoqing.

5.2 Data Results

The rankings of attention among the 11 cities exhibited substantial fluctuations prior to 2016, but subsequently reached a state of stability (Figure 1). Guangzhou, Shenzhen, Foshan, Huizhou, Zhaoqing, Zhongshan and Jiangmen have exhibited a consistent stable pattern in their rankings of attention since the year 2014. Guangzhou exhibited a persistent high ranking, whereas Zhaoqing, Zhongshan and Jiangmen consistently maintained low rankings. Both Hong Kong and Macao underwent a significant decrease.

Based on analysis of the data, it is evident that the reputation levels of 11 cities over a span of 14 years have exhibited fluctuations and overall upward trends (Figure 2). The reputation exhibited significant fluctuations between the years 2008 and 2013, followed by a consistent upward trend from 2014 to 2021. With the exception of Jiangmen and Huizhou, the remaining nine cities exhibited reputation levels surpassing 0.80 in the year 2018 and subsequent years. Moreover, the levels of reputation in Macao have consistently remained at a high level, exhibiting stability comparable to that of Hong Kong and Zhongshan.

Guangzhou was selected as the focal point for examining the emotional portraits, specifically in relation to positive and negative reviews. Based on the analysis of Figure 3, it becomes apparent that the positive feedback from visitors to Guangzhou primarily revolves around tourist attractions and infrastructure. The overall depiction of the city encompasses its picturesque landscapes, favorable ecological conditions, and the rich cultural legacy spanning a millennium in the Lingnan capital. This collective portrayal accentuates the city??s identity as a floral metropolis that harmoniously amalgamates contemporary elements with historical significance. In contrast, negative feedback, on the predominantly revolves around dissatisfaction with develop­ment of tourist attraction and the accompanying facilities. Significantly, the issue of overcrowding on cruise ships, limited diversity in entertainm­ent offerings, concerning san­itation conditions, expensive and unappealing food options, inadequate parking facilities, and challenges in locating affordable parking spaces are noteworthy concerns.

Figure 3  Positive and negative emotional images of Guangzhou tourists ^[10]

5.3 Data Verification

This study utilizes the statistical data obtained from questionnaires administered by the World Tourism Organization (UNWTO) over a period of 10 years in order to validate its findings. The findings confirm that TSE model demonstrates a significant level of reliability. For a more comprehensive understanding of the results, please consult the relevant literature ^[4].

In contrast to the other six machine learning models, the TSE model exhibits consistent accuracy. A comprehensive analysis of the specific outcomes can be accessed in the relevant academic literature^[5].

6 Discussion and Conclusion

The dataset provides a comprehensive overview of the impact and perception of cities within the Greater Bay Area. It sheds light on the competitive landscape among the 11 cities in the region and highlights the key factors that concern tourists. A decision-making framework can be established to guide the management of the tourism destination.

Author Contributions

Liu, Y., Chen, H. L., Xiao, W. J. and Bao, J. G. designed the development of the dataset; Wu, X. H. and Xu, J. L. collected and processed the data such as original reviews and reputation; Liu, Y. and Xiao, W. J. designed the model and algorithm; Xiao, W. J. conducted data verification; Chen, H. L., Xiao, W. J. and Wu, X. H. wrote the paper; Liu, Y. and Bao, J. G. participate in the revision of the paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]    Gong, J., Yang, S. Y. Study on tourism destination evaluation based on web reviews??taking 31 provinces in China for example [J]. Journal of central China Normal University (Natural Science), 2018, 52(2): 279‒286.

[2]    Shi, C. Y., Zhang, J., Shen, Z. P., et al. Review of the studies on the tourism spatial competition and cooperation [J]. Geography and Geo-Information Science, 2005(5): 85‒89.

[3]    Zhen, F., Wang, B. Rethinking human geography in the age of big data [J]. Geographical Research, 2015, 34(5): 803‒811.

[4]    Liu, Y., Bao, J. G., Zhu, Y. L. Exploring emotion methods of tourism destination evaluation: a big-data approach [J]. Geographical Research, 2017, 36(6): 1091‒1105.

[5]    Liu, Y., Meng, L. K., Bao, J. G., et al. A comparative study of sentiment computing methods: will machine learning be overwhelming? [J]. Nankai Business Review, 2021, 24(5): 63‒74.

[6]    Liu, Y., Xu, X. J., Zhao, Y. Analysis on characteristics and differences of tourism destination image based on TSE and IPA Model??the case of Guangdong province [J]. Tourism Forum, 2019, 12(6): 41‒49.

[7]    Liu, Y., Chen, H. L., Cao, Y. H. Analysis and theoretical modification of the spatial structure of tourist source market based on tourism review data [J]. World Regional Studies, 2023, 32(5):113‒124.

[8]    Liu, Y, Chen, H. L., Xiao, W. J., et al. Dataset of tourism geography sentiment evaluation model in cities of Greater Bay Area of China (2008-2021) [J/DB/OL]. Digital Journal of Global Change Data Repository, 2023. https://doi.org/10.3974/geodb.2023.05.06.V1. https://cstr.escience.org.cn/CSTR:20146.11.2023.

05.06.V1.

[9]    GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. https://doi.org/10.3974/dp.policy.2014.05 (Updated 2017).

[10]  Liu, Y., Lin, X. Y., Zhang, T., et al. Report on Sentiment Analysis of Tourists in Guangdong-Hong Kong-Macao Greater Bay Area based on Big Data (2021) [M]. Xu, H. G., Bao, J. G. Blue Book of Guangdong-Hong Kong-Macao Greater Bay Area: Report on Tourist Industry of Guangdong-Hong Kong-Macao Greater Bay Area in China. Beijing: Social Sciences Academic Press, 2022: 168‒210.