In this paper we present the motivations, the methodology and the data used to develop a platform aimed at improving the information about peripheral and shrinking areas in order to foster their attractiveness. As case study, we select the internal area of the Ufita Valley in Irpinia (Campania, Italy). The platform shows through maps and statistics the insights on the cultural attractions in the area of interest on the basis of an aspect-based sentiment analysis model trained on the Google reviews. The platform, addressed to local administrations, is intended as a tool for obtaining an overview of public sentiment towards cultural sites, understanding strengths and weaknesses, as well as for supporting governance and intervention policies for these sites.
stitutional capacities to improve and promote cultural tourism and highly depends on external resources to cope Shrinking areas, as reported in Grasland et al. [1], are with the diferent problems they may face [4, 5, 6]. internal and rural regions afected by depopulation, de- In order to support public administrations and instimographic decline and a rise in the proportion of elderly tutions in their local governance a data-driven decision people, remoteness of public services and scarcity of in- making approach could prove efective [ 7]. Several types frastructures. Shrinking areas are also characterised by of data, from reports to surveys, reviews and social media, geo-morphological fragility and poor accessibility which can be automatically leveraged to empower administracauses in some cases also economic impoverishment tions to make informed choices, guide strategic decisions [2, 3]. Furthermore, the high percentage of emigration as well as discover new insights, identify weaknesses among young generations towards larger centres, the and strengths, to enhance public services or optimise reabsence of job opportunities and employment, the aban- source allocation and investments. Efective data analysis donment of buildings, houses and land is also causing can indeed be used to derive essential information for the disappearance of traditions, customs, local knowl- making prevision, statistics, anticipate trends, and disedge and artisan expertise. The shrinkage is caused by cover areas of improvements thus transforming raw and a multitude of natural, political and economic factors disconnected data into rich knowledge to be interpreted common to many similar areas around Europe [4, 5]. In and reused wisely. order to face the problem and propose innovative and In this paper, we present the development of the KiNEefective solutions, local governments and public admin- SIS Project platform to support local administrations in istrations should work towards a common strategic plan improving attractiveness of the shrinking internal area of and the establishment of a collection of actions and activ- the Ufita Valley in Irpinia (Campania Region, Italy) with ities together with stakeholders, social groups, citizens, a specific focus on cultural sites and tourism. The paper companies and organisations [4]. is organised as follows. Section 2 ofers an overview of
Compared to larger centres, governance in shrinking the project; Section 3 delves into existing research on areas is indeed penalised by a lack of financial and in- the topic, providing context for our approach. Section 4 outlines the specific methodology we applied. Following this, Section 5 discusses the conclusions and outlines potential directions for future research.
The KNowledgE alliance for Social Innovation in Shrinking villages Project (KiNESIS)1 is an Erasmus+ Programme of
the European Union co-founded project coordinated and In the following sections, we propose a study related managed by the University of Naples "L’Orientale" which to the analysis and extraction of information related to gathers together many academic institutions, stakehold- specific categories in relation to diferent cultural sites ers and organisations across Europe, in particular Italy, in areas at risk of depopulation. Specifically, we show Spain, Germany, The Netherlands and Estonia. The Ki- the development of a model capable of extracting diferNESIS Project addresses the topic of international coop- ent dimensions of intervention regarding cultural sites eration focused on shrinking areas with the aim of pro- to guide public administrations in potential investments moting and fostering ideas, developing and sharing best for the maintenance and enhancement of cultural attracpractices, projects, workforce, productivity and attrac- tions present in the territory. In this context, some studtiveness. The Project’s objectives are to revitalise depop- ies show how investments in the tourism sector can be ulated, shrinking and marginalised areas by stimulating beneficial for the growth of inland areas at risk of depopentrepreneurship and entrepreneurial skills; to create lo- ulation [16]. cal living laboratories to promote social inclusion and entrepreneurial development; to experimenting new, innovative and multidisciplinary approaches in teaching and 4. Methodology learning; to facilitate the exchange, flow and co-creation of knowledge at a local and global level. During the years, several activities have been carried out as part of the KiNESIS Project such as co-participation tables, workshops and conferences, training sessions and summer schools, internships and Erasmus+ students exchanges, Hackathon and fairies, publication of handbooks, reports, scientific documents and best practices, the creation and dissemination of promotion materials.
Within the KiNESIS Project, with the aim of supporting the local administrations and institutions in the improvement of the attractiveness in the shrinking internal area of the Ufita Valley (Italy), we developed a user-friendly visualisation platform trained on an Aspect-Based Sentiment Analysis classification system for cultural attractions and cultural sites. In this section, the methodology adopted for training the Aspect-Based Sentiment Analysis (ABSA) classification system for cultural attractions and potentially of tourist interest in the Irpinia area will be discussed. Specifically, in Section 4.1, the data collected and on which the ABSA model was trained will be described; in Section 4.2, the model architecture and the elicited outputs will be presented; and finally, the integration into the data analysis visualisation platform will be presented.
The cultural tourism sector is increasingly driven by the use of data-driven approaches [8, 9]. In this context, data from user-generated content platforms allows for the collection of increasingly up-to-date and real-time insights capable of identifying trends to guide decisions around economic activities [10]. Specifically with regard 4.1. Data and Exploratory Data Analysis to tourism-related economic activities, in recent years, methods using Natural Language Processing (NLP) tech- The textual data used for training the ABSA model were niques have been applied to hotel reviews to extract from collected from the Google Maps platform. Specifically, user-generated content the sentiment and perceptions data on reviews of cultural attractions in Irpinia were of users in relation to various categories relevant to the collected. The names of the attractions were collected structure and the services it ofers [ 11, 12]. In addition, using the resources ofered by Sistema Irpinia2. Sistema NLP and topic modeling methods have been applied to Irpinia is an interactive platform that promotes sites of user-generated content in reference to quality dimen- historical, artistic, architectural, cultural, environmental sions in the museum field [ 13]. Additionally, other studies and food and wine heritage of Irpinia. This platform have investigated the attractiveness of Italian cities using contains 408 cultural sites. user-generated content. Specifically, users’ behaviour Therefore, the reviews related to the cultural attrachas been measured to identify the annual trend of pho- tions present on Sistema Irpinia were extracted from tographic activity in cities [14]. User-generated content Google Maps. A total of 9504 reviews were extracted. (UGC) on social media and review platforms related to Of these, about 4% are represented by reviews that do tourist attractions represents a valuable source of infor- not convey any textual information, showing only the mation for guiding decisions towards more informed user rating expressed on Google Maps through the aseconomic growth in regions that potentially benefit from signment of stars. For ABSA model training purposes, cultural tourism [15]. However, information from UGC these latter reviews were removed. Before training the has often been analysed by considering only user ratings ABSA model, an exploratory analysis was conducted on or focusing on large tourist hubs such as Italian art cities the information conveyed by the data extracted in the [14]. Type Churches Historical buildings Castles Other places Archeological area Religious complexes Castles - historic palaces Total manner previously described. This information focused on: 1) the number of cultural sites present in each munici- Figure 1: Availability of visits for cultural sites. pality in the Irpinia area; 2) the type of cultural attraction; and 3) the accessibility to the type of cultural site. The largest number of cultural sites present and represented that allow the site to be reached and to the presence of on the Google Maps platform belongs to two municipal- any explanatory totems within cultural sites. For examities (Rocca San Felice and Bonito) in the Ufita Valley, a ple, in the following review extracted from Google Maps consortium of municipalities participating in the KiNE- in reference to the Goleto Abbey located in Sant’Angelo SIS Project. Specifically, the municipality of Rocca San dei Lombardi: Felice is represented by 14 cultural sites, while Bonito has 13 cultural sites. L’abbazia del Goleto è ancora CHIUSA
Table 1 shows the number of types of cultural sites PER LAVORI DI RISTRUTTURAZIONE che extracted from the Sistema Irpinia platform and conse- dovrebbero terminare il 4 giugno 2024. Spero quently for which reviews were found. The most rep- venga rispettata la data di consegna dei laresented type is related to churches (160), followed by vori perché è sempre un piacere visitare il historical buildings (65), castles (48), and the ‘other place’ complesso. L’abbazia è spettacolare e mi astype (36), which represents noble residences. The least petto che al termine della ristrutturazione, represented, although the most extensive in terms of spa- ldoelsla’arràchanitceottrodXip.i(ùT.hCe oGmopleltimoeAnbtbi eayllaisdsittitlal tial extent, are archaeological areas and religious com- CLOSED FOR RESTORATION WORK which plexes. should be completed on June 4, 2024. I hope
For each cultural site that falls into one of these types, the deadline for the work will be respected as it categories/aspects were assigned to the reviews using the is always a pleasure to visit the complex. The distant supervision method [17, 18] in combination with abbey is spectacular and I expect it to be even the information from the overall rating score given by more so after the renovation is complete. Conthe review stars. Specifically, this method allows to build gratulations to the firm of architect X. ) an annotated dataset without or with little human intervention. In fact, rule-based heuristics are used in order The extracted aspects are related to ‘Accessibility’ and to produce labeled data and on these labeled data pro- ‘Appearance of the place’, as well as an overall score that duced being then used to train a model. The rule-based shows the ratio for all aspects. As far as the ‘Accessiheuristic consists of lists of words, primarily adjectives bility’ aspect is concerned, the identified text portion is and adverbs, that can signal positive and negative char- ‘CLOSED FOR RESTORATION WORK’, while for the ‘Apacteristics related to the aspects identified as salient for pearance of the place’ aspect the identified text portion describing the conditions of a cultural site. In the case of is ‘The abbey is spectacular’. the ABSA model, four categories/aspects were identified The application of the distant supervision method althat are able to describe the conditions of cultural sites. lowed us to annotate the data at our disposal with miniThe aspects related to cultural sites are the following: 1) mal human intervention. Subsequently, an exploratory accessibility; 2) signage; 3) appearance of the place and analysis of the annotated data at our disposal was carried 4) overall score. out. In Figure 1, for example, the frequencies related to
Specifically, ‘Accessibility’ refers both to the availabil- the accessibility of the cultural sites present in the dataset ity of opening hours for the public and to the provision are shown. In addition, as previously mentioned, the ‘Acof equipment to make the visit more accessible to a wide cessibility of cultural sites’ refers to two diferent use range of visitors, while ‘Signage’ refers both to road signs functions. In the case of Figure 1, accessibility is shown in terms of availability for visits. For example, in Figure
to religious functions. Instead, the ‘To complete’ label with 52 cases refers to closed cultural sites that are not available for visits or are undergoing renovation work.
As previously mentioned, the dataset also includes information concerning the type of cultural site among its characteristics. This information allows for more granular information in relation to the diferent aspects identified. For example, Figure 2 shows the frequencies of availability for visits only for cultural sites of the ‘Archaeological Areas’ type. In this case, for example, we can note that in 2 cases it was reported that the archaeological Figure 4: Overall score - Negative. area is open all year round.
In addition, the dataset also includes user reviews of cultural sites in the form of star ratings. These were used in conjunction with the identified aspects to balance the In Figure 4 are shown the negative textual spans about overall score. Specifically, the textual spans related to the ’Appearance of the place’ with judgements around the the aspects of ‘Accessibility’, ‘Appearance of the Place’, state of neglect of some cultural sites; about the ’Signage’ and ‘Signage’ were identified. A sentiment and emotion with the reporting of the complete absence of explanalexicon3 was then applied to these spans to add a score tory panels or damaged and in relation to ’Accessibility’ associated with each aspect. The score from the lexicon with comments on the lack of services or structural defiwas used in conjunction with the user’s rating score in ciencies. the review. These scores from the lexicon for each aspect identified in the text, together with the scores assigned 4.2. ABSA Model and Platform by the user, were used as supervision labels to train the ABSA model. Figure 3 shows the textual spans extracted for diferent aspects in relation to the positive overall score.
Specifically, Figure 3 shows textual spans related to: Accessibility with the span ‘free site’; Signage with ‘guides available’; and Appearance of the place with textual spans related to the beauty and spaciousness of the cultural sites.
This section outlines the fine-tuning process of the chosen model and the implementation of the platform prototype to be made available to public administrations for efectively directing policies towards cultural sites with potential appeal for cultural tourism. The platform’s objective is to provide insights to inform decisions on which aspects of a specific cultural site to focus on in order to prepare it for the influx of tourists.
In this context, we applied ABSA to analyze and classify user reviews of cultural sites and attractions in the 3Ttohrey:sehntttpims:/e/nsatiafmndoheammomtioand.cleoxmic/oWneubsPeadgecso/mnrecs-vfraodm.httmhilsarnedpothsie- Ufita Valley. We employed the XLM-Roberta-base model 4, lexicon for the Italian language was used. Specifically, the NRC a multilingual pre-trained transformer-based language Valence, Arousal, and Dominance (VAD) Lexicon was used. This model[19], for the ABSA task. The model was fine-tuned resource includes a list of more than 20,000 words and their va- using a dataset of user reviews described in section 4.1. lence, arousal, and dominance scores. These scores represent the emotional qualities of the words. 4https://huggingface.co/FacebookAI/xlm-roberta-base
The dataset was divided into training (60%) and testing site. For the statistics, you can select the name of the (40%) sets. The fine-tuning process involved optimis- cultural site and the aspect of interest and view the correing the model parameters to minimise the loss function, sponding information. For example, Figure 6 shows the which measured the discrepancy between the predicted information related to Abbazia del Goleto in relation to and actual sentiment labels. We employed a multi-task the ‘Appearance of the place’. learning approach, training the model to simultaneously In this case, when selecting the aspect to view for a perform two tasks: specific cultural site, the user will find the information extracted using two methods: a pie chart and a bar chart. • Aspect Category Classification : Classifying The first shows the sentiment scores associated with the each textual span into its corresponding aspect selected aspect, while the second shows the top-n words category (e.g., ‘Accessibility’, ‘Signage’, ‘Appear- (sorted by frequency of use) extracted from the reviews ance’). and associated with a particular aspect. Indeed, in Figure • Overall Sentiment Score Assignment: Assign- 6, we can observe that the majority of reviews express ing a sentiment score (ranging from 1 to 10) to a very favourable opinion of the Goleto Abbey, as more each textual span based on the sentiment ex- than half (54%) have a very high ‘Appearance of the place’ pressed towards the corresponding aspect. score (9) and use words such as "evocative," "historical," "charming," and "beautiful view."
The overall sentiment score for each review was calculated as the average of the individual aspect scores, weighted by the number of mentions of each aspect. Ad- 5. Conclusion and Future Works ditionally, the overall score was adjusted based on the user’s overall judgement of the review (positive, nega- In this paper, we present the implementation of a plattive, or neutral). The fine-tuned model’s performance form capable of extracting and classifying sentiment inachieves an F1-score of 78% in extracting the correct tex- dices for various aspects from online reviews of cultural tual spans for each aspect and assigning the correct score. sites. Specifically, the ABSA model and the platform proThis performance represents the average value across the totype were implemented within the KiNESIS project, model’s performance. The platform developed to provide which aims to investigate methods for mitigating the public administrations with insights into cultural sites efects of ongoing depopulation in rural areas across Eufeatures two main data visualisation modes: a map and rope. In this context, the platform is proposed to public cultural site-specific statistics, as illustrated in Figure 5. administrations as a tool to support their policies re
The map is populated with markers corresponding to garding cultural tourism sites of potential interest. Inthe coordinates of cultural sites and displays the weighted deed, the platform can be a valuable tool for obtaining average of the scores for each identified aspect of the cul- an overview of public sentiment towards cultural sites, tural site. This approach visualises and considers the as well as a tool for directing active intervention policies frequencies of the diferent scores (ranging from 1 to for the maintenance of specific aspects related to cultural 7, negative to positive) for each review of the cultural sites. The platform is currently under development and is site. The map provides a quick overview of the sentiment only being tested for the Ufita Valley area and for Italianassociated with the identified aspects for each cultural language reviews. Additionally, thanks to the KiNESIS site. Additionally, the platform ofers a second data visu- Project, activities and data collection have already begun alisation tool that focuses more on the specific cultural to extend the training and testing phase to the Oldambt
region (The Netherlands) in collaboration with the Dutch partner of the KiNESIS Project. In this context, therefore, the ABSA model and platform will be tuned and made available in other languages to analyse data from other European regions. planning cultural tourism, Technological Forecasting and Social Change 162 (2021) 120345. [10] M. P. A. Austin, P. Austin, M. M. Marini, A. Sanchez,
C. Simpson-Bell, J. Tebrake, Using the Google Places API and Google Trends data to develop high frequency indicators of economic activity, International Monetary Fund, 2021. [11] Y. Liu, T. Teichert, M. Rossi, H. Li, F. Hu, Big data for big insights: Investigating language-specific drivers of hotel satisfaction with 412,784 user-generated reviews, Tourism Management 59 (2017) 554–563. [12] S. U. S. Chebolu, F. Dernoncourt, N. Lipka, T. Solorio,
Survey of aspect-based sentiment analysis datasets, arXiv preprint arXiv:2204.05232 (2022). [13] D. Agostino, M. Brambilla, S. Pavanetto, P. Riva,
The contribution of online reviews for quality eval[1] C. Grasland, R. Ysebaert, B. Corminboeuf, uation of cultural tourism ofers: The experience of N. Gaubert, N. Lambert, I. Salmon, M. Baron, italian museums, Sustainability 13 (2021) 13340. S. Baudet-Michel, E. Ducom, D. Rivière, et al., [14] S. Giglio, F. Bertacchini, E. Bilotta, P. Pantano, Using Shrinking regions: a paradigm shift in demog- social media to identify tourism attractiveness in raphy and territorial development, Ph.D. thesis, six italian cities, Tourism management 72 (2019) Parlement Européen; Direction Générale des 306–312.
politiques internes de l’Union . . . , 2008. [15] A. Torre, H. Scarborough, Reconsidering the esti[2] T. Amodio, Territories at risk of abandonment in mation of the economic impact of cultural tourism, italy and hypothesis of repopulation, Belgeo. Revue Tourism Management 59 (2017) 621–629. belge de géographie (2022). [16] M. H. Guimarães, L. C. Nunes, A. P. Barreira, [3] S. De Rubertis, Dinamiche insediative in italia: T. Panagopoulos, Residents’ preferred policy acspopolamento dei comuni rurali, Perspectives on tions for shrinking cities, Policy Studies 37 (2016) rural development 2019 (2019) 71–96. 254–273. [4] A. Haase, G.-J. Hospers, S. Pekelsma, D. Rink, [17] A. Go, R. Bhayani, L. Huang, Twitter sentiment Shrinking areas: Front-runners in innovative cit- classification using distant supervision, CS224N izen participation, The Hague: European Urban project report, Stanford 1 (2009) 2009.
Knowledge Network, 2012. [18] A. Giannakopoulos, D. Antognini, C. Musat, [5] D. Rink, P. Rumpel, O. Slach, C. Cortese, A. Violante, A. Hossmann, M. Baeriswyl, Dataset construction P. C. Bini, A. Haase, V. Mykhnenko, B. Nadolu, via attention for aspect term extraction with distant C. Couch, et al., Governance of shrinkage: Lessons supervision, in: 2017 IEEE International Conferlearnt from analysis for urban planning and pol- ence on Data Mining Workshops (ICDMW), IEEE, icy, Leipzig: Helmholtz Centre for Environmental 2017, pp. 373–380.
Research (2012). [19] A. Conneau, K. Khandelwal, N. Goyal, V. Chaud[6] D. Rink, A. Haase, K. Großmann, M. Bernt, hary, G. Wenzek, F. Guzmán, E. Grave, M. Ott, C. Couch, M. Cocks, A. Violante, C. Cortese, L. Zettlemoyer, V. Stoyanov, Unsupervised crossP. Calza Bini, How shrinkage and local governance lingual representation learning at scale, CoRR are interrelated across urban europe: a comparative abs/1911.02116 (2019). URL: http://arxiv.org/abs/ view, 2011. 1911.02116. arXiv:1911.02116. [7] R. Matheus, M. Janssen, D. Maheshwari, Data science empowering the public: Data-driven dashboards for transparent and accountable decisionmaking in smart cities, Government Information
Quarterly 37 (2020) 101284. [8] T. Kalvet, M. Olesk, M. Tiits, J. Raun, Innovative tools for tourism and cultural tourism impact assessment, Sustainability 12 (2020) 7470. [9] M. T. Cuomo, D. Tortora, P. Foroudi, A. Giordano,
G. Festa, G. Metallo, Digital transformation and tourist experience co-design: Big social data for