Nowadays, the cultural heritage wealth held within museums, archives, libraries, and heritage sites is under continuous threat due to various environmental, geopolitical, and resource factors. The need for proper preservation of such invaluable cultural assets worsens the risk of their loss and hinders global access to their knowledge. This paper describes our first steps towards preserving cultural heritage knowledge through intelligent digital services. Our work is conducted as part of the PNRR e.INS “Ecosystem of Innovation for Next Generation Sardinia” project, Spoke 6, which aims to support digital transformation in various domains, including cultural heritage, in the Sardinian region, Italy. The main objectives of this project include the digitization of cultural heritage knowledge by employing innovative solutions for a proper knowledge representation, the exploitation of such digital knowledge for proposing contextual and personalized services, and suitable user-friendly interfaces and applications that enable both operators and the general public to access, explore, and engage with such knowledge.
Cultural Heritage (CH) is the legacy of tangible and intangible assets representing human
culture, history, identity, and diversity. UNESCO defines it as “ the entire corpus of material signs
– either artistic or symbolic – handed on by the past to each culture and, therefore, to the whole of
humankind”1. It includes numerous varieties of entities, such as archaeological sites, monuments,
museums, artworks, music, or festivals. CH is a priceless educational, touristic, economic, and
social cohesion resource [
The lack of appropriate instruments has often led to losing valuable content for preserving and
promoting our cultural heritage. To overcome this issue, the typical solution is the integration
of specific digital technologies and services [
In this context, an adequate knowledge representation is paramount for implementing digital
services [
However, creating KGs is a challenging task [
This contribution describes our first steps toward promoting the preservation of CH knowledge by devising and implementing innovative digital services. Our work is conducted as part of the PNRR e.INS “Ecosystem of Innovation for Next Generation Sardinia” project, Spoke 6, aimed at supporting digital transformation in various domains, including cultural heritage.
In the context of improved digital transformation in tourism, the main goal of our work is the
cooperation between technology partners and culture operators to develop a highly innovative
service for the automatic digitization and interconnection of cultural heritage through the
generation of proper KGs from structured and unstructured information and their exploitation
for devising specific novel services also tailored to individual visitors or specific groups. As
KGs can link and organize data from historical archives, museums, cultural institutions, tourist
guides, and other relevant sources, enabling an in-depth understanding of our cultural heritage
and facilitating the user experience, creating proper KGs and related digital services facilitates
the enhancement and promotion of our heritage. They also enable knowledge sharing among
various communities and make collaboration among cultural institutions possible. In detail,
appropriate KGs for cultural heritage through innovative artificial intelligence techniques that
exploit the potential of LLMs will be created, and algorithms and platforms that implement
specific contextual and personalized targeting systems will be developed. For the sake of
completeness, let us point out that nevertheless LLMs may produce wrong facts due to their
embeddings [
The devised tools and services aim to increase the competitiveness and attractiveness of the Sardinian region. The goal is guided by a series of objectives, including: • The digitization of cultural heritage by creating usable, comprehensive, robust, and coherent knowledge graphs. • The exploitation of the created KGs for innovative, efective, and eficient products and services that take advantage of the potential of the developed knowledge base. • The development of a technological infrastructure that provides the aforementioned contextual and personalized functionalities.
Our work will yield distinct products and outcomes, as follows: • First, proper KGs for the cultural heritage domain, created in accordance with the requirements analysis and use cases defined during the first step. • The KGs will be fed into contextual and personalized targeting systems that provide (groups of) visitors with suggestions based on their context, interests, and preferences. • A properly accessible, usable, immersive, and customizable technological infrastructure that provides the contextual and personalized features devised in the earlier stages.
For the sake of clarity, in Figure 1, the high-level architecture of the general platform is depicted. According to the primary outcomes described in Section 3.2, the platform is composed of three main modules: knowledge graph generation, contextual and personalized targeting, and infrastructure. Each module is briefly described in the following.
This module is aimed at generating the final KGs for CH. In particular, for each use case defined during the project, relevant triples (in the form of <subject, relation, object>) are extracted from structured or unstructured textual sources provided by stakeholders by employing innovative algorithms based on a proper LLM-prompting strategy. A specific computer-aided validation methodology will be used to investigate the correctness of the extracted triples. In detail, innovative and eficient active learning algorithms will embrace an incremental approach for determining the correctness of triples. This will be achieved through an iterative query strategy in which, at each step, a human operator is asked to validate a limited set of triples, pre-classified by the algorithm trained in the previous step, with the aim of incrementally correcting the error, and, therefore, improving the efectiveness step by step. The output will be a proper KG composed of the generated triples for a given use case.
This module aims to generate proper explainable recommendations by innovative LLM-based algorithms that explore the paths of the generated KGs. For example, if we consider the path [<user_1> visited <Castle of Cagliari> associated_to <Ancient Sardinia> associated_to <ExhibitionX>], the resulting text to explain a recommendation for the place ExhibitionX to user_1 could be “It is related to the Castle of Cagliari, which you visited recently, as both belong to the Ancient Sardinia category”. This approach improves the user experience and confidence. This module first analyzes the KGs by exploiting ad-hoc methodologies to understand entity structures and relationships, identify afinity relationships between entities, and adequately represent entities to overcome the limitations of existing methods. Afterwards, the module generates recommendations with a twofold strategy. On the one hand, specific targets, such as identifying similar entities or forming clusters of related entities, are defined. Recommendations are generated based on clusters and relationships, validating their quality (contextual targeting). On the other hand, personalized representations are created by combining afinity information and specific user characteristics, including interaction history. The personalized representations will allow us to provide recommendations tailored to users’ individual needs or specific contexts ( personalized targeting).
The final technological infrastructures will integrate (i) the essential services for knowledge ingestion, i.e., the information extraction and management from multiple sources of structured and unstructured information; (ii) the KG creation and management services, including the human-assisted triples validation; (iii) the targeting services (both contextual and personalized); (iv) the application layer of end-user services with various management services (e.g., user creation, profiling, ID management and authentication, travel experience historicization, or sharing and social media interfaces). The integrated platform will be accessible to two main categories of users: first, the entities involved in the knowledge construction phase (cultural institutions, touristic operators, or other stakeholders), i.e., users designated for the data ingestion and the iterative manual triples validation, and second, the end-users, such as a website visitor, which will be the beneficiary of the targeting services.
The final integrated tool may enable (a) high quality and customization in the search, retrieval, and fruition of knowledge in the cultural domain referring to a target territory; (b) active support and guidance of the user in the construction of custom services such as route stages and places/resources according to his or her interests; (c) proactive formulation of personalized suggestions and recommendations regarding ideas, visits, experiences, and related products, (d) support in sharing knowledge and experiences with other users and visitors connected to the platform and its services to assist in creating and developing cultural communities.
Our work is expected to positively afect the cultural communities from several perspectives. In particular, we expect an increase in the quality and quantity of available information on the CH and the related services of the involved territories, increased visibility and enhancement of the cultural heritage at a national and international level, greater collaboration and sharing among the diferent communities and cultural institutions involved in the project and augmented competitiveness and innovation of tourism and culture operators in Sardinia.
We acknowledge financial support under the National Recovery and Resilience Plan (NRRP), Mission 4 Component 2 Investment 1.5 - Call for tender No.3277 published on December 30, 2021 by the Italian Ministry of University and Research (MUR) funded by the European Union – NextGenerationEU. Project Code ECS0000038 – Project Title eINS Ecosystem of Innovation for Next Generation Sardinia – CUP F53C22000430001- Grant Assignment Decree No. 1056 adopted on June 23, 2022 by the Italian Ministry of University and Research (MUR). //doi.org/10.1145%2F3409256.3409834. doi:10.1145/3409256.3409834. [16] J. Baek, A. F. Aji, A. Safari, Knowledge-augmented language model prompting for zero-shot knowledge graph question answering, 2023. arXiv:2306.04136. [17] R. M. Harrison, A. Dereventsov, A. Bibin, Zero-shot recommendations with pre-trained large language models for multimodal nudging, 2023. arXiv:2309.01026.