This paper describes a practice for live social reuse of Schema.org annotations and Linked Open Data in the realm of events. The events of the “Palermo Capitale della Cultura 2018” initiative of the Ministry of Culture of Italy, were semantically enhanced by interlinking the available open data with related information inferred from the Linked Open Data cloud (namely DBpedia and Geonames). The resulting dataset - stored as an RDF graph within a triplestore was exposed via API in a knowledge graph by using the Schema.org vocabulary .The dataset was made also accessible to an event search assistant for tourists implemented as a bot for the instant messaging application Telegram. This effort shows how plain open metadata can be powered by Linked Data and semantic vocabularies like Schema.org, to became rich machineunderstandable descriptions usable by automatic bots to provide improved question answering experiences for the social user.
1.1
The evolution of the Web is strictly connected to the way users interact with it. Nowadays, the potential users of web data are not only human beings but also software services and software agents. For this reason data should be published on the web using standards and technologies which can be understood and elaborated automatically.
At present, the most popular Web applications, such as Facebook and Youtube,
offer Application Program Interfaces (APIs) that allow software agents to access the
information they host. Semantic Web technologies provide an adequate technological
substrate for supporting the representation of concepts and the relationships between
them through ontologies, and the recent evolution of Linked Open Data is the natural
way to publish, integrate, and link data semantically described. The information
available on the web uses different typologies and is published in heterogeneous
formats. Linked Open Data (LOD) aims to provide a technological substrate for
publishing structural data in a standardized format. The advantages of such an approach are
tangible and it is increasingly common for data on the web to be published following
the LOD principles [
SPARQL Query Language1 is the standardized language to query and retrieve Semantic Web data stored as RDF2 triples thus allowing and facilitating access to LOD resources. DBpedia3, can be seen as the semantic version of Wikipedia; it is the core of the Linked Open Data cloud, and provides a main access point for semantic enrichment. 1.2
Schema.org was launched in 2011, as a result of a join effort of the big players in the search engine field: Google, Microsoft, Yahoo!, and Yandex, with the aim of defining a shared vocabulary on common concepts of the real world.
Starting from the top-level concept, represented by the most generic type named
Thing, main sub-concepts have been defined to represent concepts related to
CreativeWork, Organization, Person, Place, Product and Event. Moreover, specific
subtypes have been defined to represent concepts in different popular domains such as
medicine4, or education [
As detected by the Web Data Commons5 project, nowadays, the adoption of schema.org is more than 39% of all Web pages.
Recent studies report that over 2.5 billion people have installed an instant
messaging app in its mobile phone, and already in 2015, interactions between people on the
Web were mediated by instant messaging apps more than social network. More
popular instant messaging platforms are: WhatsApp, Telegram, Viber [
Amongst them, Telegram is cross-platform and provides an appropriate API for
building chat-bots to interact with a user or with a group of users. [
“Palermo Capitale della Cultura 2018” is one of the initiatives of the Ministry of Culture of Italy supporting coordination of cultural events in order to promote tourism in a chosen city. This paper describes a bot developed for instant messaging Telegram platform is presented to provide and share rich information about the initiative events. The bot enrich sparse open event descriptions with LOD data and provides the social 1 https://www.w3.org/TR/rdf-sparql-query/ 2 https://www.w3.org/RDF/ 3 http://wiki.dbpedia.org 4 http://schema.org/docs/meddocs.html 5 http://webdatacommons.org user an easy to use interface to browse them. The interface hides the complexity of queries required to semantically build useful descriptions according to user spatiotemporal context.
Figure 1 shows the overall system architecture. The system serializes information about events in a knowledge graph containing not only the sparse open data published in the official website6, but also enriched entities extracted from the Linked Data silos that were interlinked as described in Section 2.1. This approach, based on knowledge graph and enriched entities, allows users to obtain additional information that were not included in the original website or in all the other services and mobile apps that are based on it.
Data extracted from the PCC2018 web site are firstly imported in a Drupal CMS in order to perform lexical cleaning of data and to improve improve efficiency of user query .
A mapping module translates CMS data into triples and stores them on a Virtuoso RDF triple-store. A interlinking module implemented via web-services enriches the resulting knowledge graph. A SPARQL endpoint anwers semantic queries on the graph.
Semantically enriched event descriptions are then reimported and delivered to a Telegram bot through an API endpoint supporting field selection and range queries on temporal and spatial data. 6 http://palermocapitalecultura.it/ 2.1
Semantic Enrichment is a term used to describe the process of transforming plain data into structured data that contains machine-readable statements. This enrichment can happen using ontologies or taxonomies of controlled terms with semantics defined by the data owner or, in the context of Linked Open Data principles, reusing machineunderstandable vocabularies with metadata values defined by external data providers as Europeana7 or DBpedia8). Knowledge graphs published on the LOD cloud can be traversed to extract references or descriptions of related entities.
The main issue with semantic enrichment lies in the way to automate the process,
in order to apply the enrichment to large volumes of data, instead of using manual
domain expert annotation. Interlinking rules, distance measure algorithms as well as
natural language processing techniques can support automated enrichment processes
and generate well-formed semantic data that exploit the LOD cloud [
In the context of this work, data enrichment was implemented using a mixed
approach including programmatic tagging via external services and federated SPARQL
queries to provide interlinking enrichment [
To semantically enrich event data, three specific enrichment techniques were used: • Semantic named-entity recognition • Geocoding enrichment • Spatial interlinking
The interlinking module developed in Python, implementing appropriate connector interfaces between the external services and the triple-store.
For example, the text referring the locality where the event happens was geocoded using Google Geocode APIs9 , then, expressed using Schema.org relations (Address, Administrative Areas, Latitude/Longitude coordinates) and finally reconverted in plain text address of the event place for sake of simplicity.
For the recognition of named entities the textual contents of the “title” and
“description” fields of each events was sent as input text to DBpedia Spotlight APIs10.
DBpedia Spotlight [
The geocoding and named-entity inferred triples were used as source data for the third type of interlinking, exploited via SPARQL federated queries12 to merge facts about the same event extracted from different sources of the Linked Open Data cloud. In particular, the SPARQL query implemented, infers from DBpedia nearby entities (places, historical monuments, etc). A threshold to the Haversine geospatial distance from the Event place coordinates was used to define the effective region of interest around the event place. 7 https://pro.europeana.eu/page/linked-open-data 8 https://wiki.dbpedia.org 9 https://developers.google.com/maps/documentation/geocoding/intro 10 http://spotlight.dbpedia.org 11 https://www.w3.org/TR/rdf-schema/#ch_seealso 12 https://www.w3.org/TR/sparql11-federated-query/
The semantically enriched events stored in the triple-store, were mapped back to the application main CMS, and served to the telegram BOT via the CMS output API endpoint.
Table1 shows the semantics and formats of event descriptors at various steps of the enrichment chain. 2.2
The Telegram Bot presented in this paper was designed to guide users in searching for events organized in the framework of the “Palermo Capitale della Cultura 2018” initiative. The interface facilitates the searching process by simplifying users’ interactions in order to provide access to information related to events matching their interest in a minimum number of clicks. In particular, customized keyboards were designed to help users in selecting straightforwardly the most commonly used search options.
The customized keyboard shown by the bot allows users to search along three specific dimensions: temporal, spatial and categorical (Fig 2a).
The spatial dimension allows users to search events in the nearby of their current position. Telegram Bot API supports the possibility to transmit to the bot the user coordinates, by using the request_location parameter associated to the keyboard button. After selecting the options along the three dimensions, a list of events matching user’s preference is shown as selection list (Fig 2b).
(a) (b) (c)
User can select from the list an event in order to access event description. In fact, for each event the bot answers with a message containing the title, an image related to the event and a brief description. Moreover, the bot can provide additional content related to opening hours, description details and a list of and nearby points of interest.
The enriched knowledge graph is used to provide this additional information. Specifically, the rdf:seeAlso property is used to provide detailed information to the entities related to the event, while the geonames:nearby property is used to provide information about the points of interest located in the neighborhood of the place in which the event takes place.
Descriptions of more than about 200 events (plus replicas) in the framework of the initiative “Palermo European Capitale della Cultura 2018” were enriched with full geo-location information, completed with images related to location or named subject, and with references to Linked Data entities, extracted via Named-Entity recognition and via geospatial inference rules. A compact semantic search GUI was provided through a Telegram bot to allow users to easily search and share information.
This effort shows how plain metadata (and Open Data) can be powered by Linked Data and expressed through semantic vocabularies like Schema.org, to became richmachine understandable data used unambiguously by automatic bots to provide improved question answering experiences for the social user.
Topics of current research are the enhancement of the enrichment inference engine and the upgrade of the bot to a conversational bot (chat-bot).