The BookSampo Linked Data (LD) portal was deployed in 2011 by the Public Libraries of Finland and has today nearly 1.6 million annual users. Its large knowledge graph (KG) covers virtually all Finnish Fiction literature but has not been fully exploited in Digital Humanities. This paper discusses how the KG can be used for literary search, data exploration, and research by presenting a new BookSampo user interface (UI) based on faceted semantic search and browsing with seamlessly integrated data-analytic visualization tools. This application makes it possible for the first time to analyze the BookSampo data in versatile ways without programming skills. The analysis results presented suggest interesting spatial, temporal, and topical trends in how the Finnish fiction literature has evolved during the last decades.
BookSampo1 [
BookSampo is used by library users and librarians for finding literary works of interest
and related contextual information. The original Drupal-based UI in use since 2011 provides
traditional text search engines for finding records and then related contents as links for data
exploration [
To facilitate this, this paper presents a new semantic user interface (UI) for for the BookSampo knowledge graph (KG). First in Section 2 the Sampo model underlying our work is overviewed.
After this the BookSampo KG is presented (Section 3). Based on using the Sampo-UI framework
[
Support collaborative data creation and publishing Use a shared open ontology infrastructure Make clear distinction between the LOD service and the user interface (UI) Provide multiple perspectives to the same data Standardize portal usage by a simple filter-analyze two-step cycle
Support data analysis and knowledge discovery in addition to data exploration
The Sampo model4 [
Principles P1–P3 lay a foundation for developing LOD services. The model is based on the idea of collaborative content creation (P1)6. The data is aggregated from local data silos into a global service, based on a shared ontology and publishing infrastructure (P2). The model supports the idea of separating the underlying Linked Data service completely from the user interface via a SPARQL API (P3). This arguably simplifies the portal architecture and the data service can be opened for data analysis research in Digital Humanities.
The idea of principles P4–P6 is to “standardize” the UI logic of Sampo portals to be created on top of a LD service SPARQL endpoint. The goal is to make the portals easier to use and implement. Principle P4 articulates the idea of providing diferent thematic application perspectives by reusing the data service. They are used by a two-step cycle for research (P5): First the focus of interest, the target group, is filtered out using faceted semantic search [ 12, 13, 14]. Second, the target group is visualized or analyzed by using ready-to-use data analytic tools of the application perspectives. A novelty of the model is to support data analysis, visualization, and knowledge discovery with seamlessly integrated tooling, even solving research problems using AI [15] (P6).
To create Sampo portals, the Sampo-UI framework [
The user interface (UI) on top of the underlying knowledge graph (KG) in a SPARQL endpoint has been created using traditional search and data exploration methods, and the full potential of the KG—nearly 9 million triples today—has not been fully utilized: the data covers all Finnish fiction literature and beyond and is interesting from a Digital Humanities (DH) research perspective, too. Table 2 lists the number of instances of diferent entity types in the data from back in 2013 and today.
Class (Type) Literary works Editions Book covers Fictional characters Contemporary reviews Literary series Literary awards Literary award series People (e.g. authors) Author’s pictures Publishers
Instances
Fig. 1 illustrates how the novel Pride and prejudice is modeled in the BookSampo KG. The kaunokki:romaani entity represents the abstract work level of the novel. That entity has links to the entity of the author Austen, Jane and the kaunokki:fyysinen_teos entity representing the Finnish edition of the work translated by Joutsen, O. A.. This Finnish edition further has the link to the publisher entity of the WSOY publishing house.
The BookSampo data divides works into two levels: abstract and physical work levels. The data model is based on the FRBRoo model [16] but simplified to the aforementioned two levels. The abstract work level is equivalent to the work level in the FRBRoo model while the physical work level represents what would be the manifestation level in the FRBRoo model. In practice the abstract work level deals with information that is shared between all editions of a work. The physical work level on the other hand contains edition-specific information, e.g., number of pages, and publisher, that are specific to the edition. In the case of the BookSampo data this edition-specific information is often recorded for the first editions in all relevant languages. That is to say, an average work will have a physical work level entities for at least its first edition in the original language as well as for the possible first editions of translations of the work into Finnish and/or Swedish. If a work is translated again later in time, an additional physical work level entity might be added for the new updated translation. Fig. 2 illustrates the split for Mika Waltari’s novel Sinuhe Egyptiläinen.
This section first present the standard model of Sampo-UI for designing and implementing semantic portals. After this, it is shown how the model was used in the BookSampo portal.
Generic Sampo-UI model The new BookSampo User Interface (UI) is built using the
SampoUI framework [
A Sampo portal is implemented by configuring the pages types depicted in Fig. 3, i.e., the
landing page, application perspective pages, and the instance pages. This is done using JSON
(JavaScript Object Notation) configuration files [
Structure of the UI The user of BookSampo first lands on the landing page (shown in Fig. 4) when opening the portal. The portal consists of five application perspectives based on corresponding classes:
2. Publications. This perspective deals with the physical work level of all works. 3. People. This perspective deals with authors and other people related to literature, e.g., illustrators, translators, and reviewers. 4. Covers. This perspective deals with book covers and information related to them. 5. Nonfiction books . This perspective deals with the abstract work level of nonfiction books.
All of the perspectives query the data from the same SPARQL endpoint of BookSampo KG. Selecting a perspective represents the data from that perspective. For example, choosing the Novels perspective shows the data through novels. The search view of that perspective lists all novel entities in a list with links to the rest of the KG as their properties.
The basis for choosing the aforementioned perspectives was to cover all aspects of Finnish
literature as comprehensively as possible with few perspectives. Novels perspective was chosen
as the perspective to cover fictional books due to novels being the largest subgroup of fictional
literary works in the data. Nonfiction books was chosen to supplement the Novels perspective
with the nonfictional works, although in the data it only represents a non-comprehensive subset
of nonfiction published in Finland. The Publications perspective covers all literary works on
the physical work level. The split between 1) novels and nonfiction books and 2) publications
follows the split made in the original BookSampo data [
To supplement the data on literary works, the People perspective was added to provide information on all people relevant to the presented data on literary works, whether it be the authors behind the books or other people relevant to them like the illustrators and translators of works. As the BookSampo KG also includes data on contemporary reviews, information of reviewers is also included in this perspective. To finish of the available perspectives, the Covers perspective was added due to the popularity of the book cover search function on the original BookSampo Portal, the search capabilities of which could be even further improved with the use of faceted search.
Clicking on a card for an application perspectives leads to that particular perspective’s faceted search view (see Fig. 5). The faceted search view consists of three key elements: 1) the facet menu on the left, 2) the results view on the right, and 3) the diferent visualization tabs on top of the results view.
The facet menu includes all the available facets that can be used to filter the data in that particular perspective. The Sampo-UI framework ofers various types of facets for diferent types of data. The BookSampo Portal utilizes three diferent facets types:
1. Checkbox facet. A facet for filtering results by selecting one or multiple checkboxes for wanted property value entities. Results are automatically updated when a checkbox is checked. Selecting multiple checkboxes works in a disjunctive way: Selecting both the genres romance and thriller for the genre facet would return works that belong to either (or both) genre. If the property values have a hierarchical structure, e.g., yearly literary awards (Finlandia Prize 2022) and the award series they belong to (Finlandia Prizes), the facet can be configured to show the entities hierarchically as well. 2. Integer range facet. A facet for filtering results by limiting the integer range a property’s value should be in, e.g. searching for works that have a page count in the range of 200–300.
The facet is applied by pressing the ’apply’ button. 3. Text facet. A facet for searching for results based on text string, e.g. searching for works by their names. The facet is applied after the user presses enter.
The results are shown in a table format on the right side of the screen. Each of the rows represents one entity. The diferent columns represent the diferent properties and property values these entities have. Column values that are underlined denote links to more information about that particular entity. By default the result set includes all entities that match the type of the application perspective with no filters applied. The results are automatically updated when any facets have been applied.
The links in the results table lead to the instance pages (shown in Fig. 6) of entities. Instance pages aggregate all the information about that particular object in the same page. This includes the information shown about the entity in the table view as well as possible further information not deemed relevant to be included in the table view. In the BookSampo Portal the general choice was to include information that could be used as facets as columns in the table view and leave the rest of the relevant information to the instance pages.
Similarly to the faceted search view, the instance page view can have multiple tabs for diferent ways of visualizing the data. These tabs depend on the type of the entity in question. Novels and nonfiction books, for example, have a specific tab for showing detailed information about the diferent publications of that particular work (shown in Fig. 7) that exist in the data.
Visualizing the data with integrated data-analytic tools The various tabs available in the faceted search views and instance pages of application perspectives ofer the user diferent ways of looking at the data as well as analyzing the data. The diferent visualization types available can be roughly split into three diferent categories: 1. Pie/bar charts. Charts that show the ratio of property values in comparison to each other, e.g., the top genres of novels (shown in Fig. 8). The component behind these visualizations is created with the ApexCharts7 library. 2. Maps. Charts that show entities on a map based on some location information related to the entity, e.g., settings of novels on a map (shown in Fig. 9). The components behind map visualizations are created with the Leaflet 8 and deck.gl9 libraries. 3. Time series. Charts that show the evolution of entities or some of their properties as a function of time, e.g., the evolution of average page counts throughout years (shown in Fig. 10). The components behind these visualizations are created with the ApexCharts and AmCharts10 libraries.
The Sampo-UI framework ofers some ready-to-use visualization types for all of these categories. To better utilize the potential of the data, some custom visualizations were developed. These visualizations use the libraries already included in the framework, but have custom configuration options, mappers and/or data processing functions to expand on the possible visualization types available in the base Sampo-UI framework. The inspiration behind these components came from the data itself, where no existing visualization type could readily be used to visualize an aspect that could potentially turn out to be interesting.
Two examples of such components are shown in Fig. 11 and Fig. 12. Fig. 11 shows a deck.glbased component for visualizing the gender ratio of authors, who have written novels with a location as a novel’s setting. The circles on the map represent diferent locations, where the size of the circle is determined by the number of novels with that location as their setting and the color of the circle represents the gender ratio for that location. Red circles indicate a
strong female author majority and blue circles a strong male author majority, with the purplish shades in between indicating a more even split between female and male authors. In Fig. 11, for example, interestingly the major locations in the UK have at least a slight female author majority, while Germany on the other hand has more locations with a male author majority. With the faceted search tools the UI ofers, the user could easily explore in more detail whether, for example, the top genres for these locations could play a part in the diference.
Fig. 12 is based on an existing ApexCharts time series component configuration, but with new data preprocessing functions to enable it to handle dynamic sets of series. This enables the visualization component to render the top themes and keywords and their evolution throughout the years for publications dynamically based on the applied facets instead of having to predetermine the visualized themes and keywords. In the bottom visualization for the top 10 keywords, for example, there is an interesting visible spike for two keywords – kielletyt kirjat11 and myrkkykaappi12 – during the late 1930s to 1940s, when the Winter and Continuation Wars were fought in Finland.
The general idea behind the inclusion of visualizations is that they should visualize the entities being filtered in that particular perspective. For example, in the Novels perspective all the visualizations mainly deal with the novels and their properties. Ideally the set of the 11Finnish for ’banned books’. 12Lit. ’poison cabinet’. This is a term used for a cabinet in certain libraries that would be used to store controversial/banned books. visualized entities should be the exact same result set as shown in the table view of the results. The problem, however, arises with properties having varying annotation coverage. This is especially apparent with the map visualizations, where the ratio of the set of visualized entities to the whole result set can become low, e.g., only around 12,000 people out of 62,000 people having an annotated place of birth. The problem is also further exacerbated by issues with the geographical location data in the BookSampo KG, where a single location can have multiple diferent latitudes and longitudes listed in the data. In these cases those locations are just filtered out of the results to prevent them from skewing the visualization by showing the ’same location’ and thus instance counts at multiple diferent locations.
The choice on whether to even include a visualization in the first place was based on how potentially interesting the data itself could be: The settings of novels as well as the places of birth and death could potentially show some interesting patterns and correlations between factors, but on the other hand something like the place of education like won’t ofer much due to the extremely low annotation coverage (less than 2,000 people out of 62,000 people) as well as the most common annotations largely being Finnish cities with major universities.
For pie/bar charts this problem with coverage is somewhat easier to mitigate by being able to visually show the ratio of known and unknown values by including a slice with the label ’Unknown’. This also helps the user more easily assess how well the annotated values might represent the whole data set. For map visualizations and most of the time series visualizations this ratio of known to unknown isn’t as clear unless the user manually counts together the instance counts in the visualizations.
Related Works Linked Data and ontologies have been used in libraries [17], museums [18,
19, 20], and archives [21, 22]. Using LD is advocated by major library organizations, such as
IFLA13 and OCLC 14, and several libraries provide their collections as data in this form [23]. LD
has been used in building infrastructures, such as ARIADNEplus 15 for archaeology, Linked
Art16 in the U.S., and in local eforts in Italy [ 24], the U.K. [25], Spain [26], and Finland [27]
to list a few examples. Cultural Heritage and DH have become a major application domain
for LD technologies [28, 29]. The focus in related research on portal UIs has been usually on
(explorative) search and browsing [
Evaluation The new BookSampo UI has been available mostly to members of the Semantic
Computing Research Group as well as people from the Finnish Public Libraries while the most
important issues of the underlying KG are being corrected. The UI has been improved based on
the wants and needs of the current users, e.g. by including additional facets the users felt the
UI was missing and by implementing new custom visualizations to better visualize aspects of
the data, but hasn’t yet been evaluated with outside testers. Based on previous in-use semantic
13https://www.ifla.org/references/best-practice-for-national-bibliographic-agencies-in-a-digital-age/
service-delivery/linked-open-data/
14https://www.oclc.org/research/areas/data-science/linkeddata/linked-data-overview.html
15https://ariadne-infrastructure.eu/
16https://linked.art/
portals utilizing the Sampo-UI framework, the framework is however suggested to have good
usability and scalability for the end user [
In terms of the evaluation of the contents of the BookSampo Portal itself, the SampoUI framework’s structure enables easily developing future improvements. The Sampo-UI framework ofers various ways to extend and customize the portal through things like custom components based on the needs of the users, which was already done during development based on the comments from the members of the research group as well as the people from the public libraries. Therefore possible problems or lacking features could be implemented and/or fixed in the future when the portal is made public.
The BookSampo KG has already shown itself to be interesting from the perspective of DH research [32], which the new UI could potentially make more accessible to both non-academic people as well as researchers not previously familiar with the data and its potential. The new UI ofers ways to explore and browse the data without the need for much technical knowledge or skills in query or programming languages like SPARQL and Python that are often used when conducting research with these kinds of data sets.
The ease of use, however, also means less ability for the user to afect the data being visualized. For example, the data on geographical locations has its problems with the excess longitude and latitude values as well as a missing hierarchy system for connecting cities and towns to the countries they are in. Issues like these would greatly benefit from the ability to preprocess the data before visualizing it, which would be available to the user if they were working directly with SPARQL and Python instead.
Another problem arises from the nature of simply how things are annotated. The annotations are done by humans who naturally might have their own ways and preferences of how they annotate things. For example, one person might annotate a book’s setting as ’Helsinki’ while another might annotate it as ’Finland’. Both of these choices could be correct on a level, but the ifnal choice is dependant on the annotator. With geographical locations, adding hierarchy and visualizing things on a country level could lead to most accurate visualizations in some cases, but with something like themes and genres, aggregating things to an upper shared concept might not be as easy as stories with similar elements can still be described in diferent ways.
Future Work The new UI ofers the users a new and more intuitive way to explore the BookSampo KG compared to the old portal with its limited search functions. With the integrated data-analytic tools the users can also easily analyze the data they are presented without having to learn languages like Python to work with the result data from SPARQL queries. The new UI could potentially serve as an intermediary step in DH research. The underlying BookSampo KG has a lot of potential to be used in literary DH research, but has not widely been used for it yet. With the new UI researchers could easily search, explore, and analyze the underlying data without needing to be familiar with technology like SPARQL or Python that would be needed otherwise. In addition to not needing to be familiar with the technology, the UI could be helpful in just narrowing down the are of interest to be researched. In comparison to trying to scour through the KG itself with its nearly 9 million triples, finding interesting topics or questions is be easier with the new UI, where one can expand on search in an iterative way by adding new restrictions on the results through the facets as you go.
Due to the nature of the Sampo-UI framework, the portal could easily be extended in the future. The BookSampo KG includes data on other types of works, e.g., poems and short story collections, that could easily be added to the portal as perspectives after assessing the quality of that data and the annotations related to them. The components and visualizations present in the portal could be expanded as well based on the needs and wishes of users and their feedback. The data could also be enriched by linking it to other linked data sources, e.g., BiographySampo KG [33], which includes data on over 700 people present in the BookSampo KG.
The new portal presented in this paper is planned to be opened for public use in autumn 2023. Before this can be done, the data issues need are being addressed by the Public Libraries of Finland. We also plan to publish (most of) the data for open use with the CC BY 4.0 license at the Linked Data Finland platform17.
Acknowledgements Thanks to Matti Sarmela, Kaisa Hypén, and Tuomas Aitonurmi for their collaborations as well as providing a newer version of the BookSampo KG to be used in the development of the new user interface. This project was funded by the Aalto University; Computing resources of the CSC – IT Center for Science were used.