Widely employed digital archives of literary works are not necessarily neutral sources of knowledge, but rather literature ecosystems where the relevance of authors and works are shaped by communities of users. This could afect the visibility of authors belonging to historically marginalized groups of people, as it has been demonstrated by previous works on the under-representation of non-Western writers. In this work, we present an exploratory analysis on how divergent representations of works written by African and Asian writers emerge from diferent platforms. More specifically, we compare the reviews gathered from an Italian and an international social networks of readers, in order to highlight how authors from these continents are perceived by two diferent communities of users.
and their expressed sentiment about authors vary between the two platforms, a common Western-centric attitude of readers emerges from these communities.
The paper is structured as follow: in Section 2 the URW-KG is briefly described. Section 3 summarizes the data gathering process from Wikidata and illustrates a mitigation strategy of underrepresentation. In Section 4, the analysis of Anobii and LibraryThings communities of readers is presented. In Section 5 conclusion and future works are presented.
Several digital resources providing information about literary works and authors are available
online, created with diferent scopes and purposes. Some of them are general-purpose archives
that host user-generated contents, while some others are the outcome of monographic researches
curated by Digital Humanities scholars. Example of the former type are websites like Goodreads,
owned by Amazon, LibraryThing, owned by AbeBooks, and Anobii, owned by the Italian
publisher Mondadori, namely, social networks where users share their reads and review books.
A project of the Internet Archive Foundation, Open Library is an online service for book loan
where users can have an active role by adding and editing records in the archive and reviewing
books. Wikidata is a Knowledge Graph linked to Wikipedia, which has a section devoted to
literary works. Monographic platforms focus on specific literary genres across countries and
ages. It is the case of The European Literary Text Collection1 [
The Under-Represented Writers Knowledge Graph (URW-KG)4, a dataset of writers and their
works [
Our resource is built upon the Under-Represented (UR) Ontology Network, which organizes
knowledge about authors and their works and aligns information gathered from diferent
sources. The Ontology Network is composed of two main modules: the Under-Represented
Writers Ontology (URW-O) [
Building the URW-KG was a two-step process. We first gathered from Wikidata all the entities of type Person (wd:Q5) with occupation (wdt:P106) writer (wd:Q36180), novelist (wd:Q6625963), or poet (wd:Q49757). Consistently with the definition of underrepresentation provided above, we filtered out all the people born before 1808, which marks the beginning of the SpanishAmerican conflict – considered as the first process of decolonization. After this first step, we obtained a KG of 194, 065 authors and 145, 375 works, that allowed us to identify a significant underrepresentation of Transnational Writers. In fact, they are only 17, 368 on Wikidata, the 9%, and the number of their works is even more represented, since they are only the 8, 380 (5.7%).
In order to understand if such underrepresentation is specific to Wikidata or if is shared by other archives, we integrated into the knowledge graph three external sources of knowledge specialized in archiving books: OpenLibrary8, Goodreads9, and Google Books10. For this integration, we considered the writers obtained from Wikidata as a constant, and gathered all their works from Open Library and Goodreads: by doing so, we could assess the diference in the number of works on Wikidata and on the external resources.
The first part of the data integration strategy consisted in collecting writers’ ids from OpenLibrary and Goodreads. We first considered all the 55, 834 authors with an OpenLibrary id, retrieving all their works and editions; we thus obtained 891, 037 works and 1, 742, 252 editions. Results include a set of useful information, including language, publisher, synopsis, subjects, year of publishing, place of publishing, and ISBN. Then, in order to bypass the limitations posed by the lack of APIs from Goodreads, we scraped from Goodreads all the 9, 963 writers who do not have an OpenLibrary id (instead of considering all authors with a Goodreads identifier). Finally, using the Google Books API, we retrieved from Google Books additional information about the collected works through their ISBN. By doing so, 1, 177, 801 new items were obtained and mapped onto OpenLibrary and Goodreads.
The impact of the data integration strategy is presented in in Table 1. As it can be observed, the mapping resulted in a significant and general increase of information about writers literary production. The number of works rose from 145, 375 to 1, 371, 326 (8 times more), and information about them grew accordingly: there are 20 times more blurbs and 34 more subjects. The mapping did not only generically increased the number of works and all the correlated information, but also contributed in rebalancing the proportion between Western and Transnational 8https://openlibrary.org/ 9https://www.goodreads.com/ 10https://books.google.com/ works. The former, which are only the 5.7% on Wikidata, become the 10.1% in the KG after the integration.
The mappings between Wikidata and external resources showed that the role of communities of contributors can be crucial in reducing (or spreading) under-representation: platforms based on diferent groups of users result in diferent representations of World Literature. In this section, we deepen this intuition by providing a comparative analysis between two social networks of readers, in order to understand how and to which extent readers from diferent digital platforms discuss about Transnational writers.
For our investigation, we chose two platforms where users can upload, review, and discover books: LibraryThing11, which is composed of English readers, and Anobii12, a website owned by the Italian publisher Mondadori and thus having a prominent audience of Italian people. Such a choice paves the way not only for a generic comparison between two website, but may also provide insights about the diferent attitudes about books and writers from a local and an international community of people.
Given the high number of writers in our KG, the absence of APIs for gathering data from
Anobii and LibraryThing, and the lack of a structural mapping of authors with these archives,
we performed an analysis of a subset of writers. We selected a sample of the Transnational
writers who have the highest number of editions on Open Library (the topmost 20 born in
Africa, and the topmost 20 born in Asia), and manually found their personal pages on the two
platforms. We then gathered all their works and readers’ reviews. The number of reviews from
LibraryThing is 16, 758, while reviews gathered from Anobii are 12, 141. For each review we
computed the Sentiment score with existing tools that reached the State of the Art for Italian
[
First of all, a diferent distribution of writers by continent can be observed in the two communities. Reviews about African authors are the 70.5% on LibraryThing, but only the 53.6% on Anobii. Diferences between the number of reviews on LibraryThing and on Anobii may be also be observed at a fine grained level. Figure 1 shows the 10 authors with the highest delta between the number of reviews on LibraryThing and Anobii. Such diferences may be a signal of the impact of certain authors in a local market like Italy (e.g., Tahar Ben Jelloun), compared to international trends. In this sense, widening the scope of the analysis by augmenting the number of digital archives from diferent countries may brings light on which representations of “literatures” emerge from readers at global and local contexts (see [18] for a similar approach to Wikipedia).
Secondly, it is worth mentioning that readers from LibraryThing and Anobii seem to show more interest for authors of Western origin, even when they are not born in Western countries. As it can be observed from the data in Tables 2 and 3, 15 out of 20 African-born authors are children of European or American parents, or belong to white minorities. The number of reviews increases this gap, since the 91.4% of reviews are about European or American descents. Proportion are inverted for Asian-born authors: only the 25% of them have European origins (however, the 42% of reviews are about them). This fact highlights the vulnerabilities of the criteria upon which the distinction between Transnational and Western writers is drawn (see Section 2), which should be integrated with other features.
The two tools adopted for detecting the sentiment of reviews [
The Sentiment Analysis of reviews of books of African writers (Table 2) seems to be consistent with diferences emerged from the quantitative analysis. Except from J. M. Coetzee, who obtained the most negative sentiment score both on Anobii and on LibraryThing, readers opinions significantly difer. Simon Scarrow, namely the first ranked author for sentiment with a score of 0.681 on LibraryThing, is only ranked 8ℎ on Anobii. On the contrary, Wole Soynka obtained a score of 0.600 on Anobii while on LibraryThing he is only 16ℎ. Other significant diferences regard Chinua Achebe, Nadine Gordimer, and Tahar Ben Jelloun, who respectively obtain a score of 0.401, 0.304, and 0.494 on LibraryThing against the following negative values on Anobii: − 0.169, − 0.100, and − 0.194.
The trend of readers’ sentiment about Asian writers (Table 3) is not as simple to interpret as the previous one. This is probably due to the lowest number of reviews per author who may have had an impact on the average scores. Many authors receive comparable scores. Eg: 13Bidirectional Encoder Representations from Transformers: https://github.com/google-research/bert Rajneesh (0.637 vs 0.563), Deepak Chopra (0.461 vs 0.420), and Stephanie Laurens (0.320 vs 0.381). Anita Ganeri, namely the author who receives the highest sentiment on Anobii (0.666) is also the 2 ranked on LibraryThing, while Sri Chinmoy (0.949) and Solomon Volkov (0.008) are respectively the author with the best and the worst scores on LibraryThing, but these results are not comparable with Anobii, since an adequate number of reviews from this platform is missing. In general, readers’ sentiment seem less controversial between the two platforms for Asian authors: the only two examples of high divergence are Amin Maalouf (0.702 on LibraryThing; 0.186 on Anobii) and Vandana Shiva, who obtained the lowest sentiment on Anobii with − 0.632 while on the LibraryThing it scored 0.437.
In this paper, we presented a quantitative analysis of readers’ reviews about Asian and African writers on Anobii and LibraryThing. The analysis shows that diferences emerge in the reception of authors on the two platforms, both quantitatively and in the term of the sentiment expressed by reviews. Such diferences may be interpreted as country-specific, since Anobii is mainly frequented by Italian readers while LibraryThing is more international. However, a common aspect between the two communities of readers is that they both give more attention to Africanborn and Asian-born writers with Western origins. Future work will be devoted to widening the analysis with a higher number of platforms and writers, in order to provide tools and resources for exploring country-specific receptions of literary works. American Chapter of the Association for Computational Linguistics (Demonstrations), Association for Computational Linguistics, Minneapolis, Minnesota, 2019, pp. 54–59. URL: https://aclanthology.org/N19-4010. doi:10.18653/v1/N19-4010. [18] C. Hube, F. Fischer, R. Jäschke, G. Lauer, M. R. Thomsen, World literature according to wikipedia: Introduction to a dbpedia-based framework, arXiv preprint arXiv:1701.00991 (2017). [19] S. Schweter, A. Akbik, FLERT: Document-level features for named entity recognition, 2020. arXiv:2011.06993.