kotobli: a homegrown experiment in resisting dominant algorithms

If you search for books about Lebanon on the internet —not even the Lebanese civil war—, you are most likely to find war books by two white men, Robert Fisk and Thomas Friedman, on top of many lists. Why not Emily Nasrallah and Etel Adnan? These two women who were born and raised in the region, two great authors who have skillfully and beautifully written about their lives and their communities, rarely show up in general searches.

Book searches on the internet are biased towards the “Big Five” publishers based in the global North — their publications account for more than 75% of the annual bestsellers in the U.S. If you don’t have a book title in mind and are browsing for your next read by genre or topic, it's difficult to find a genuinely good recommendation that is not Western centric. If you’re searching in Arabic, the feat becomes near impossible.

In the cacophony of new technologies, and the overwhelming incessant flow of book recommendations on all and any social media platforms, how can authors find a genuine and equitable space to showcase their work? And how can readers find a good book that moves them?

We, at kotobli.com, set out to fill the gaps in representation, and to give Southwest Asia and North Africa (SWANA) authors the chance to be discovered, read, and loved. We embarked on this journey because of our own need for this tool to exist. Before kotobli, we were frustrated at how difficult it was to find a good book recommendation online, especially about our own communities. We were frustrated at popular book recommendation platforms suggesting the same orientalist books about war when we marked an interest in Arab history.

kotobli (all lowercase as a nod to its Arabic origins), is a book discovery platform and database for authors, books, publishers, and literary magazines from the SWANA region and its diaspora. We are committed to the intentional process of centering and amplifying the work of small authors and publishers from our native region, especially highlighting their work on our own communities and struggles. We aim to challenge existing algorithms and digital economies by setting up anti-colonial recommendation algorithms, exploring sustainable business models that would give back to book creators, and experimenting with new technologies that serve the readers’ best interest without exploiting the work of authors.

In the cacophony of new technologies, and the overwhelming incessant flow of book recommendations on all and any social media platforms, how can authors find a genuine and equitable space to showcase their work? And how can readers find a good book that moves them?

Re-imagining the book ecosystem is a big feat, and we’ve been trying to break it down into smaller, hopefully achievable steps: building a database, writing recommendation algorithms, finding sustainable business models, and finally, integrating new AI technologies to the benefit of readers while fairly compensating and crediting authors.

The database challenge

Many SWANA-based publishers don’t have a digital catalog, let alone a website or online presence, either because they don’t have the human financial resources, or because they have little to no tech literacy. As a result, there is no comprehensive and well-indexed database of books from and about the region.

To solve our database issue and simultaneously support rather than marginalize small publishers, kotobli decided to invite them to collaboratively build this database. We are offering a free webpage to small SWANA-based publishers in which they can access a simple and safe content management system where they have agency in how their books are displayed on our platform. The first to take interest were those with no websites, but soon, bigger publishers with a strong online presence followed suit. With a grant from Culture Resource, a regional cultural non-profit, we have been able to digitize the metadata for 2500 books, and 15 publishers in the Levant and North Africa over a period of four months.

Snoubar Bayrout was among the first small publishers that we added to our website. It’s run by Hala Bizri, one of our bookish heroes who has patiently given us feedback as we perfected our tools. For 10 years, Hala recently told us, she had been trying to get a website up and running but it was difficult for her as a small publisher. “kotobli solved this problem,” she said. “Our webpage is now within a platform with many more books, so people can stumble upon us by coincidence,” she added. Most importantly, Hala has been able to manage this website through the content management system. “It is nice and organized. I can add pictures, excerpts from the books” (which we enabled per her request).

We have recently opened our virtual doors to authors and literary magazines as well, allowing everyone to claim their profiles, add their books, and even curate reading lists if they wish. Slowly but steadily, we have been populating our database and setting the foundation for our online book community.

Creating Recommendation Algorithms

Generally, to facilitate discoverability, book platforms and online libraries tag books based on genres (e.g., fiction), topics (e.g., 1977 Egyptian bread riots), and locations (regions, countries, or cities). The question of recommendation then boils down to answering the following two questions: which books and tags to recommend to our readers? And in what order do we show the books in each tag?

We aim to challenge existing algorithms and digital economies by setting up anti-colonial recommendation algorithms, exploring sustainable business models that would give back to book creators, and experimenting with new technologies that serve the readers’ best interest without exploiting the work of authors.

Generally, existing book recommendation algorithms are based on the similarity in users’ reading patterns. If you and X have read two similar books, the algorithm will recommend the third book that X read. This is problematic because the algorithm naturally reflects the bias of the user base, which often leads to the overrepresentation of western authors and trendy books that have big marketing budgets regardless of their quality and relevance to you. This creates almost a cultural colonialism where a privileged minority dictates what anyone anywhere is recommended regardless of their own interests. And in the context of tags, these algorithms would recommend topics that the dominant user group of the platform is interested in. For example, if a particular user shows interest in books on politics and war in the SWANA region on Goodreads, they will be recommended books about 9/11 (based on our own experience on the platform).

At kotobli, we do not rely on user generated pattern in this way. Instead, we are building an algorithm that takes into consideration, not what others have read, but what the books you read have in common. Books on our database are tagged with genres, topics, geographies and author identities. So if, for example, you read history books and novels about Lebanon, you'll be shown historical fiction set in the civil war. We currently also have an algorithm that prioritizes local authors in topics related to their own communities. For example, if you are looking for books about Lebanon, first you'll see books by Lebanese authors, then Levantine authors (Palestinians, Syrians, Jordanians), then authors from SWANA, then others. Robert Fisk is at the bottom of the list. We are considering applying this not only to nationality but also to ethnicities, gender and sexuality (if disclosed by the author) so that marginalized authors have a chance of showing up in searches about books that tackle their own identities and communities. This is not to say that “outsiders” can never write good books about a certain issue, but it is to counter the dominant algorithms who barely give “insiders”  the spotlight or others/fetishizes them when they do showcase them.

The implementation of our algorithm has not been simple because it not only requires proper book tagging, but also relies on information about the book authors. This means that the majority of online databases that only store information about the book will not be able to implement nuanced recommendation algorithms like ours. Instead, they end up implementing simple but tired algorithms such as the one mentioned above. One way book platforms and online libraries circumvent the algorithms is to offer list curation pages where the books are manually chosen. But this is not an ideal solution as, (1) it’s difficult to create and maintain a list on each topic, and (2) it doesn’t solve the problem at all, and users can still end up seeing algorithmic recommendations instead of curated lists (which are not always properly highlighted and accessible). As far as we are aware, kotobli is the only platform that implements a general culture-sensitive recommendation algorithm.

Book curations

Our thematic curations also gives women, queer authors, and underrepresented authors a spotlight on our homepage and in relevant topic, genre and geography pages. For example, we chose to highlight the “Beirut Decentrists” in our very reading list within a larger curation on the Lebanese civil war. We dug into academic texts and browsed through many books to highlight books by the Beirut Decentrists; who are iconic underappreciated women who lived the war in Beirut and wrote about it from different geographic, intellectual, national lenses.

If you are looking for books about Lebanon, first you'll see books by Lebanese authors, then Levantine authors (Palestinians, Syrians, Jordanians), then authors from SWANA, then others. Robert Fisk is at the bottom of the list.

One of the lists that most people thanked us for (privately and publicly) has been the “Queerness in the SWANA region,” a comprehensive but incomplete list of books by lesbian, gay, queer, and trans Arab authors. One person sent us a message saying, “It lit up my heart seeing this section.” We have touched the thirst for the representation of our communities in their fullness, and the love and joy with which it has been received and shared.

More recently, feminist author pascale ghazaly helped us curate a list on the works of and about May Ziadeh, an early Levantine feminist, author, and journalist who is often left out of our literary lessons. Earlier this year, pascale published an engaging and inspiring graphic novel about May Ziadeh, and this sparked our interest in collaborating. Aside from curating a list around May, we also published an interview with pascale about her process, the influence of feminist methodologies, and the “right to narrate.” The conversation was really insightful and we were delighted to learn more about May Ziadeh, and the importance of correcting common misconceptions that even we had about her. At the same time, we were pleased to see that pascale’s book reached new and larger audiences through our platforms; exactly our goal.

Unanswered questions

The lack of diversity is not the bookish world’s only woe. Recently, more and more readers are weary with current options of where to buy books as the predominant corporations threaten the survival of small independent bookstores, and e-book platforms force authors to settle for low pay. The Arabic language industry is not safer; though the data is lacking, we have anecdotally heard from authors and publishers that even local platforms replicate exploitative models that pocket the profit from readers without fairly paying book creators.

In parallel, we are witnessing major advancements in Artificial Intelligence capabilities especially in parsing “unstructured data” like images and texts. With the advent of ChatGPT and other similar tools, authors and artists now risk losing their jobs to tools that were likely trained on their own writing and art —mostly without consent or compensation. 

At kotobli, we believe there is better way to sustain a platform, and we also believe there must be a better way to create, train, and use AI. We are currently considering building an AI tool which would “speak” Arabic, and answer smart search queries as well as summarize, compare and extract information from books on all things literature. Of course, before we embark on such a journey, we are considering the ethical before the technical: how can we ask for consent from authors to use their writings in the training of this AI? And how would copyright work? How can we compensate them fairly? And how can we sustain our project while experimenting?

We don’t pretend to have all the answers yet but we promise a genuine reflection on the matter.