plenLiteracka Sp. z o.o.    |   Olsztyn, ul. Lanca 2/5
+48 883 00 88 22

Automatic book reviewer

Dr. Marek Maruszczak, R&D manager at Literacka, will speak at the conference Media Business Culture. Pomerania 2021, which will take place on October 14-15, 2021, where it will present the course and results of Literacka’s work on an innovative book recommendation system. Below we present an announcement of the speech that will be given during the symposium.

The aim of Literacka’s research was to create and implement a prototype of a book recommendation system using automatic content analysis by artificial intelligence, i.e. with the use of neural networks and natural language processing methods.

Traditional recommendation systems used in online bookstores, but also in libraries, are based primarily on three main methods: content filtering, collaborative filtering and association rule mining . The collaborative filtering method uses the so-called the wisdom of the crowd, that is, it is based on the assumption that similar users make the same decisions. The content filtering approach focuses on the analysis of the user’s subject of interest (content). The recommendation system based on this assumption proposes items similar or in a different relation to items that were previously of interest to a given recipient. The third of the most frequently used methods is the so-called basket analysis . The so-called baskets, i.e. shopping lists of various customers or readers.

The research problem was to create a recommendation system based on the so-called soft plot characteristics . For this purpose, the software has implemented algorithms that assign the written content an overtone rating based on a set of ratings for individual books.

The research that led to the creation of the automatic reviewer was the first on this scale to research long texts in Polish. For their needs, a database of 60,000 book titles in Polish was created, 30,000 full-text electronic editions obtained in cooperation with Publio.pl, including 10,000 titles in the category of non-fiction and fiction for adults. These books were later used to teach artificial intelligence algorithms. The current version of the automatic reviewer is able to “read” a book in about 5 seconds and on that basis evaluate its soft features with an effectiveness of 90% of human resemblance. In the assessment of genre affiliation (comedy, crime fiction, etc.), artificial intelligence is even more effective than humans.

The created recommendation system also fulfilled its task, increasing the conversion (visiting the purchase / rental page) on the partners’ websites by as much as 65% . Work on the system has led to the development of two completely new recommendation strategies: Gravity Mean and More Expert, the effectiveness of which has been confirmed in test implementations on three websites: codoczytania.pl, Poszamksiazki.pl and Publio.pl – in the form of a plugin. The newly developed strategies turned out to be so promising that research into their effectiveness and other applications will continue.

Related Posts

Leave a Reply