Book recommendation services in the bibliographic activities of libraries

Full Text

Book recommendation services in bibliographic activities of libraries¹

There are many services that help users choose books and other types of publications based on their personal interests, preferences, and habits. These include sites with recommendations and ratings of books, for example: LiveLib (https://www.livelib.ru), BookMix.ru (https://bookmix.ru), AvidReaders.ru (https://avidreaders.ru) , Readly (https://readly.ru), ReadRate (https://readrate.com/rus), Goodreads (https://www.goodreads.com), Skoob (https://skoob.com) and others Services that use artificial intelligence algorithms are developing, such as Semantic Scholar (https://www.semanticscholar.org), a free publication search tool for researchers that indexes more than 200 million scientific publications from reputable partner publishers (Science, Springer Nature , Wiley, Wolters Kluwer and others). Recommender technologies are actively used in scientific social networks as a means of communication for the rapid exchange of both bibliographic and full-text information between individual researchers and entire teams. Book recommendation services have become available through mobile applications, for example, the MTS service "Stroki" (https://stroki.mts.ru), which recommends books based on user genre preferences.
Let's consider how these services are used in Russian libraries. The rapid development of recommender technologies in various information systems has led to a situation where, in order to find the necessary and sufficient information, readers do not have to refer to the bibliographic resources of libraries. In order to increase their demand, to reveal the fund to the fullest extent and, as a result, to expand the user's reading circle, to stimulate his cognitive activity, it is important to introduce recommendation services into library practice. In this regard, the purpose of the study is to consider the possibility of using services in library and bibliographic activities that provide recommendations on the choice of books, as a means of disseminating scientific knowledge in the process of reading popular science literature. The following tasks contribute to the implementation of the goal:
1) determine the degree of study of advisory services for the selection of publications;
2) formulate a definition of the term "book recommendation service";
3) experimentally identify the need to use such services when creating recommendatory bibliographic resources aimed at disseminating scientific knowledge.
Literature review
In the RSCI, 240 publications were found for the query “recommendation services”, of which 79 were identified using a continuous scan, dedicated to recommendation services in libraries, while 48 were articles in journals. In total, the publications were cited 134 times. The total number of authors is 115. The largest number of publications belongs to I.Yu. Turchanovsky and A.A. Fomina - 4 works each.
The main number of publications (14) was published in 2019, while in 2001 and 2009 only one paper was written. The earliest publication, a didactic manual on library management for universities and colleges of culture and arts [1], dates back to 2001.
Most of the works from the selection are thematically distributed as follows: culture, cultural studies (26 publications); computer science, automation, computer technology (22 publications); public education and pedagogy (14 publications). This suggests that in domestic research on recommendation services in library and bibliographic activities, two main approaches have developed - humanitarian, including cultural and pedagogical areas (40 publications), and information technology, including computer science, automation and computer technology (22 publications).
More detailed information is provided by the keywords in the generated selection. There are 426 of them in total. Among the 10 most frequently occurring keywords identified by automated analytical tools of the RSCI, the phrase "recommender system" ranks first.
Recommender systems are understood as computer programs capable of “predicting objects of interest to a particular user” [2, p. 121]. This definition is given by Yu. S. Poletaeva. We find a similar interpretation by V.N. Kozub and I.I. Piletsky: “A recommender system is a system for searching and predicting materials that may be of interest to the user” [3, p. 277]. Note that today there are several types of recommender systems, depending on the principle of their operation.
1. Systems based on collaborative filtering use user preference data to recommend content that best suits them. This model is based on the assumption that users who are interested in certain materials will also be interested in similar ones.
2. Recommender systems based on content filtering - use information about the properties and characteristics of materials available in the system to recommend new documents. Characteristics can be authorship, subject matter, genre, language, etc.
3. Recommender systems based on a hybrid model combine the two previous ones, which makes it possible to use both data on user preferences and information on the properties and characteristics of materials to obtain more accurate recommendations.
4. Interactive model-based systems - allow users to influence recommendations by giving them the ability to rate and comment on content and indicate which topics they are interested in.
5. Recommender systems based on machine learning (artificial intelligence) - automatically analyze and classify materials according to predetermined characteristics and properties, and these characteristics are set by the user himself.
These species have been formed gradually since the early 1990s. In 2015 M.C. Kim and C. Chen [4] conducted a bibliometric analysis of different areas of research on recommender systems and found that collaborative filtering is the earliest way (the user is offered something that has already been evaluated by others). Such recommendations are widely used in various fields (science, education, production, commerce). Over time, recommender systems began to use information about users from social networks, recommender systems and services based on content filtering developed. The latest trend of M.C. Kim and C. Chen named the development of better recommender systems, improving the accuracy of recommendations.
It should be noted that this trend is important from the point of view of the search capabilities of electronic catalogs of libraries. The user, when receiving a response to a request, often encounters the problem of redundant and irrelevant information. The solution of this problem is facilitated by the introduction of recommender systems and services in electronic catalogs and electronic libraries. Back in 2010, recommendation services were considered as a tool for the formation of both individual collections and the fund of an electronic library [5].
O.S. Kolobov et al. [6] proved the possibility of functioning of the library electronic catalog as a recommender service, provided that the catalog will work as a hybrid recommender system. P. Laforce and S. Ratté [7] described the successful implementation of a recommendation service based on collaborative filtering in the electronic catalog of the National Library and Archives of Quebec in 2017. The developers associated document data with bibliographic records in the MARC 21 format and analyzed search query histories, namely: user subscription histories, materials viewed by the user on other platforms, and subscription histories of other users interested in similar topics. The results showed that using a recommendation system adapted for public libraries improves the user experience with minimal impact on pre-existing automated information systems.
Recommendation services have not yet become widespread in domestic libraries, although the literature presents the experience of some Russian university libraries in their implementation, for example, in the scientific library of the National Research Tomsk State University [8], in the scientific and technical library of Tomsk Polytechnic University [9], scientific library of the East Siberian State Institute of Culture [10], in the scientific and technical library of the Irkutsk National Research Technical University [2]. It should be noted that in the catalog of the scientific library of Tomsk State University, the search is carried out on all electronic resources, namely databases simultaneously - local and in the public domain, as well as on the fund. At the end of the bibliographic description of each book or article there are options "Similar books" and "Other books by this author".
In F. Liu, S.P. Asaithambi, R. Venkatraman [11] describes the practice of creating personalized hybrid systems that help digital library users find suitable books based on their own interests and other users' ratings.
The study of models on the basis of which recommender systems are built has led scientists to address issues of quality and relevance of recommendations.
Assessing the relevance of recommendations by users is complex and multifaceted. It is studied both from the point of view of the addressee (for a scientist user or for an ordinary user), and from the point of view of the way it is received (through bibliographic resources, social networks, specialized sites for readers). In electronic libraries, searches are carried out mainly by title, author or keywords. And here one of the main problems in developing recommendations arises - insufficient understanding of the needs of a particular user and what he already knows. This can lead to the fact that the system will recommend books, articles and other materials that are not at all interesting to the user, but related to the subject of his request. To solve this problem, many experts, such as B. Amini et al. [12] suggest using the so-called "background ontologies". They are a set of knowledge that contain general concepts and relationships between them. The use of background ontologies allows the recommender system to better understand the interests and preferences of a particular user, based on knowledge about his habits and hobbies. This can help the system more accurately identify suitable materials and reduce false recommendations. In general, the development of recommender systems based on background ontologies can significantly improve the quality of recommendations and simplify the search for the necessary materials for digital library readers.
The search for ways to improve the accuracy of recommendations led to the study of their perception by scientists in social networks. Thus, in the work of E. Olshannikova, T. Olsson, J. Huhtamäki, P. Yao [13], criteria for the relevance of recommended publications are formulated:
- similarity of views, values, beliefs, research goals;
- mutual complementarity of the recipient and the sender of the recommendation (professional roles, skills, knowledge);
- personal and emotional compatibility of the recipient and the sender for direct cooperation (personally valuable cooperation);
- willingness to cooperate.
As for the relevance of recommendations for a wide range of users, they should be based on relevant, correct, unique, complete, structured data [14].
In foreign sources devoted to the quality of recommendations, the term "usability evaluation of recommender systems" has come into use, which includes the user's interest in the recommendation and overall satisfaction with the process of obtaining it. In 2001, a study was conducted among students at the University of California [15] comparing the ratings of users of recommender systems (book - Amazon.com, RatingZone, Sleeper and movies - Amazon.com, MovieCritic, Reel.com). It has been established that interest depends on the usefulness of the recommendation and trust, and satisfaction depends on the time spent on registration and receiving recommendations, as well as on the interface of the recommender system (amount of input required from the user, screen layout, graphics, navigation, color, instructions) . In addition, it has been found that recommendations received from friends are more trusted by users than recommendations from search engines.
The study of trust in recommendations by users continues to this day. Scientists seek to find out which recommendations are more preferable for readers: those created by recommender systems or people (friends, acquaintances, experts, groups of like-minded people from online communities or clubs, etc.). Some authors, such as Zhang H. et al. [16], contrast these types as “algorithmic recommendations” and “recommendations with a social source”, and the latter include: content created by others; online comments on a book or article; requesting reading recommendations from other readers or in a discussion group (reading club); film adaptations based on the book; videos created by other users with recommendations for reading; blog entries.
Not all authors support such a contrast, believing that recommendations from social sources serve as additional material for recommender systems. Indeed, users evaluate the books they have read, create lists for reading in the future in their personal accounts or profiles, thereby providing the recommender system with additional data for building models of reading behavior and further developing more accurate, personalized recommendations. Systems enriched with detailed information about readers are seen as programs that promote specific cultural selection. The work of such programs is studied on the material of websites (social networks in some sources) for book lovers, such as Goodreads [17]. It should be noted that the Goodreads site provides rich material for studying the degree of scientific influence of publications in the social and human sciences [18]. Recommendations, user reviews, and book ratings on Goodreads are considered as altmetrics that complement traditional bibliometric assessments [19].
A separate area is the study of recommendation services for scientists, such as DBLP, CiteSeerˣ, Microsoft Academic Search, ArnetMiner, Google Scholar, Semantic Scholar [20], etc. With some differences, these services are characterized by three properties: personalized search, data exploration, effective and scalable methods (for example, machine learning, graphical data processing, etc.) [21].
So, the main efforts of researchers are aimed at improving the quality of book recommendation services and the recommendations created with their help. At the same time, other important aspects have not been studied:
- the impact of user behavior on the effectiveness of mental algorithms;
- what factors determine the satisfaction of users with book recommendation services;
- how to avoid showing recommendations of publications that are potentially dangerous, harmful to the reader or contrary to his moral and ethical principles;
- how compliance with the rules on the protection of personal data and confidentiality may affect the use of book recommendation services;
- what information and communication technologies and tools can improve the quality of recommendations for the selection of books and other publications.
Definition of the term "recommendation book service"
In the works of the authors who offer definitions for the term "recommendation service", there is no single approach. AND I. Mudrina, L.R. Mendigazieva [22] and A.S. Punda [23] by this term means "electronic resources of a recommendatory and evaluation nature" [22, p. 171]. We believe that it is difficult to agree with such a definition, since a service (service) by its nature cannot be a resource. Recommendation services are, rather, tools for generating a response in the form of a list of sources from available resources. In the above formulation, the method of obtaining the product is replaced by the final product itself.
AND I. Mudrin and L.R. Mendigazieva, understanding electronic resources as advisory services, chose social networks as the object of research, namely the scientific social networks Socionet, Mendeley, SSRN. Their authors also consider them recommendation services: “At present, social scientific networks perform the functions of recommendation services for scientists” [22, p. 172]. We believe that this approach can be accepted with some reservations. First, recommendation services, as noted above, usually use various algorithms to determine the relevance of the recommended information. In social networks, such algorithms are not provided. The content is mostly displayed in chronological order. Secondly, the types of connections between users can be different (personal, business, about an event, etc.), and not all of them involve recommendations. We can only talk about recommendations as one of the functions of the network community and social indexing of content, which are implemented through scientific social networks [24].
The most reasonable approach can be considered O.S. Kolobova, A.A. Knyazeva, Yu.V. Leonova, I.Yu. Turchanovsky [6], who consider recommender service as a form of functioning of a recommender system based on the ABIS already developed in the library. Indeed, ABIS eventually accumulates information about users, their information needs, interests, degree of activity - everything remains in the history of search sessions. The automated library system already has some recommending features, for example, it can suggest books similar to the one that the user ordered or viewed. When creating a recommender system based on ABIS, these functions can be supplemented with more complex recommendation methods, such as collaborative filtering and content analysis. The recommender system uses this information and, through recommender services, makes the library and bibliographic service individualized.
In our opinion, a book recommendation service should be understood as the function of a recommender system for selecting bibliographic information about publications that are most interesting and useful for a particular user or group of users.
Using book recommendation services to disseminate scientific knowledge by means of bibliography
In order to identify the need for the use of book recommendation services in bibliographic resources, from October to December 2022, a study was conducted by the method of questioning the readers of the State Public Library for Science and Technology of the Siberian Branch of the Russian Academy of Sciences in the amount of 209 people.
Respondents were asked to choose one of four options for presenting information in a recommendatory bibliographic resource, namely an electronic thematic collection of popular science literature on the website of the State Public Scientific and Technical Library of the Siberian Branch of the Russian Academy of Sciences (http://www.spsl.nsc.ru/news-item/populyarno-o- nauchnom). The variant was evaluated by a particular reader as preferable and sufficient, in terms of the bibliographic information contained in it.
Option A included a minimum of information about the book: author, title, place and year of publication, name of the publisher, number of pages. Option B contained extended bibliographic information - information about age restrictions, illustrations, series, number of sources in the list of references, number of copies. This option provided for the possibility to order the book from the electronic catalogue. Variant C, in addition to those listed in Variant B, included a brief annotation of the edition. Option D is similar to B, but is provided with a table of contents and a link to the book recommendation services LiveLib and Bookmix with an extended annotation, reviews, reviews, reader and expert ratings assigned to the book, with information about the number of readers and the plan those who want to read this edition, as well as about online stores where this book can be bought. Each of the four versions included an image of the book cover.
The distribution of survey participants by age is as follows:
- up to 21 years old - 7% (15 people);
- 21-30 years old - 20% (42 people);
- 31-40 years old - 24% (50 people);
- 41-50 years old - 27% (57 people);
- 51-60 years old - 14% (29 people);
- 61–70 years old - 6% (12 people);
- over 70 years old - 2% (4 people).
As you can see, the largest group among the respondents were readers aged 41-50. The group of 31-40 years old is close to them in size.
The choice by readers of the forms of presenting information about publications in a recommendatory bibliographic manual aimed at disseminating scientific knowledge is as follows:
- A - 3% (7 people);
- B - 11% (23 people);
- B - 56% (117 people);
- G - 30% (62 people).
This suggests that users, turning to a recommendatory bibliographic resource in order to expand knowledge on a particular topic, for a deeper understanding of a particular scientific problem, choose a book primarily on the basis of an annotation, and secondly, on the basis of ratings and ratings from other readers.
Answering the question about what information is missing in the submitted versions, respondents (11%) noted that a more detailed annotation is needed, supplemented by information about the author and comments from other readers. In addition, the answers indicated that we needed an introductory fragment from the book and information about its availability in the catalogs of other libraries. In some questionnaires, bibliographic information from options B and C was rated by readers as redundant, while option A turned out to be sufficient.
It should be noted that only 27% of respondents use recommendation services when choosing books, of which:
- up to 21 years old - 4% (8 people);
- 21-30 years old - 10% (22 people);
- 31-40 years old - 1% (3 people);
- 41-50 years old - 6% (12 people);
- 51-60 years old - 4% (8 people);
- 61–70 years old - 1% (2 people);
- over 70 years old - 0.5% (1 person).
Based on the data obtained, it can be assumed that the age of readers affects the use of book recommendation services. Mostly they are young people under 30 years old.
The survey showed that readers, turning to a recommendatory bibliographic resource in order to expand their knowledge on a particular topic, expect that, in the ideal case, the information in it will be presented as follows:
- the cover image is accompanied by an expanded set of bibliographic information necessary for identifying the book - the author, title, place and year of publication, publishing organization, quantitative characteristics (volume, illustrations, circulation), information about the series, the availability of bibliography;
- detailed abstract, supplemented with information about the author;
- through links to sites with book recommendations, you can see the ratings, ratings and comments of other readers;
- it is possible to go to the catalog and order a book, and in the absence of a free copy, order it through the catalogs of other libraries;
- access to the table of contents and a text fragment for introductory reading.
So, the survey showed that when choosing scientific and educational literature, readers have a need not only for extended bibliographic information about it, but also for services that provide recommendations on choosing the appropriate publications. When deciding whether or not to read a specific book on a particular scientific problem, library users are guided not only by the annotation, but also by the ratings, comments, ratings of other readers.
When creating bibliographic resources aimed at disseminating scientific knowledge, it is important to use book recommendation services for another reason. This will allow libraries to keep bibliographic resources up to date, improve the quality of bibliographic information by strengthening the analytical component, create personalized recommendations, and save time for readers to search for relevant information scattered across different sources, contribute to regular replenishment of the list of found documents on a topic of interest, expand the range of reading, as well as a deep understanding of the topic through multi-level search using related data.
In general, it can be stated that, firstly, one of the promising areas of library and bibliographic activities can be the inclusion in the recommendatory bibliographic resources of links to the websites of reputable publishers, online bookstores, scientific social networks, to sites with reviews and reviews, such as LiveLib , Goodreads, ReadRate, Bookmix, etc. Another direction could be the placement of recommendatory bibliographic resources (lists, indexes, reviews, thematic collections) on social networks.
Secondly, resources created using recommendation service technologies can not only increase the appeal to the library collection, but also develop the skills of selecting and critically evaluating sources of information by users, and also provide an opportunity to readers indirectly (through reviews and ratings, suggestions regarding the content content and the form of information presentation) themselves to participate in the creation of recommendatory bibliographic resources.
Thirdly, the experience of developing bibliographic resources using book recommendation services can be useful for libraries of various types, primarily public and university ones. It seems that this is a promising area of advisory bibliographic activity aimed at disseminating scientific knowledge.

About the authors

Olga L. Lavrik

State Public Scientific and Technical Library of the Siberian Branch of the Russian Academy of Sciences

Author for correspondence.
Email: ollavrik@rambler.ru

Doctor of Pedagogical Sciences, Professor, Head of the Information System Analysis Laboratory, Chief Researcher

Russian Federation, 15 Voskhod Str., Novosibirsk, 630102

Anna V. Yuklyaevskaya

State Public Scientific and Technical Library of the Siberian Branch of the Russian Academy of Sciences

Email: nad83770559@yandex.ru
ORCID iD: 0000-0001-9837-9423
SPIN-code: 3028-9706

Junior Researcher at the Scientific Bibliography Department, Postgraduate Student

Russian Federation, 15 Voskhod Str., Novosibirsk, 630102

References

Supplementary files