The association of disciplinary background with the evolution of topics and methods in Library and Information Science research 1995–2015

The paper reports a longitudinal analysis of the topical and methodological development of Library and Information Science (LIS). Its focus is on the effects of researchers' disciplines on these developments. The study extends an earlier cross‐sectional study (Vakkari et al., Journal of the Association for Information Science and Technology, 2022a, 73, 1706–1722) by a coordinated dataset representing a content analysis of articles published in 31 scholarly LIS journals in 1995, 2005, and 2015. It is novel in its coverage of authors' disciplines, topical and methodological aspects in a coordinated dataset spanning two decades thus allowing trend analysis. The findings include a shrinking trend in the share of LIS from 67 to 36% while Computer Science, and Business and Economics increase their share from 9 and 6% to 21 and 16%, respectively. The earlier cross‐sectional study (Vakkari et al., Journal of the Association for Information Science and Technology, 2022a, 73, 1706–1722) for the year 2015 identified three topical clusters of LIS research, focusing on topical subfields, methodologies, and contributing disciplines. Correspondence analysis confirms their existence already in 1995 and traces their development through the decades. The contributing disciplines infuse their concepts, research questions, and approaches to LIS and may also subsume vital parts of LIS in their own structures of knowledge production.


| INTRODUCTION
In Library and Information Science (LIS for short), the share of authors with disciplinary background outside LIS has grown markedly (Chang, 2019;Chang & Huang, 2012;Lund, 2020;Urbano & Ardanuy, 2020;Vakkari et al., 2022aVakkari et al., , 2022b. For example, Chang (2019) found, based on 75 LIS journals, that non-LIS authors had the majority in about 70% of the journals. This multidisciplinarity in practice likely leads to two-way impacts between LIS and the other contributing disciplines. For LIS' part this means, among others, influencing which LIS topics are studied and which methodologies are applied. In the present paper we explore longitudinally the relationships between LIS article authors' disciplinary backgrounds, the research topics, and research methods.
Content analysis is widely used to analyze LIS research topics (Aharony, 2012;Asubiaro & Badmus, 2020;Ma & Lund, 2021) and research methods (Asubiaro & Badmus, 2020;Ma & Lund, 2021). However, only few studies have examined the relationship between study characteristics and the disciplinary backgrounds of study authors-even fewer longitudinally. Vakkari et al. (2022a) is a cross-sectional study on these relationships. Applying correspondence analysis on the year 2015 they found three clusters of research, one focusing on traditional LIS with contributions by authors from Humanities and survey-type research; another on information retrieval (IR) with contributions from Computer Science and experimental research; and the third on scientific communication with contributions from Natural sciences and Medicine and citation analytic research. However, a cross-sectional study gives only a weak basis for analyzing trends. In the present paper, we adopt the study approach by Vakkari et al. (2022a) and extend their study to a longitudinal one by adding compatible datasets from years 1995 and 2005. This facilitates trend analysis.
Following Vakkari et al. (2022a), we understand scholarly contributions in LIS as research publications, especially as LIS journal articles, which often are crafted in collaboration with scholars of varying disciplinary backgrounds. These backgrounds influence the construction of new LIS knowledge. Authors' affiliations in each article suggest their disciplinary affiliations at the time of writing (Chang, 2018). Such a cognitive affiliation means sharing a given domain of interest, metatheoretical assumptions, and methodological ideas. These are likely to change when the background of the author population changes. Järvelin and Vakkari (1990) presented a widely used classification of LIS research into a hierarchy of research topics (e.g., Armann-Keown & Patterson, 2020;Hider & Pymm, 2008;Ma & Lund, 2021;Tuomaala et al., 2014). The extensive use of the scheme hints that it is conceived as a valid structuring of LIS research in the research community. An updated version (Järvelin & Vakkari, 2022) lists the main research topics as follows: LIS context, L&I services, information retrieval, information seeking, scientific communication. Table B1 of Appendix B lists the classification, providing a working definition of the scope and structure of LIS.
In addition to changes in research topics, we shall analyze possible changes in research strategies. By research strategy we mean the abstract methodological approach to performing a study above the level of specific methods of data collection, etc. Table B1 of Appendix B lists this classification as well.
Our main research question is: How have the associations between contributions to LIS from various disciplinary backgrounds and the choice of research topics and methodology evolved from 1995 to 2015? Our specific research questions are as follows: • How did each discipline's overall share of contribution vary from 1995 to 2015? • How did each discipline's share of contribution vary in each LIS main research topic from 1995 to 2015?
• How did each research strategy's share of contribution vary in each LIS main research topic by discipline from 1995 to 2015? • Using correspondence analysis (CA), how did the relationships between the contributing disciplines, main topics, and research strategies develop from 1995 to 2015?
These RQs are longitudinal versions of those by Vakkari et al. (2022a) and approached using a longitudinally coordinated dataset bridging the years 1995, 2005, and 2015. Therefore, the answers to these RQs provide a much more reliable basis for analyzing and debating the development of LIS and trends likely to continue, and their scholarly driving forces.

| LITERATURE REVIEW
LIS has long been conceived as an interdisciplinary field with contributions from various disciplines (Saracevic, 1992). Many central figures of LIS have their disciplinary roots in other disciplines like Bradford (mathematics), Farradane (chemistry), Garfield (linguistics), or Mooers (physics). It is evident as Wilson (1983, p. 390) claims that "the entrants from the other fields brought naturally their tools with them as well as their expectations about the proper way of tackling problems in the bibliographic sector." Thus, methodological-theoretical assumptions common in external disciplines have influenced in LIS for decennia. Vakkari et al. (2022a) have made a cross-sectional analysis of how authors' disciplines are associated with the topics and methods used in research articles in 31 leading LIS journals in 2015. They presented a comprehensive literature survey on this topic. Our longitudinal analysis uses the data from 2015 by Vakkari et al. (2022a) but extends the analysis to also cover the years 1995 and 2005. To avoid unnecessary repetition in literature review, we present the major characteristics of studies dealing with disciplinary background, topic, and method (Table 1).
The studies reviewed show that the share of contributions to LIS from other disciplines has increased. However, the comparison of findings is somewhat limited due to the variation in the unit of analysis. In most of the studies the unit of analysis is authors' discipline, and in some studies the disciplinary composition of authors in an article. The former ones calculate distributions of authors' disciplines, while the latter ones analyze the distributions of articles with author compositions. Also, the number of journals as the source of articles varies greatly. In addition, most of the studies are cross-sectional, only two are longitudinal. Only four studies analyze the topics and two the research methods applied. Chua and Yang (2008) reported a decline from 61 to 48% in the share of LIS authors in the articles published in JASIST between 1988 and 2007, while the share of authors representing information technology increased from 16 to 25%. Chang (2018) showed that between 2005 and 2014 the share of articles authored by non-LIS scholars increased from 31 to 38%, while articles authored by LIS scholars decreased from 55 to 49%. The share of articles by LIS authors was on average 47%. Of non-LIS authors, 39% belonged to Computer Science and 19% to Business.
Other studies reported cross-sectional findings with the share of LIS authors from 27% (Aharony, 2012) to 50% (Chang & Huang, 2012). In most of the studies Computer Science was the second largest contributor to LIS and Business the third. The share of Computer Science varied from 8% (Chang & Huang, 2012) to 18% (Urbano & Ardanuy, 2020) and Business from 7% (Chang & Huang, 2012) to 15% (Prebor, 2010). Using contribution instances as the unit of analysis, Vakkari et al. (2022a) showed that LIS contributed 36%, Computer Science 21% and Business 16% of all cases.
Despite the variation in the characteristics of data in studies reviewed, it seems that disciplines external to LIS are increasingly contributing to LIS, especially Computer Science and Business. Although most of the studies do not observe topics and methods, Vakkari et al. (2022aVakkari et al. ( , 2022b show that the preferred research topics and methods vary according to authors' disciplinary background. Scholars representing LIS focus on traditional professional topics, LIS context, and L&S services, while Computer scientists on information retrieval and information seeking and Business scholars on information seeking and scientometrics.

| METHODOLOGY
The construction of the dataset for the year 2015 has been reported in Järvelin and Vakkari (2022) and Vakkari et al. (2022a). The same process was followed in the construction of datasets for 1995 and 2005. The former dataset is completely new and the latter partially new-the  (Järvelin & Vakkari, 2022;Vakkari et al., 2022aVakkari et al., , 2022b. Below we give a concise description of the content analytic methods and emphasize the longitudinal ones. The characterization of research in LIS requires a valid definition of LIS, which can be used for selecting articles for analysis. Although it may be challenging to give a definition of LIS shared by all scholars, it is widely accepted that the unifying characteristic of LIS is the study on the provision of access to desired information (Vakkari, 1994). This brief definition does not provide sufficient tools for operationalization, but rather opens the view to LIS. The definition of LIS is nowhere available for picking up like a ripe fruit. It is always a construction. We are explicit about the proposed scope of LIS by providing an ostensive definition of LIS through the classification scheme for LIS topics (Table B1 of Appendix B). Publications whose topic can be placed within its classes belong to LIS. This classification has been adopted in several studies analyzing the trends in LIS research (e.g., Hider & Pymm, 2008;Ma & Lund, 2021), which hints that it is considered by the research community as valid characterization of research topics in LIS. Wide adoption also makes comparison of findings based on different datasets easier.

| Data collection and classification
Each annual dataset was collected in two main phases ( Figure 1). The first phase involved selection of journals to represent LIS in 1995represent LIS in , 2005represent LIS in , and 2015 (Table A1 of Appendix A), identification of scholarly articles in the journals, and performing content analysis of the articles (see the content classification scheme in Table B1 of Appendix B). In the second phase, each annual dataset was extended by data on disciplinary backgrounds of the article authors based on their affiliations. Statistics on the dataset are given in Table 2.
Journal and article selection. To facilitate longitudinal analysis, the journal set used as the source of articles should remain representative and comparable through the years. A partial approach toward this is to use the same journal set for each year. This is not perfect, since across long time spans some journals die and new ones are born, some are joined, some drift outside/inside LIS. At the same time, LIS develops, and some topic considered external to LIS at an early time point may be a focus area at a later point-the original content and structure of LIS may turn obsolete over the years. The speed of such changes depends on the changes in the disciplinary backgrounds of the authors. An alternative method, sampling articles from a bibliographic database (e.g., LISA) is likely no better: the editors change and judge, they also select journals and identify LIS articles based on their views which are bound to change over the years. This said, the journal/article set collected is reasonable while the findings derived depend on the nature/biases of this set.
Noting name changes, there are altogether 33 journal titles and 29-31 for each year in the dataset. For 5/33 titles (15%) there is less than full 3-volume coverage. When ranked by LIS article productivity, 25/33 journal titles (76%) having 3-volume coverage, contribute 94% of the units of observation. The annual datasets are well compatible and represent their year's LIS research.
We collected full articles, brief communications, and critical reviews-using digital versions of the journals whenever possible-and excluded other types as well as articles falling outside LIS. A total of 2,552 articles published in 31 LIS journals in the 3 years were collected initially for the longitudinal analysis. Because many articles had more than one contributing discipline, the total number of units of observation-identified by the pairs (a-id, cd) of article identifier a-id and contributing discipline name cd-grew up to 2,866.
Content analysis of each article was based on title, abstract and keywords, or title and first page, depending on what was available. In insufficient cases the entire article was skimmed. Each article was classified into one content class for each content dimension by two of the authors and a third expert, all broadly experienced in LIS. Table 3 lists the content variables used in the present study, and Table B1 of Appendix B their classes.
The classification of research topics is presented in Table B1 of Appendix B. We used the five main research topics for analysis: LIS context, L&I services, information retrieval (IR), information seeking, and scientific communication. The topic class non-LIS research was coded but excluded from the analyses. Classification reliability was measured by Fleiss' Kappa (Table 4).
To increase the degrees of freedom in the analysis we merged classes of some variables. In research strategies, historical and evaluation strategies were merged with other empirical strategies as other empirical strategies; citation analysis was merged with other bibliometric strategy as citation analysis; verbal argumentation and concept analysis was merged as conceptual strategy; literature review and bibliographic strategy were merged with other strategy as other strategy.
Identifying authors' disciplines. Chang's method (Chang, 2018) was used to identify each author's discipline based on their affiliation information of articles. Authors who were affiliated with LIS-related institutions were coded as LIS authors. Other authors were classified as authors in Business-and-Economics, Computer Science, Engineering, Humanities, Medicine, Natural Sciences, and Social Sciences (Table C1 of Appendix C). The first affiliation of multi-affiliated authors was used to infer their disciplinary attributes. Such authors were infrequent, for example, only 6% in 2015 data.
Classification of articles' disciplines. Each article was classified into each unique discipline present among its authors. Because it is very difficult to measure the total contribution of an article, or the share of each author in producing it, each discipline was credited by one for each article. Because some articles were contributed by two or more disciplines, the total number of units of observation in disciplinary analysis rose to 3,078 (Table 2). Classification reliability was measured by Fleiss' Kappa (Table 4). Kappa value ranges from À1 for complete disagreement, to ±0 for random choices, and to +1 for complete agreement. Kappa values 0.41-0.60 are moderate, 0.61-0.80 good, and 0.81-1.0 very good.
Data analysis. The data matrix for analysis was constructed by combining the encoding of the authors' disciplines with the content analysis data. We report cross tabulations, and χ 2 significance test results. In addition, we visualize the relationships between the disciplines, topics, and research strategies through correspondence analysis (CA) (Hair et al., 2010). It plots variables and objects in a visual map based on their association. Visual proximity indicates their relative association and is not intended for making precise statements (Hair et al., 2010).
For the longitudinal analysis, we constructed annual cross-sectional maps for each year of data. These allow the analysis of the associations of variables and objects on each year and support the analysis of changes between the years.

| Research topics
During the years observed the share of pure LIS contribution decreased essentially from 57 to 27%, while the contribution of external disciplines correspondingly increased from 29 to 53%. The joint contribution of LIS with other disciplines increased somewhat from 1995 to 2005 stabilizing in 2015 on the level of one fifth. LIS scholars have dominated the production of knowledge in their discipline at least to 2005, after which representatives of other disciplines have taken the dominant role.
Although external disciplines contributed most to LIS in 2015, among individual disciplines the contribution of LIS has remained the largest (Figure 2). The contribution of LIS scholars has decreased notably from 67 to 36%, F I G U R E 1 The two phases of data collection, preparation, and analysis process (Vakkari et al., 2022a) while the contribution of scholars in major external disciplines has increased in the following way: Computer Science from 9 to 21% and Business and Economics (Business) from 6 to 16%. The contribution of Computer Science increased most between 1995 and 2005, while Business increased contribution most during the next decennium.
Next, we analyze the contribution of various disciplines to main research topics in LIS. Due to small number of cases, we combine Engineering and Medicine with Natural Sciences (other sciences), and Social Sciences with Humanities when analyzing information seeking, information retrieval, and scientific communication.
LIS was the largest contributor to LIS context with a decreasing share from 80 to 66% (Figure 3). Social Sciences and Humanities played a minor but increasing role. In 2015 Humanities scholars contributed to library history, while social scientists contributed to publishing.
LIS scholars dominated the contribution to L&I services with a declining share from 85 to 68% (Figure 4). Social Sciences and Humanities, Business and Computer Science increased their shares of contribution. In 2015 Business and Computer Science contributed to library automation and digital libraries, in particular.
LIS and Computer Science have been the major contributors to information retrieval. LIS dominated the research in 1995 with a share of 65%, while the share of Computer Science was 21% ( Figure 5). In 2015 Computer Science contributed most with a share of 48%, while the share of LIS had declined to 22%. The contribution share of Social Sciences and Humanities, and Business increased somewhat during the period 2005-2015.
In 1995 LIS dominated the contributions to each subtopic of IR with a share of at least 50%, while Computer Science contributed most to classification and indexing, text-retrieval methods, and retrieval methods in other media. In 2005 LIS contributed most with a share of at least 50% to metadata and cataloguing, digital information resources and interactive IR, while Computer Science dominated all other subtopics. In 2015 Computer Science dominated contributions to all subtopics except metadata and cataloguing, which were the focus of LIS scholars.
LIS has contributed most to information seeking with declining share from 67 to 43% (Figure 6). The change has been rapid after 2005. The contribution of Business and Computer Science increased notably during 1995-2015. In 1995 the former covered 12% and the latter 5%, while in 2015 the respective figures both were 19%.
LIS was still in 2015 the major contributor in all subtopics of information seeking with a share 40-50% except for information management, which Business focused on with a share of 30%. In addition, Computer Science contributed notably to task-based information seeking, and Business to other types of information seeking studies like serendipity or presence in social media.
During the period observed LIS also dominated the contribution to scientific communication, although not to the same extent than in other research topics. However, its contribution share declined from 40 to 24% (Figure 7). Also, the contribution of Social Sciences and Humanities to Scientific Communication decreased from 27 to 16%. The latter decline is associated with the decreasing interest in scientific and professional publishing. The share of other disciplines (Engineering, Medicine, and Natural Sciences) has been stable (22-24%) throughout the years, while the shares of Business and Computer Science increased notably. As a result, among the research topics in LIS, the contribution to scientific communication has distributed most evenly between

| Research strategies in topics
Next, we analyze differences in the use of research strategies between disciplines in each main research topic. Due to the small number of cases, we combined research topics LIS context and L&I services. The only notable difference between disciplines in the use of research strategies in these combined topics was the application of system analysis. In 2005, Computer Science used system analysis in 28% of cases, while LIS only in 2%. In 2015, the share of system analysis in studies by Computer Science was 15%, by other disciplines (Engineering,  Medicine, and Natural Sciences) 39%, and by LIS 2% (χ 2 , df = 45, 100.9, p < .000). Thus, toward the end of the period observed, Computer Science, Engineering, Medicine, and Natural Sciences applied significantly more system analysis compared to LIS. As our previous results hint, system analysis was applied to problems of digital libraries.
In information retrieval, research strategies varied by disciplines across the years. In 1995, the share of evaluation/experiment was 36% both in LIS and Computer Science, while the share of system analysis was in LIS 10% and in Computer Science 38% (χ 2 , df = 45, 108.1, p < .000). In 2005, the share of evaluation/experiment increased in both disciplines being 55% in LIS and 70% in Computer Science. Also, other disciplines (Engineering, Medicine, and Natural Sciences) applied evaluation/ experiment frequently (73%). System analysis was popular in Computer Science (21%) and in other disciplines (23%), but rare in LIS (6%) (χ 2 , df = 40, 101.4, p < .000). In 2015, the popularity of evaluation/experiment decreased in these three disciplines to the level of 25-30%, while the use of system analysis (23%) and mathematical strategy (32%) gained footing in Computer Science and mathematical strategy (36%) also in other disciplines (χ 2 , df = 45, 122.3, p < .000). Throughout the years LIS has preferred in IR experimental evaluation strategy as much as Computer Science and Engineering, Medicine, and Natural Sciences, but significantly less system analytic and mathematical research strategies. Thus, it seems that the grown influence of Computer Science in There was a divergent trend between the disciplines in the use of research strategies in scientific communication. In 1995 there were no significant differences between disciplines (χ 2 , df = 35, 35.4, p = .45). Citation analytic strategy was most popular in all disciplines. Its share was 53% in LIS, 64% in other disciplines (Engineering, Medicine, and Natural Sciences) and 57% in Sociology. In 2005 the share of citation analytic strategy had decreased to the level of 40% among these three disciplines but was significantly higher in Computer Science (64%) (χ 2 , df = 45, 62.0, p = .047). In the same year, the popularity of survey increased in LIS (28%), in other sciences (Engineering, Medicine, and Natural Sciences) (30%) and in Social Sciences (35%). Mathematical strategy was relatively popular in Business (18%), LIS and other sciences (18%).
In 2015 citation analytic strategy remained still popular on the level of 40%, although it lost popularity in Computer Science. Survey lost while case study strategy gained popularity in all disciplines, in Business (18%) and in other sciences (Engineering, Medicine, and Natural Sciences) (32%), in particular.
The variety of research strategies applied in scientific communication increased by the years. In 1995, citation analytic strategy was clearly the dominating one, and remained as dominating with diminishing share, although in the following years first survey, and then mathematical strategy and case study gained popularity. There are three clusters in each correspondence map a-c: Cluster I involves the topics LIS context, L&I services and information seeking which are close to survey, qualitative, case study, conceptual and other empirical research strategies-reflecting the variety of approaches in these topics. Cluster II focusses on the topic scientific communication with citation analytic research strategy and Social Sciences and other sciences (Engineering, Medicine and Natural Sciences) in its vicinity-echoing the application of scientometrics in many disciplines. Cluster III travels with the topic information retrieval which is close to evaluation and experiments, system analytic and other strategies-a realistic view on mainstream IR research. Interesting changes include:

| Correspondence analysis
• Cluster I: Business starts at the edge of Cluster I but travels to Cluster II in 20 years' time. • Cluster II receives mathematical strategy for a visit from the marketplace between the three clusters but hands it over to Cluster III in 2015. • Cluster III becomes quite closely-knit by 2005 after a dispersed beginning and receives mathematical strategy by 2015.
Overall, it seems that within the observed 20 years research clusters in LIS have differentiated somewhat. Clusters II and III around information retrieval and scientific communication consist of tighter associations with respective disciplines and research strategies. In addition, professionally oriented LIS Cluster I has differentiated internally to some extent between LIS context, L&I services, and information seeking.

| DISCUSSION
Our study is the first to analyze, longitudinally, the characteristics of research contributions to LIS by various disciplines. It confirms and elaborates findings from crosssectional studies (e.g., Chang, 2018Chang, , 2019Chang & Huang, 2012;Urbano & Ardanuy, 2020) which show that disciplines external to LIS are responsible for over half of the research articles contributing to LIS. The study by Vakkari et al. (2022a) analyzed in detail topical and methodological contributions to LIS by various disciplines but was cross-sectional, focusing on year 2015. We extend their analysis longitudinally covering years 1995, 2005, and 2015.

| Major findings
Our research questions were as follows: • How did each discipline's overall share of contribution change from 1995 to 2015?
The contribution of LIS scholars has decreased essentially from 67 to 36%, while the contributions by scholars in major external disciplines grew greatly: Computer Science from 9 to 21%, and Business from 6 to 16%. Scholars in LIS have dominated knowledge production in their discipline at least to 2005, after which external disciplines have taken the dominant role. Thus, currently most of the research in LIS is created by representatives of disciplines other than LIS.
• How did each discipline's share of contribution and their use of research strategies change in each LIS main research topic from 1995 to 2015?
Our findings indicate a growing differentiation between LIS research topics caused by uneven interest by various disciplines. Authors' disciplinary background differed essentially between major LIS topics. The contribution of Computer Science increased especially between 1995 and 2005, while the contribution of Business strengthened between 2005 and 2015. During these periods Computer Science focused on information retrieval, while Business on scientific communication.
Although the contribution of LIS decreased notably in LIS context and in L&I services from about four fifths to two thirds, LIS remained as the main knowledge producer in these profession-oriented topics. Humanities and Social Sciences increased their contribution somewhat in LIS context, and Social Sciences, Business, and Computer Science in L&I services. Interestingly, the two latter contributed especially to library automation and digital libraries. Computer scientists used system analysis significantly more frequently in the problems of L&I services compared to LIS scholars, which hints that it was applied in research on digital libraries and library automation.
In information retrieval, the contribution share of LIS decreased from two thirds to one fifth, while the share of Computer Science increased from one fifth to one half. LIS dominated in 1995 the contributions to each subtopic of IR with the share of at least 50%, while in 2015 Computer Science dominated contributions to all subtopics except for metadata and cataloguing, which were the focus of LIS. Computer Science has taken the leading role in all subtopics of IR, while LIS has focused on the professional subtopic classification and indexing. Evaluation/experiment has been the most popular research strategy through the years in both disciplines. In the later years, however, system analytic and mathematical research strategies have gained footing in Computer Science. Thus, Computer Science seems responsible for the F I G U R E 8 Correspondence maps for research strategies applied in topics by contributing disciplines in 1995, 2005, and 2015 application of these research strategies in the topic information retrieval.
During the years observed LIS has contributed most to information seeking, although its share declined from 67 to 43% especially after 2005. The share of both Social Sciences and Humanities, and Business has increased from one tenth to one fifth. In 2015 LIS was the major contributor in all subtopics of information seeking with a share of 40-50% except for information management, which was dominated by Business.
In scientific communication in 1995 the major contributors were LIS (40%), Social Sciences and Humanities (27%) and other sciences (Engineering, Natural Sciences, and Medicine) (23%). In 2015 LIS, other sciences, and Business contributed about one fourth each, while the share of Social Sciences and Humanities had declined to 16%. The variety of research strategies applied in scientific communication grew from the initial dominance of citation analysis to later varied application of survey, mathematical, and case study strategies.
In all, the early dominant role of LIS scholars in knowledge production in all LIS research topics has declinedmost dramatically in information retrieval (À43% units), but also in all other research topics (À14 to À24% units). LIS scholars were in 2015 still major contributors in LIS context and L&I services with a share of about 2/3, while in information retrieval their contribution share was only 1/5, bypassed by computer scientists with their share of about 50%. In information seeking the contribution share of LIS was in 2015 still the largest one, but in scientific communication both Business and other sciences (Engineering, Natural Sciences, and Medicine) caught up LIS.
With the strong shares of LIS knowledge production, other disciplines than LIS, especially Computer Science and Business, will certainly continue to bring their own concepts, research questions and approaches to reorient LIS. Computer Science and Business are big and have a mighty ally: the techno-economic interests in digitalization and the Web-for LIS it is better be agile. Historians may judge whether the likely reorientation looks like organic growth or perilous twist for the cognitive and social institutions of contemporary LIS. The former means continuation and the latter leads to fragmentation and/or subsumption. Figure 8 points to the road of fragmentation through visualization of the formation of the three clusters growing apart.

| On the road toward fragmentation of LIS
Although the contribution share of LIS was the largest of all disciplines, how long may it prevail after 2015? We calculated linear trends to 2025 based on percentual contribution shares of LIS and other disciplines between 1995 and 2015. Figure 9 shows the 1995-2015 datapoints for five disciplines and their linear trend to 2025. In 1995, LIS produced two thirds of the contributions while the other disciplines produced 6-9% each. Following clear trends, LIS's share drops to 18% while the others grow up to 15-26%. Even if the absolute volume of LIS contributions grows, their share drops within the estimated total 2000 contributions in 2025. Nevertheless, the message is clear: in the near future LIS will no more be the master in its own house. The other, bigger players will likely step in with their own agendas, approaches, and methodologies. If they do not fit in, the house of LIS will break.
In 2025 LIS scholars are no more likely the main knowledge producers in LIS but share this position with scholars from Computer Science and Business. In 2025, LIS may still be the major contributor in LIS context and services because of marginal interest by other disciplines.
Business and Computer Science evidently will enhance their roles in information retrieval, information seeking, and scientific communication. The contribution share of LIS discipline in information retrieval will probably marginalize. The major player will be Computer Science, which produced about half of the contribution in 2015. The growth trend likely continues.
Business and especially Computer Science increased their contribution between 2005 and 2015 in information seeking. If the trend continues, they challenge the leading role of LIS by 2025. The trend strengthens by the integration of research on information retrieval and information seeking based on digitalization of information and communication (Vakkari, 2008). This integration calls for concepts and approaches offered by Computer Science and Business.
Business shared with LIS in 2015 the position of dominating contributor in scientific communication. Assuming linear trends, Business will be by 2025 the largest contributor followed by LIS and Computer Science. The increasing use of mathematical research strategy likely continues in scientific communication favoring contributions by Computer Science.
Based on cross-sectional data from 2015, Vakkari et al. (2022b) reported that traditional professional topics are cultivated mostly by LIS scholars, while other disciplines contributed most to IR, scientific communication, and information seeking. They conjectured that LIS is under fragmentation, with IR and scientometrics on the road toward integration with their external contributing disciplines.
Our longitudinal results corroborate this. The fragmentation process has accelerated since 1995. Other disciplines, especially Computer Science and Business, produce an increasing share of research in information retrieval, information seeking and scientific communication. This suggests that fragmentation continues, leading to the integration of information retrieval with Computer Science. This is in line with Computer Science Classification by ACM, which has conceived IR as part of Computer Science since 1960s. It also produces scientific communication and scientometrics as a discipline of its own-or as part of Science Studies-with established social structures like journals and conferences. Figure 8 illustrates this as formation of clusters.
The research profile of LIS scholars has become over the years increasingly narrow. In 2015 it consisted of research areas LIS context, L&I services, and information seeking from traditional channels and sources including L&I institutions, and some minor subareas of information retrieval and scientific communication. We agree with Vakkari et al. (2022a) that scholars not affiliated with LIS maintain research in two major areas of LIS, information retrieval and scientific communication. Without their contribution, LIS would be narrower and more traditionally oriented toward L&I institutions. This trend may strengthen in future by including also information seeking. We are afraid that under pressure by larger disciplines, LIS shrinks into a marginal field of research focusing on L&I institutions with some minor exceptions. Information retrieval and scientific communication would become marginal parts within other disciplines.

| Limitations
There are several limitations in our study, many of which have been discussed in earlier articles using the dataset (e.g., Järvelin & Vakkari, 2022;Vakkari et al., 2022aVakkari et al., , 2022b. • The type of publications, journal articles, and the choice of journals might be questioned for biased representativeness. However, journal articles are regularly used for the analysis of disciplines, and a stable journal list greatly helps longitudinal comparability. • The authors of the articles, and thus their disciplines, are assumed equal contributions. This problem persists because: (a) the total contribution of an article varies F I G U R E 9 Linear trends of disciplinary composition shares in annual production of articles from marginal to revolutionary, (b) the order of authors is not systematically related to their contribution, and (c) the contribution types vary and are valued differently. • An author's affiliation may be unclear in indicating their discipline or there are several for one author. High reliability of the affiliation coding suggests that the former problem is negligible. Multiple affiliations were rare and, when present, often indicating the same discipline. • The content classification scheme (Table B1 of Appendix B) leads to aka WYSIWYG principle and its corollary WYDSIWYDU (What You Don't See Is What You Don't Understand). Topics and research methods not in the classification are not identified as areas of non-LIS contributions. In the long run an unchanging classification does not help identify organic growth of LIS from non-LIS developments. A slowly evolving classification helps longitudinal comparability. • Each article was assigned to one class of each content dimension, challenging the classification of multitopic or multimethod articles. To relieve the problem, the classifiers were instructed to identify the primary class among alternatives, the articles were considered at the level of main classes, and the class "multiple x's" was offered for some dimensions.

| Further studies
The current dataset covers 21 years, ending 2015. We look forward to the opportunity of collecting the 2025 dataset. It would be of interest to extend the annual datasets by data on the cited and citing literature and analyze, whether the hypothesized fragmentation process materializes in citations. Another extension would be the analysis of content-bearing title words of the articles which were assigned to the topic class other discipline.

| CONCLUSION
How have contributions to LIS, coming from various disciplinary backgrounds, been associated with the choice of research topics and methodology across decades? We used content analysis of scholarly articles published in 31 LIS journals in 199531 LIS journals in , 200531 LIS journals in , and 2015 to find the answer. The findings form a clear message: the share of LIS as authors' disciplinary background is shrinking from 67 to 36% while Computer Science and Business increased their share from 9 and 6% to 21 and 16%, respectively. The latter disciplines increased their share of contributions especially in information retrieval, information seeking, and scientific communication. These trends are likely to continue.
Consequently, Computer Science and Business are not only infusing their concepts, research questions and approaches to LIS but also likely to subsume vital parts of LIS in their own structures of knowledge production. The restructuring of information, media and commercial practices around the Web will cause a corresponding restructuring of the related research fields. In the cracking and frictions which are unavoidable in the pursuit of a new constellationboth cognitive and social-LIS is too small a player to determine the result. It better stay agile and reinvent itself in the new situation. The current trends point toward fragmentation of LIS as a discipline.