Introduction

Quantitative analysis of the bibliographic elements contained in formal scientific publications can be applied to determine the intellectual structure of research specialities (Small and Griffith 1974). This study explores the library and information science (LIS) output attributable to authors working at Spanish institutions from 1985 to 2014. The purpose is to define nationwide research subjects and thematic trends in this domain and identify any changes taking place across those years.

Library and information science was institutionalised in Spain in 1975 with the creation in the country’s primary research body, the National Research Council, of three research institutes, one each for scientific information and documentation-related education, research and service provision (Plaza 2014). Library and information science was acknowledged as a university discipline in 1978. Education in the field began in 1981–1982 at ‘university schools’ (comparable to US junior colleges) at the Universities of Barcelona and Granada, with other universities following suit in the nineteen nineties. At this time, the discipline is taught in 17 public and private Spanish universities (Moneda-Corrochano 2016).

The importance attached to research on library and information science output in Spain has grown steadily from the outset. Spanish production from 1985 to 1994 was mapped in 1998 by Moya-Anegón et al. (1998), using co-citation techniques when the discipline was still in a stage of pre-development. The first 17 years of scientific production in Spain (1977–1994) were the object of bibliometric analysis by Cano (1999). Other quantitative studies (Jiménez-Contreras 2002) or papers on specific factors such as co-authorship in listed Spanish LIS research (Ardanuy 2012) have been published in the interim. The contribution and impact of Spanish institutions’ output and their publication practices has been compared with those of other European Union countries (Olmeda-Gómez and Moya-Anegón 2016). A recent analysis of national production in a worldwide context found that authors affiliated with Spanish institutions ranked third in a total of 27 countries (Walters and Wilder 2016).

This article analyses the co-occurrence of entities such as keywords or jointly cited bibliographic references (i.e., co-citations) in scientific articles. The year 1985 was chosen as the start date for an earlier study (Arquero Aviles 2001) on author co-citation in Spanish library and information science output covered the period 1975–1984, although the literature analysed in that Ph.D. thesis was not indexed in the WoS.

The aim was to determine: (1) the core knowledge around which LIS research revolved in Spain in the years studied by creating co-keyword networks to ascertain frequency and proximity; (2) the temporal changes in the trends in research subjects; and (3) the structure of the intellectual base and research front identified on the grounds of co-citation networks.

The article is organised as follows. “Review of the literature” section describes the two most widely used approaches to identify knowledge structures in scientific fields. “Materials and methods”section explains how the data were gathered and the techniques and metrics used to analyse LIS articles attributed to Spanish institutions. The results of co-keyword and co-citation analysis are listed in “Results”section. The article concludes with a discussion of the results and pointing out some conclusions.

Review of the literature

To date, bibliometric studies have used primarily two methods to determine the subjects explored in a given research speciality: co-word analysis and document co-citation analysis (DCA). Both build on academic networks based on similarity and can be generated for different units of analysis (authors, journals, institutions or scientific fields).

In co-word analysis, content is explored through the co-occurrence of pairs of terms or lexemes (such as words or phrases) in a corpus of papers. Macro- or ‘signal-words’ (Rip and Courtial 1984) can be mathematically aggregated by building indexes that count the frequency of co-occurrence of terms in a given group of papers. The co-occurrence rates of the terms chosen are assumed to show the strength of the relationship between them. These pairs of terms may be grouped into clusters of the co-words that tend to be the most frequent in the subject area analysed and subsequently mapped. If the words appear together in many articles, the community of authors may in all likelihood be thought to see some manner of logical connection between them (Small 2003). The terms represented on these maps express and channel the conceits, interests and problems common to a given domain at a given time (hot topics), as well as areas of minority interest and the vocabulary associated with the subject (Callon et al. 1983, 1991).

Document co-citation analysis (Small 1973) is based on the citation in a given paper (X1) of two previously published ‘co-cited’ works (A and B). The strength of this co-citation bond is determined by the number of times that other citing articles (X2, X3, X4) jointly list the cited articles (A and B) in their references. The papers cited can be quantitatively pooled and represented on a graph, for instance, where they are shown as nodes and their inter-relationships as connecting arcs. Texts cited jointly are assumed to have some degree of thematic similarity or cognitive relationship (Garfield et al. 1978). Reference clusters with the highest co-citation values have been identified as the intellectual base for a given speciality (Persson 1994) and to represent and symbolise the conceptual structure of the corpus (Small 1980). The variety and validity of such methods have been examined (Gmür 2003), along with their most prominent variations, author (Rousseau and Zuccala 2004) and journal (Ding et al. 2000) co-citation analysis.

Many authors have analysed specialities using such traditional techniques either exclusively or in combination (Chang et al. 2015; Liu and Mei 2016) or by comparing them (Yan and Ding 2012) with others. Recent examples of co-word analysis can be found in research on nutrition (Blázquez-Ruiz et al. 2016), information metrics (Ravikumar et al. 2015; Sedighi 2016), the internet of things (Yan et al. 2015), recommendation systems (Hu and Zhang 2015), marketing (Wang et al. 2015), institutional repositories (Cho 2014), fuzzy set theory (López-Herrera et al. 2009), consumer behaviour (Muñoz-Leiva et al. 2012) and the economic transition in eastern and central European countries (Topalli and Ivanaj 2016), to mention but a few of the high diversity of subjects addressed.

Other recent studies have used document co-citation analysis to identify the knowledge base for specialities including transport geography (Liu and Gui 2016), organisational ambidexterity (García-Lillo et al. 2016), intercultural relations (Chi and Young 2013), innovation (Shafique 2013), substance abuse (González-Alcaide et al. 2016), corporate information systems (Shiau 2016), magnetic nanoparticles (Liu et al. 2016) and business and intra-company trust (Yang 2016), among others.

Both co-word and document co-citation analysis are characteristic of “knowledge domain analysis”, a library and information science topic of research. “Knowledge domain analysis” is defined as an interdisciplinary area of research geared to mapping, content mining, classifying, analysing knowledge and enabling its presentation and navigation (Shiffrin and Börner 2004).

Materials and methods

Data

The search conducted in the Web of Science core collection yielded 2247 papers of all types published in journals classified under the WoS heading ‘Information Science and Library Science’ in which at least one of the authors was affiliated with a Spanish research institution, university or body. Of those, only the ones classified as articles (2209) were selected i.e., conference communications, reviews, notes and others were excluded. The citations analysed were the ones contained in the Web of Science’s core collection when the papers were downloaded (2 September 2016).

The co-word and co-citation analyses of Spanish output were conducted with Java Citespace software (Chen 2004), which supports the construction and visualisation of bibliographic record networks (Chen 2006) and is suitable for visualising relative small ones. Each network, built for the number of years defined as the study period, was composed by the application to analyse the knowledge domain. Inasmuch as the fundamentals and details of these techniques are well known and fully explained in Chen et al. (2010), only the specific procedures deployed in this study are described below.

Co-keyword analysis

In this first procedure, the keywords listed in the set of 2209 bibliographic records were retrieved. No pre-processing of the keywords included in the fields DE e ID have been done. CiteSpace does not mix the terms in the keywords fields. We believe nuances might be lost if they were merged. Merging synonyms is potentially more computationally consuming than what you aim to achieve with your analysis in the first place. It is not cost-effective, you won’t gain any more insight by merging them. In this particular study, analyzing the intellectual landscape of a specific scientific field, we decided that extra pre-processing was not worthwhile. The same approach is followed in similar works (Chen et al. 2014; Kim and Chen 2015).

The duration defined for all the networks, 1985–2014, was divided into 1 year intervals (“time slices” in Citespace jargon). Each network was formed from the 50 articles most frequently cited in each interval. The co-word frequencies in the matrix were converted using the value of the cosine as the similarity measure (Leydesdorff 2008). The networks were subsequently merged into a single entity in which the links were pruned where necessary using the Pathfinder network technique. Pathfinder network scaling simplifies the co-word matrix by retaining the strongest co-word links with reference to the triangle inequality condition (Schvaneveldt 1990). The Kamada and Kawai (1989) algorithm was applied to visualise the spatial organisation of the co-keyword node graph.

Burst

A second procedure to determine thematic trends consisted in identifying the noun phrases contained in the titles and abstracts of the articles in the record base created. The terms appeared at a given time in the period analysed in the wake of their intense use by the research community. These ‘bursts of activity’ were detected with Kleinberg’s (2002) algorithm. Instead of simple frequencies, this algorithm deploys a probabilistic “automaton” whose states are conditioned by noun phrase frequencies. Inter-state transitions are the time points around which term frequencies change significantly. The algorithm generates a list of the strongest noun phrase bursts found in the set of records, as well as the years when they appeared.

Document co-citation analysis

Spanish LIS document co-citation analysis was conducted by building co-citation networks for the references contained in 2202 articles (the software excluded seven articles for unknown technical reasons). It detected the structural subjects informing the speciality. Co-citation counts were used to calculate the similarity between pairs of references. The interval, selection thresholds and link normalisation criterion were the same as used to build the co-keyword network. The largest connected or main component on the network built was selected for subsequent analysis.

Spectral clustering (Luxburg 2007) was applied to that main component to subdivide the network into clusters or modules, subsequently subjected to dual analysis. The references cited in each cluster with the highest co-citation values were identified as the intellectual base for the speciality and the subjects of the citing articles as the research fronts.

Two procedures were used to summarise research front subjects on the grounds of the descriptors for the keywords of the citing articles. The first consisted in listing subjects based on the keywords with the highest term frequency-inverse document frequency (tf*idf) value (Salton and McGill 1986). In the second, keywords were retrieved using the log likelihood ratio, comparing the observed to the expected keyword frequencies based on contingency tables generated from the corpus (Dunning 1993). The clusters on the network graphic created as a result of these operations were labelled with the keyword having the highest inverse document frequency.

Metrics

The analytical and representational procedures used to create networks included the calculations supported by the software to generate a series of metrics. (1) A betweenness centrality value, which is the proportion of all the geodesic distances between network pairs including the keyword at issue, was calculated for each network node (Freeman 1977). A keyword with a high value meant that it occupied a significant position as a network intermediary and was associated with a large number of other keywords. (2) Q modularity is a quality measure that expresses the goodness of the overall network breakdown into clusters (Newman 2006). A high value for this parameter means that the initial network was divided into clusters in which the constituent nodes were closely interconnected and any weak inter-cluster links were deleted. A low value means that a given network division failed to locate more node clusters than would have appeared randomly (Shibata et al. 2008). (3) The silhouette value indicates which terms fit well into the clusters and which are located between several, without assignment to any. Silhouette values denote the relative quality of the breakdown into clusters. The mean silhouette value assesses clustering validity, in which perfect network subdivision is 1 and the worst possible −1 (Rousseeuw 1987). (4) Sigma is a metric that combines the value of a structural measure, such as betweenness, with a time-related node property, burst strength. When used in co-keyword network analysis to identify the most novel co-keywords, it is defined as: (betweenness centrality + sim1)burstiness (Chen et al. 2009).

Results

The yearly output of authors affiliated with Spanish institutions publishing in journals listed under Journal Citation Reports’ (JCR) category ‘Information Science & Library Science’ is shown in Fig. 1. Annual output never exceeded 35 papers in 1985–2004. That number was surpassed for the first time in 2005 and from 2007 onward, production never dipped below 130. This abrupt rise was largely due to the provisional inclusion in JCR of two Spanish professional journals, El Profesional de la Información and Revista Española de Documentación Científica, prior to their permanent inclusion in 2008 and 2010. These publications are major vehicles for Spanish LIS research (Olmeda-Gómez and Moya-Anegón 2016).

Fig. 1
figure 1

Number of articles with Spanish affiliations published yearly under the WoS category ‘Information Science & Library Science’

Full size image

Table 1 lists the universities and institutions publishing the largest number of papers in the journals analysed in this study. With the sole exception of Catalonia’s Universitat Oberta, a private institution, only public universities are listed. The universities in the highest positions (1–13) all had departments or faculties delivering library and information science courses. The top spot was held by the University of Granada, the first Spanish university to offer a degree in this area, beginning in the nineteen eighties. The only non-university on the list was the Spanish National Research Council (CSIC). Specific centres and institutions under the aegis of the CSIC are responsible for gathering and managing information on science, technology and the humanities and employ staff whose research is geared to studies on science and technology.

Table 1 Number of Wos category ‘Information Science & Library Science’ papers published by Spanish institutions, 1985–2014
Full size table

The data on the two networks built are given in Table 2. For the co-keywords, the individual networks were merged into one with 407 co-keywords. The initial network had 1566 links. The proportion of actual to possible links was a very small 1.9%. The largest connected sub-graph grouped 356 co-keywords or 87% of the total identified. The silhouette values suggested good clustering and the modularity value denoted good but not exceptional interconnections among them.

Table 2 Co-keyword and document co-citation networks (both based on 1 year intervals)
Full size table

The co-citation network was built from an analysis of 2202 records, which yielded 51,674 valid (97.5% of the total) and 1297 (2.4%) invalid (incomplete and therefore excluded) references. This single network had 4746 nodes and 10,506 links, for a network density of 0.0009. The main component had 931 nodes, or 19% of the total.

Thematic trends. Co-word analysis and burst detection (1985–2014)

The thematic trends informing the domain were defined from the frequencies of the co-keywords most used by researchers. As the data in Table 3 show, the main poles of attraction for researchers working in the domain in the period analysed revolved around the terms ‘science’, ‘Spain’, ‘internet’, ‘citation analysis’, ‘impact’, ‘indicators’ and ‘university’. With such abstract terms, research subjects can only be identified on a macro-scale. The betweenness values for each co-term might reflect the power of the individual term to match subordinate subjects. ‘Information’ (0.44), ‘internet’ (0.36) and ‘citation analysis’ (0.29), for instance, exhibited the highest values.

Table 3 Most frequent keywords (frequency over 15), 1985–2014, in descending order
Full size table

An additional way to illustrate the subjects preferred in the Spanish domain is by representing the keywords as a network. Figure 2 shows part of the largest component in the Pathfinder-pruned co-keyword network, with the most frequent terms highlighted. Adjacent terms were the ones with a high joint assignment rate. Near the top of the graph, for instance, the terms ‘absorptive capacity’, ‘information technology’, ‘firm’ and ‘knowledge management’ are positioned close to one another. The terms most frequently used are circled, with purple outer circles identifying terms with structural betweenness value. The thicker the outer ring, the higher the value.

Fig. 2
figure 2

Co-keyword network of Spanish library and information science papers, 1985–2014, containing 407 nodes and 1566 links in 2209 records. Pruning with Pathfinder. Purple rings: keywords with high betweenness. (Color figure online)

Full size image

The presence of concurrent terms on the network denotes relative consensus about the macro-scale research issues addressed. Network density values are an indirect measure of the degree of diversity and cohesion. The low values found here were an indication that consensus in the use of common elements related to the subjects contained in the articles was also low. This is an initial sign that scientific discourse in this domain is only weakly integrated.

A more detailed breakdown of the noun-phrases or terms retrieved by CiteSpace from the title and abstract fields in 2202 records and derived from the burst analyses described above is given in Table 4. Citespace generates networks of co-occurring terms or noun phrases. This is undertaken using a Part-of-speech (POS) tagging algorithm. By means of this procedure, the informative content of the concepts is represented. Studies show that such n-grams preserve far more semantic content than individual term extraction (Ding and Chen 2014). By furnishing more specific information on single words, these findings facilitate interpretation, for such phrases can be more readily linked to conceits (Schneider 2006).

Table 4 Noun phrases with the strongest bursts (in bold type, duration >3 years)
Full size table

The period studied, 1985–2014, was divided into three sub-periods to determine the thematic trends in each, including the most recent years based on the publication trends observed in the data analysed (Fig. 1). The first period, 1985–1993, can be regarded as a preliminary stage, followed in 1994–2003 by development, in which papers began to be written in the university departments created in the early nineteen nineties. The last sub-period may be regarded as a growth or consolidation stage characterised by a sustained rise in output.

The earliest sub-period lists noun phrases found in a Spanish journal (Revista de Informática y Automática) classified under the subject category until 1989 when it ceased publication. The phrase ‘information professionals’ first appeared in 1988–1990. In the second (1994–2003), high burst strength values were found for ‘electronic library’ (1.87), ‘scientific journal’ (1.93) and ‘information retrieval system’ (1.74). The most recent sub-period confirmed consolidated subject trends, such as ‘library science’ (3.24) and others relating to science more generally: ‘scientific research’ (5.09), ‘scientific production’ (3.14), ‘scientific output’ (3.93), ‘scientific collaboration’ (2.88), ‘scientific information’ (4.8) and ‘scientific field’ (4.76). Note the high strength value for ‘social media’ (5.4).

Figure 3 shows the subjects with the highest structural values that also attracted researchers’ attention for more time, according to their burst activity scores. They included ‘information retrieval’ (3.9), ‘co-citation’ (2.36), ‘information technology’ (1.73), ‘database’ (1.5), ‘web’ 2.0 (1.4), ‘world wide web’ (1.3), ‘Scopus’ (1.28), ‘semantic web’ (1.09), ‘bibliometric indicator’ (1.07), ‘digital library’ (1.12), ‘research performance’ (1.04), ‘information literacy’ (1.04), ‘information science’ (1.04), ‘impact factor’ (1.03) ‘media’ (1.03), ‘web site’ (1.03), ‘map’ (1.03), ‘scientometrics’ (1.01) and ‘social network’ (1.01).

Fig. 3
figure 3

Co-keyword network of Spanish library and information science papers with the highest sigma values. 1985–2014. Purple rings: keywords with high betweenness. (Color figure online)

Full size image

Document co-citation analysis: research fronts and intellectual base (1985–2014)

The research fronts and intellectual base for the speciality were identified using document co-citation analysis for objective classification and independent and automatic generation. The boundaries were set on the grounds of the breadth of the period studied, the size of the clusters resulting from the breakdown, the validity of that process as measured by the silhouette values for each cluster and the consistency of the subject contents identified and the research front labels.

The breakdown of the co-citation network into nine sub-specialities is given in Table 5. The 771 references involved accounted for 82.8% of the total contained in the largest connected network. The silhouette values indicative of the relative validity of the clusters were very high, denoting a clearly separated structure. Values of 1 mean that the respective cluster was perfectly separated and uniform. The silhouette values for each cluster varied depending on the size of the sets of references from which it was derived. Somewhat lower values were found for clusters with a larger number of references, such as clusters #0 and #1, for their internal structure and contents were more diverse. The most recent clusters were identified on the grounds of the average year of publication.

Table 5 Number and size of co-cited reference clusters (minimum ≥59)
Full size table

The clusters were described on the basis of the keywords in the articles citing the references classified (Table 6), in turn used to label the research fronts active during the period. Assuming the same p value, the identifiers with the highest log likelihood ratios (LLR) were the most unique, i.e., the most suitable labels.

Table 6 Major co-cited reference clusters (1985–2014)
Full size table

The four sub-specialities labelled generically (#1, #3, #4, #6) (Figs. 5 and 6) exhibited research metrics that differed fairly slightly. Five other sub-specialities were also identified and labelled using the term frequency-inverse document frequency formula, as follows: #5, ‘electronic books’; #7, ‘information systems’, ‘structured document retrieval’, ‘information retrieval’, ‘indexing services’ and ‘hypertexts’; #2, ‘translation and abstracting services’; #8, ‘world wide web’, ‘maps’, ‘information retrieval’ and ‘author co-citation analysis’; and #0, ‘digital rights management’, ‘copyright law’ and ‘intellectual property’.

Fig. 4
figure 4

Document co-citation cluster timeline

Full size image
Fig. 5
figure 5

Networks of co-cited references, with clusters #1 and #6 as the intellectual base

Full size image

The four clusters with references related to scientific information metrics have prevailed to date. They also contain the most recent references and may have developed further after the end date for this study. The aforesaid references constitute the primary intellectual base for Spanish LIS. Of these four clusters, described in greater detail below, only #1 (citation analysis | similarity measure) and #6 (webometrics | bibliometrics) contained references with citation bursts (the nodes with red concentric rings in Fig. 6).

Fig. 6
figure 6

Networks of co-cited references, with clusters #3 and #4 as the intellectual base

Full size image

Cluster #1 was the second largest, with 108 references and mean year 2007. The five most cited papers are listed in Table 7. J. Hirsch’s well-known study on the h-index stood out above all the others, which dealt with bibliometric rankings, the g-index, assessment bibliometrics and the application of citation analysis to social science and humanities output. This was the cluster with the largest number of citation burst papers.

Table 7 References with the highest citation frequency in cluster #1
Full size table

Cluster #3 (Table 8), with a mean year of 1996, contained 89 papers, including some by Spanish authors. The following terms appeared in the titles: ‘national journals’, ‘international collaboration’, ‘multi-disciplinary research’ and ‘journal internationalisation’.

Table 8 References with the highest citation frequency in cluster: (a) #3, (b) #4
Full size table

The mean year for the 88 references in cluster #4 (Table 8) was 1996. They addressed subjects similar to the ones found for cluster #3: ‘citation analysis problems’, ‘research fronts’, ‘education study bibliometrics’, ‘domain analysis’ and ‘authorship delimitation’ (Figs. 5, 6).

Cluster #6, identified by the term ‘webometrics’, focused on quantitative studies of websites and traffic. The most frequently cited references, listed in Table 9, addressed ‘link analysis’, ‘website interlinking’, ‘academic and university web analysis’ and the ‘Google Books impact measure’.

Table 9 References with the highest citation frequency in cluster #6
Full size table

Table 10 shows the seven references with the strongest citation bursts listed in the 50 articles with the strongest bursts in the period studied (see Table 12). The Hirsch (2005) paper proposing the h-index as an indicator for individual researchers’ scientific output, which drew much international attention, had the strongest citation burst, with a score of 18.3.

Table 10 Seven references (only first authors’ names shown) with the strongest citation bursts, 1985–2014 (bursts detected from an analysis of all the citations in the 50 articles most cited in each year in the period) (* beginning and end of period when activity was most intense)
Full size table

Table 11 shows the twelve references with the most recent citation bursts, from 2012 onward. In some cases, citation bursts began in 2012 for papers published earlier. The highest burst strength was recorded for a paper by Moed (2010), who proposed a new indicator denominated the source-normalised impact per paper (SNIP) to measure journal impact based on citations. Zitt and Small (2008), with the second highest score, introduced a new normalisation strategy known as the audience factor and Albarrán et al. (2011), with the third, published a study on the distribution of citations and references by categories and subject sub-fields. In the fourth ranked paper, Radicchi et al. (2008) proposed a method to compare the impacts of articles in different disciplines. Thelwall and Sud’s (2011) article, with the fifth strongest burst, compared web link counts delivered by two alternative methods to measure organizations’ online impact. The sixth, by Aguillo (2010), introduced website assessment metrics, and the seventh by Bordons et al. (2010) analysed the relationship between research findings based on Spanish universities’ publications and structural factors and socio-economic data.

Table 11 Twelve references (only first authors’ names shown) with the most recent citation bursts, 1985–2014 (bursts detected from an analysis of all the citations in the 50 articles most cited in each year in the period) (* beginning and end of period when activity was most intense)
Full size table

Discussion and conclusions

This paper identifies the subjects addressed in information science and library science literature authored by researchers working out of Spanish institutions by building co-keyword and document co-citation networks. The main subjects are represented graphically by discipline in a national context. The burst activity algorithm is deployed to detect the noun phrases attracting the greatest attention and ascertain the evolution over time of the subjects of greatest interest. The articles with the largest and most recent citation bursts included in the intellectual base for the Spanish domain are also identified.

Globally speaking, library and information science rests on a combination of essentially three major subject categories: library science, information retrieval and bibliometrics, although the first is vanishing (Lariviere et al. 2012). The present analysis reveals that in Spain, all the sub-specialities and hot topics fall under one of those three categories, with no deviations relative to the worldwide domain.

The document co-citation analysis and subsequent characterisation conducted here detect nine sub-specialities. These fronts represent the degree of consensus around a given combination of conceits found in the corpus analysed. Four of these are inter-related and readily characterised. They concur with the research fronts detected in global analyses under the heading ‘analysis of scientific literature’ (Zhao and Strotmann 2008, 2014).

The references cited in the ‘scientific literature analysis’-related fronts and the ones with the strongest bursts (Table 12) are the result of convergent interests among researchers working out of Spain’s largest public scientific body, the National Research Council and, once the discipline was institutionalised, out of the country’s universities. They also include papers by outstanding non-Spanish bibliometricians.

Table 12 Top 50 references with the strongest citation bursts, 1985–2014
Full size table

The intellectual content of the oldest ‘scientific literature’ research fronts (#3 and #4) has been updated and diversified. This is attested to by the intellectual base and the subjects of interest. The use of references is more intense and uniform. The fact that nearly all the references with the strongest burst factors in the Spanish domain fall under the two most recent fronts (#1 ‘citation analysis’ and #6 ‘webometrics’) denotes the existence of shared cognitive objects with their respective discourses and symbolisms. Such circumstances make it easier to assess the nature and significance of the findings (Whitley 2000), for they reveal a certain intellectual identity.

Based on the mean year of the references in the intellectual base, the remaining research fronts (Table 6, clusters #0, #2, #5, #7 and #8) are of an intermediate age. Here the interpretation of keyword retrieval and subsequent labelling is less clear-cut. ‘Information retrieval’, for instance, appears in three clusters, in the company of different mixes of other phrases. In cluster #7, where its log likelihood ratio is highest and consequently the most representative, it is identified with terms such as ‘indexing services’ and ‘hypertext’, but not others related to the use of information and characteristic of LIS globally such as ‘information behaviour’ or ‘information seeking’. The detection of two related fronts, world wide web (#8) and webometrics (#6), denotes the interest in the technical activities supported by this technology and in the activities, situations, people and social practice interconnected by these technical artefacts.

The duration of the burst periods for the noun phrases related to these clusters is brief, denoting low stability in the attention paid to the subjects formulated by the Spanish community of citing researchers. Only five of the 50 references in the intellectual base are not related to at least one of these biliometric fronts (Table 12: Maybee 2006; Ranganathan and Ganapathy 2002; Guallar and Abadal 2009; Abadal and Guallar 2010; Subramaniam and  Youndt 2005). Moreover, the presence of several media—(‘social media’, ‘digital media’, ‘communication’, ‘television’, ‘web 2.0’), semantic web- and information literacy-related terms with high sigma values is an invitation to speculate on the possible configuration in the near future of one or several research fronts in these connections.

Overall, semantic differentiation is weak and the references fairly old in all the sub-specialities. Network density values are indicative of low relationship density. From that standpoint, the domain was thematically ‘open’ in the period studied. That may be a sign of fragmented and weakly related cognitive structures in which mutual dependence is not particularly high. These are features associated by Whitley (2000) with specialities in which the degree of technical tasks performed is uncertain.

The main component of the co-citation network covers only 19% of the references. Therefore, although research fronts emerge in this study, most of the research does not lie within thematic structures. The inference may be that the research skills with which the findings are generated and the materials from which they are drawn are not standardised or generalised across the domain. That in turn may denote a prevalence of local research cultures.

One limitation to this study lies in the use of a single source of data, the Web of Science, which lists peer-reviewed journals meeting the citation level requirements in the respective disciplines. A second is related to the deployment of both traditional techniques and others which the authors intend to continue using in future research. That notwithstanding, the simultaneous use of a series of techniques, including structural and content analysis indicators, automatic generation of cluster labels and detection of the subjects appearing with highest intensity in a given period, helps understand the recent past and prevalent subject trends in Spanish library and information science research.