Historically, researchers have not fully capitalized on the potential synergies that exist between bibliometrics and information retrieval (IR). Knowledge of regularities in information production and use, as well as citation relationships in bibliographic databases that are studied in bibliometrics, can benefit IR system design and evaluation. Similarly, techniques developed for IR and database technology have made the investigation of large-scale bibliometric phenomena feasible. Both fields of study have also benefitted directly from developments in natural language processing (NLP), which have provided new tools and techniques to explore research problems in bibliometrics and IR. Digital libraries, with their full text, multimedia content, along with searching and browsing capabilities, represent ideal environments in which to investigate the mutually beneficial relationships that can be forged among bibliometrics, IR and NLP. This brief presentation highlights the symbiotic relationship that exists among bibliometrics, IR and NLP.
Both information retrieval (IR) and bibliometrics have long histories as distinct areas of investigation in information science. IR has focused on the storage, representation and retrieval of documents (text or other media) from the system and user perspectives. Bibliometrics and its allied areas (informetrics, scientometrics, webmetrics, or simply “metrics”) have focused on discovering and understanding regularities that exist in the way information is produced and used--but this simple definition belies the breadth of research undertaken. Units of analysis may be words, metadata fields, publications, authors, journals, research groups, institutions, sub-fields, disciplines or geographic regions. Applications of the study of these regularities and underlying processes extend to equally diverse areas such as science policy, subject indexing and IR system design, including digital libraries (DLs).
The documentary contents of bibliographic IR systems and their
associated indexes provide much of the data that metrics researchers rely
on today to conduct their research. Conversely, many of the processes
within IR are directly informed by metrics research. Surprisingly, there
has been little overlap in the research agendas of these two areas of
study over the history of information science. This is changing with the
growing recognition of the common interests and the tools and
techniques used by IR and metrics researchers that can help to advance the
research agendas of each field [
In a sense, digital libraries represent an ideal environment in which to study the intersection of IR, bibliometrics and NLP, whether the DLs consist of repositories of formal publications or heterogeneous collections of multimedia documents. With content representation and search functionality found in more traditional IR environments, hyperlinks that mimic relationships similar to those found in citation analysis, and full text contents that lend themselves to NLP analysis. This brief overview outlines how developments in IR and bibliometrics, along with language-based methods, have helped to advance research in both areas. These intersections are most evident in bibliographic databases, but also extend to more heterogeneous digital libraries. 2
IR research has focused on the design of more efficient and effective
systems to match documents to queries to meet the information needs
of users. Early IR research was more system-centered, but
usercentered approaches are now employed alongside system-centered
approaches in the design and evaluation of IR systems. Bibliometrics
provides useful methods for the analysis of both system content and
system usage, to better understand system processes and user dynamics. In
essence, IR processes represent forms of information production and
use [
Aspects of IR system content that lend themselves to bibliometric
modeling include index term frequency distributions, indexing
exhaustivity or term assignment, term co-occurrence frequency
distributions, database and index growth, and more recently, aspects of
webbased IR such as frequency distributions of inlinks/outlinks and
document persistence [
User interactions with IR systems as recorded by transaction logs
may be similarly modeled. Characteristics that can be modeled include
search terms used per query, frequency distributions of query terms,
query term co-occurrence, query frequency distributions, search session
length based on queries or other actions, user search and browsing
regularities, and site/document visitation frequency. Search and
browsing patterns may be modeled using network analysis methods, similar
to those applied in citation analysis, to identify issues with interface use
[
Beyond mathematical modeling of specific bibliometric
characteristics of IR systems, higher level aspects of science models also can
inform the search and retrieval process of scientific literature. Mutschke
and Mayr [
Citation-based connections between bibliographic records serve as a
readily exploitable data source for expanding searching and browsing
options for users. These ideas have been implemented in experimental
systems over the past several decades such as I3R [
For much of the history of IR, the research focus was on the representation and retrieval of document surrogates with limited natural language. Representation and retrieval were based on metadata fields, keywords or controlled vocabularies. Models to support query and document matching were based on bag-of-words approaches that treated terms independently of one another. Context and semantics played no role in determining the aboutness of documents. At the same time, computational challenges increased as the size of databases grew. IR models such as the vector space model became more computationally expensive due to the high dimensionality associated with representing documents and processing document-query matching.
IR research has been more responsive to taking advantage of
advanced NLP techniques to provide a more natural search environment
for users and to extend retrieval evaluation beyond simple term
independence models. Dimensionality reduction techniques that lower the
computational overhead associated with IR processing have become
commonplace since early approaches based on latent semantic analysis
were introduced over 25 years ago. Successors based on language
modeling [
Citation analysis has been a core area of bibliometrics research for
over 50 years. Citation relationships in the form of direct citations,
cocitations or bibliographic coupling create a network of linkages among
documents, authors, and publication venues that allows one to visualize
the intellectual structure of a field [
This situation can be addressed using approaches that integrate the
vocabulary used by the entities of interest, usually employing keyword
or controlled vocabulary co-occurrence, which takes bibliometric
studies more into the realm of IR research. Still, there are concerns with
keyword-based approaches in bibliometrics research that mirror the
same bag-of-words issues found in IR research. The assumption of
independence of vocabulary in bibliographic entities, whether as
keywords or subject terms, creates the same issues observed earlier in IR
studies. The ability to work with full text and natural language can
improve these analyses. More recent research has moved beyond simple
co-occurrence to identify relationships among entities of interest. Of
particular note has been the application of topic modeling techniques,
such as Latent Dirichlet Analysis (LDA) [
The application of bibliometric methods for IR research and vice
versa has evolved over the past forty years, from bibliometric modeling of
IR system processes to the exploitation of citation relationships to
provide extended browsing capabilities to identify potentially relevant
documents based on citation linkages. More recently, the adoption of
language-based methods from NLP and computational linguistics has
benefitted both IR and bibliometrics research. Applications of
language-based approaches are still relatively new in bibliometric
contexts. Link-based analysis (citations, co-authorship, hyperlinks) in
combination with textual analysis can capitalize on the strengths of
both approaches [