Skip to content
Back to search results

Ellis Information-Seeking Behavior Model

Introduced by
David Ellis
Year
1989
Type
Behavioural-features model of information-seeking activities
Family
Information behavior / information seeking
Font size

What does it do?

David Ellis's model, set out in his 1989 article 'A behavioural approach to information retrieval system design,' characterizes information seeking through a set of generic behavioural features rather than a fixed sequence of stages.

From grounded-theory studies of how academic researchers actually look for information, Ellis identified features such as starting, chaining, browsing, differentiating, monitoring and extracting. The crucial claim is that these are recurring activities whose detailed pattern and interrelation vary from person to person and task to task, so information seeking is better described as a flexible repertoire of behaviours than as a single ordered process. Because each feature maps onto a concrete capability, the model was explicitly framed to inform the design of information retrieval systems that support real seeking behaviour.

The intuition behind the method

Think about how a researcher actually digs into a new topic. They start somewhere — a known expert's paper, a colleague's tip. From there they follow references backward and forward, chaining their way through the literature. They browse promising journals or shelves semi-directedly, they differentiate sources by quality and relevance to filter the flood, they keep monitoring a few key outlets to stay current, and at some point they extract systematically from the most valuable sources. Ellis noticed that almost everyone does some mix of these things, but no two people do them in the same order or proportion, and the same person varies by project. So instead of forcing this into a tidy step-by-step pipeline, his model just names the recurring moves and lets their combination be flexible. The payoff is practical: if you know the moves, you can build a search system that supports each one.

History

David Ellis introduced the behavioural-features model in his 1989 Journal of Documentation article based on grounded-theory studies of social scientists' information-seeking behaviour. He deliberately framed it as a behavioural alternative to the prevailing cognitive and system-centred approaches, with the explicit aim of informing information retrieval system design. Subsequent studies by Ellis and collaborators extended and tested the features with academic physicists and chemists, English literature and history researchers, and industrial scientists and engineers, adding features such as verifying and ending, and confirming the model's adaptability across disciplines. The model became one of the canonical frameworks of information behaviour, frequently set alongside Kuhlthau's ISP and Wilson's models, and continues to inform research on scholarly search and the design of discovery systems.

When to use it

Use Ellis's model when you want a behaviourally grounded description of how people seek information in order to understand practice or to design retrieval systems and services that support real search activities. It is well suited to studying scholarly and professional information seeking, to specifying features a search interface or discovery system should support, and to comparing seeking behaviour across disciplines and roles. It is less appropriate when you need an account of the emotional or cognitive experience of searching (where stage models like the ISP fit better), when you require a strict temporal sequence, or when the population and task differ sharply from the academic researchers from whom the features were derived. Because the features come from qualitative, grounded study, applying the model well usually means observing or interviewing seekers rather than relying solely on system logs.

Limitations and considerations

  • Derived from academic researchers, so the specific feature set may need extension or adaptation for other populations and tasks.
  • It describes activities but says little about the affective or motivational dimensions of seeking.
  • Because order is left open, the model offers limited predictive power about how a search will unfold.
  • The features can overlap and shade into one another, making consistent coding across studies difficult.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Grounded empirically in how researchers actually seek information, giving the features strong descriptive validity.
  • Avoids forcing seeking into a rigid stage sequence, capturing the flexible, non-linear character of real search.
  • Each behavioural feature maps cleanly onto concrete system capabilities, making the model directly useful for IR design.
  • The feature vocabulary (starting, chaining, browsing, etc.) is portable and has been adapted across many disciplines and contexts.

Applications

  • Specifying the search activities a digital library or discovery interface should support (chaining, monitoring, browsing).
  • Comparing scholarly information-seeking behaviour across disciplines and professional roles.
  • Diagnosing gaps in a search service by checking which behavioural features it fails to support.
  • Designing citation-navigation and alerting features grounded in observed chaining and monitoring behaviour.
  • Providing a behavioural vocabulary for studies and training in scholarly and professional information literacy.

How it works

How it works

1

The model is built bottom-up from empirical study rather than imposed from theory. Ellis used a grounded-theory approach, interviewing academic researchers about how they actually sought information and inducing the behavioural features from those accounts. This methodological grounding matters: the features are not a priori categories but patterns observed in real practice, which is why the model claims descriptive fidelity to how seeking happens rather than how it ought to happen in an idealized scheme.

Formula
F = \text{generic features induced from data},\quad \text{grounded theory: data}\Rightarrow F,\ \text{not } F\Rightarrow\text{data}
2

From the data Ellis enumerated a set of generic features of information-seeking behaviour. The core set comprises Starting (the activities that begin a search, such as identifying initial sources or key people), Chaining (following citation connections among materials), Browsing (semi-directed searching in areas of potential interest), Differentiating (using known differences among sources to filter and prioritize), Monitoring (keeping up to date by following particular sources), and Extracting (systematically working through a source to identify material of interest). Naming this repertoire gives a precise vocabulary for the components of seeking.

Formula
F=\{\text{Starting},\text{Chaining},\text{Browsing},\text{Differentiating},\text{Monitoring},\text{Extracting}\}
3

Chaining illustrates how each feature has concrete structure. It means following chains of citation or referential connection between documents, and it runs in two directions: backward chaining follows the references a source cites, while forward chaining follows the later works that cite a given source. This citation-graph movement is one of the most recognizable and system-relevant features, and characterizing it precisely shows how an abstract 'feature' corresponds to a concrete, supportable search behaviour.

Formula
\text{Chaining}:\ d_0 \to \{d : d\ \text{cites}\ d_0\}\ (\text{forward}) \cup \{d : d_0\ \text{cites}\ d\}\ (\text{backward})
4

Critically, the model does not impose a fixed order on the features. The detailed interrelationships and sequence in which a seeker deploys starting, chaining, browsing, differentiating, monitoring and extracting depend on the individual and the circumstances of the search. The model therefore describes a flexible, non-linear interaction of behaviours rather than a stage sequence, distinguishing it from staged process models and capturing the messy, iterative reality of real information seeking.

Formula
\text{sequence not fixed:}\ \sigma=(f_{i_1},f_{i_2},\dots),\ f_{i_k}\in F,\ \text{order } \sigma \text{ context-dependent}
5

Finally, because the features were identified with system design in mind, the model translates directly into requirements for information retrieval systems. Each behavioural feature implies facilities a system should support — for example, chaining implies linked references and citation navigation, monitoring implies alerting, differentiating implies source-comparison cues. Mapping features to supported facilities is the model's design payoff, turning a description of behaviour into a checklist for building retrieval environments that fit how people actually search.

Formula
\text{IR design}\;\mapsto\;\bigl\{\text{support}(f) : f\in F\bigr\},\quad \text{features}\to\text{system facilities}

Common pitfalls

  • Mistaking the feature list for an ordered sequence and reading it as a fixed stage pipeline.
  • Assuming the original feature set is exhaustive for every domain without checking whether new behaviours are needed.
  • Coding behaviours from system logs alone, which can miss browsing, monitoring and differentiating done outside the system.
  • Treating the features as system functions to be imposed on users rather than as observed behaviours to be supported.

Frequently asked questions

What are the behavioural features in Ellis's model?

The core features are Starting (beginning a search by identifying initial sources or contacts), Chaining (following citation links among documents, backward to references and forward to citing works), Browsing (semi-directed searching in areas of potential interest), Differentiating (using known differences among sources to filter and prioritize them), Monitoring (keeping current by following selected sources), and Extracting (systematically working through a source to pull out relevant material). Later studies added features such as verifying and ending. Together they form a repertoire of recurring activities rather than an ordered set of stages.

How is Ellis's model different from Kuhlthau's ISP?

Kuhlthau's Information Search Process is a staged model that follows a single search over time and emphasizes the affective experience — the rise and fall of uncertainty across initiation, exploration, formulation and so on. Ellis's model instead identifies generic behavioural features whose order is not fixed; it describes what activities seekers perform rather than the emotional arc of one extended task. The two are complementary: ISP captures the temporal and affective shape of a search, while Ellis captures the behavioural repertoire deployed within it, and Ellis's design orientation makes it especially useful for building retrieval systems.

Why does the model leave the order of features unspecified?

Because Ellis's empirical studies showed that seekers combine the features in different orders and proportions depending on the person, the discipline and the specifics of the task. Imposing a fixed sequence would misrepresent this observed variability. By describing a flexible, non-linear interaction of features instead of a pipeline, the model stays faithful to how real searching unfolds — iterative, overlapping and context-dependent — while still giving a precise vocabulary for the component activities and the system facilities that support them.

Related methods

Sources

  1. Ellis, D. (1989). A behavioural approach to information retrieval system design. Journal of Documentation, 45(3), 171-212.

    Open DOI reference
    Copied.