ArticlePDF Available

Why People Don't Swarm: Evidence for a Dual-process Memory Model in Collaborative Tagging



We propose a dual-process memory model for incorporation into current generative models of collaborative tagging. This would be in line with a large body of research in cognitive psychology showing dissociation of implicit and explicit memory processes. We present initial findings from experimental studies that show the reality of the two memory processes in collaborative tagging. Our proposal has both theoretical and practical implications, mainly in terms of gaining a better understanding of the underlying processes in tagging and the differential impact that different manipulations in tagging environments will have on memory processes.
Why People Don’t Swarm: Evidence for a Dual-process
Memory Model in Collaborative Tagging
Tobias Ley
Center for Educational Technologies,
Tallinn University
Narva Mnt 25, 10120 Tallinn, Estonia
+372 6 409 355
Paul Seitlinger
Knowledge Management Institute,
Graz University of Technology
Inffeldgasse 21a, 8010 Graz, Austria
+43 316 873 9566
We propose a dual-process memory model for incorporation into
current generative models of collaborative tagging. This would be
in line with a large body of research in cognitive psychology
showing dissociation of implicit and explicit memory processes.
We present initial findings from experimental studies that show
the reality of the two memory processes in collaborative tagging.
Our proposal has both theoretical and practical implications,
mainly in terms of gaining a better understanding of the
underlying processes in tagging and the differential impact that
different manipulations in tagging environments will have on
memory processes.
Categories and Subject Descriptors
H.1.2 [Information Systems]: User/Machine Systems - Human
information processing, H.5.3 [Information Interfaces and
Presentation]: Group and Organization Interfaces - Web-based
interaction, J.4 [Computer Applications]: Social and Behavioral
Science Psychology,
General Terms
Experimentation, Human Factors
Collaborative Tagging, Implicit and Explicit Memory, Process-
dissociation Procedure
It has become customary to regard human behavior in the use
of social software systems as a form of swarming behavior. It
seems to be appealing to regard human action as simple imitating
behavior that only in combination with the behavior of others
leads to some form of intelligence. We think this disregards a long
history of research into human information processing.
Instead, we suggest that human behavior in the use of social
software is driven by a sophisticated internal knowledge
representation highly adaptive to the current task, and flexibly
used by means of metacognitive processes. In this paper, we take
collaborative tagging as a case in point and demonstrate how the
internal processes can be examined with a process dissociation
procedure. We show initial results that provide evidence for a
dual process account of encoding and retrieval from memory.
When people use a collaborative tagging environment,
individual memory processes are likely to play a major role. In the
simplest case, one could differentiate these processes into
encoding and retrieval. When a user browses a collection or
searches, she observes the tags assigned to resources or displayed
as tag clouds and, hence, encodes the terms in individual memory.
When, on the other hand, she tags a resource herself, retrieval
from memory needs to take place that allows her to assign
appropriate tags.
In recent attempts to understand tagging behaviors, several
generative models of tagging have been suggested that allow for
simulating a user’s behavior during tagging. The purpose of these
models is to gain a better understanding of the emergent processes
that are assumed to play a major role in these types of
environments. These models implicitly or explicitly make
assumptions on memory processes during encoding and retrieval.
Our reading of the literature suggests that two alternative
conceptions have been proposed to account for these processes.
Either it is assumed that a simple form of imitation takes place
where tags previously used by others are used as tags for the
current resource. Terms are then drawn in a random process from
existing terms, and this process is influenced among others by
the frequency and recency of use [4], [8]. Alternatively, it has
been suggested that cues during encoding or retrieval (e.g. the
tagged resources or other tags displayed) lead to deep conceptual
processing in which a conceptual semantic memory system is
called to generate appropriate terms to be used as tags [5],[6].
In cognitive psychology research, it has long been suggested
that two types of memory processes play a role in all types of
retrieval tasks from memory (e.g. [1][12][14]). We will refer to
these here as implicit and explicit memory processes. While
explicit memory preserves the context from the study episode,
implicit memory does not support the conscious retrieval of the
study context. Retrieving from explicit memory involves an
experience of conscious and deliberate recollection, while
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. To copy
otherwise, or republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee.
WebSci ’11, June 14-17, 2011, Koblenz, Germany.
Copyright held by the authors.
retrieving from implicit memory is relatively automatic and
effortless and is associated with an experience of familiarity or
“just knowing” [10]. While explicit memory is mainly responsible
for the conscious recollection in typical memory tests (e.g.
recognition or recall tests of studied material), implicit memory
facilitates performance on certain tasks (e.g. identification of
words or indicating preference for a stimulus) without conscious
intention of recalling the study episode.
Support for the empirical reality of this distinction has been
obtained through numerous experimental studies in which
dissociation of performance attributed to these two memory
processes could be observed. This means that both processes
contribute independently to a certain memory performance, and
that the systems can be influenced independently by different
manipulations, both during encoding or retrieval (e.g. [15][16]).
For example, semantic elaboration during encoding (e.g.
conceptually elaborating on an item when studying it) has a large
positive effect on explicit memory (e.g. consciously recalling the
item), which is consistently larger than on implicit memory (e.g
completing a word stem [7]). On the other hand, several
manipulations of processing fluency during retrieval (e.g. how
easy it is to read an item) have effects on implicit memory and no
effect on explicit memory (e.g. [3]).
Because implicit and explicit processing can not simply be
observed from behavioral data or from tag data in a tagging
environment, it is necessary to carefully design experimental
studies in which the two types of processes can be studied, and
devise sophisticated models that allow estimation of parameters
for these internal processes.
In our own research, we look at how users process tags they
observe in a collaborative tagging environment. For example, we
have tried to discern explicit and implicit memory processes with
an adapted process dissociation procedure initially suggested by
Jacoby (1991) [9]. In this procedure, users first process tags in an
incidental processing conditions: they have to make decisions on
which content is more suitable by looking at tags assigned to that
content. In a second phase, they are then asked to intentionally
learn a number of tags from the same environment. When they are
tested for their memory of the tags in a later memory test, one
group is asked to name all tags (incidentally learned or
intentionally learned) while another group is asked to produce
only those tags that were intentionally learned.
A multinomial model [2] can then be used to obtain
independent estimates of probabilities for two memory processes
during retrieval, an explicit recollection (explicit) and a
familiarity-based judgment process (implicit).
Our initial findings firstly suggest that a dual process model
that involves independent retrieval processes from explicit and
implicit memory is suitable to explain performance in the memory
test after the search task. This is shown by significance of the
overall model and means that both processes contribute to
memory performance. Secondly, across several search task
conditions, we find that processes of explicit recollection are more
important for performance than implicit processes. This is
indicated by a significant difference in the two parameters.
Finally, we have found evidence that the two memory processes
can be selectively manipulated. To show this, we have instructed
users to search with different intentions (e.g. looking for material
for a private leisure activity vs. looking for material for a
homework assignment). Initial results indicate that estimates for
the memory processes do in fact vary with search intention. This
is in line with research showing that metacognitive strategies can
influence implicit and explicit memory processes [13].
In concluding from the above, a dual process model as we
are suggesting should help us gain a better theoretical
understanding of collaborative tagging processes and the
underlying cognitive mechanisms. We see mainly two areas where
such enhancements would be beneficial. First, current generative
models that look at microprocesses in collaborative tagging would
benefit from incorporating a two-stage memory process as we are
suggesting. A formal approach as suggested above would lend
itself for inclusion into generative models of tagging. Familiarity-
based retrieval is usually conceptualized as decision process
involving a continuous variable with a varying response criterion
(strict or lenient), much the same as in a signal detection model of
memory. Explicit recollection, on the other hand, is characterized
as a discrete memory process that involves distinct states and
transitions of learning or forgetting [16] which could be captured
through a markov model.
Secondly, we assume that our dual process memory model
might shed new light on a recent attempt to explain tagging
motivations. Körner et al. [11] have proposed a typology of user
motivations, classifying users as describers or categorizers. This
assertion is mainly based on the observation of empirical tagging
behavior, whereas the theoretical underpinning of this model is
limited. We suspect that the two tagging patterns observed by
Körner et al. could be mapped to the two memory processes we
are suggesting, where the “describer pattern” would be driven
mainly by implicit memory processes, and the “categorizer
pattern” mainly by explicit ones.
Furthermore, our research has practical significance as well.
A dual process memory model for collaborative tagging would
suggest that different manipulations made in the tagging
environment will impact the two memory processes in different
ways, and hence have a differential impact on the generation of
tags in these environments. As a result, concrete hypotheses could
be proposed. As to manipulations during encoding, one would
expect that generating tags vs. simply reading through tags should
impact explicit memory processes more than implicit ones. The
same is to be expected for other manipulations that target deep
semantic processing during encoding (e.g. displaying tags as a
network of semantically related tags). During retrieval, on the
other hand, we would expect that tag recommendations given at
retrieval time would influence implicit memory processes more
than explicit memory processes. Also readability of these
recommendations (e.g. varying font size as is common in tag
clouds) should influence processing fluency and therefore impact
implicit memory processes more than explicit ones. Moreover,
different forgetting curves have been obtained for explicit as
compared to implicit memory representations [16].
We have only listed some of the more obvious hypotheses
above. We would argue that an integrated dual process model of
memory could make predictions about a large number of memory
phenomena during tagging, and given the right experimental
paradigms these could be systematically studied.
[1] Brainerd, C.J., Wright, R., & Reyna, V.F. 2002. Dual-
Retrieval Processes in Free and Associative Recall. Journal
of Memory and Language, 46, 120-152.
[2] Buchner, A., Erdfelder, E., and Vaterrodt-Plünnecke, B.
1995. Toward Unbiased Measurement of Conscious and
Unconscious Memory Processes Within the Process
Dissociation Framework. Journal of Experimental
Psychology: General, 124, 137-160.
[3] Cameron, T.E., & Hockley, W.E. 2000. The revelation effect
for item and associative recognition: Familiarity versus
recollection. Memory & Cognition, 28, 176-183.
[4] Cattuto, C., Loreto, V., & Pietronero 2007. Semiotic
dynamics and collaborative tagging. PNAS, 104, 1461-1464.
[5] Fu, W. 2008. The microstructures of social tagging: a
rational model. Proceedings from CSCW’08, pp. 229-238.
New York: ACM Press.
[6] Fu, W., Kannampallil, T.G., Kang, R. 2009. A semantic
imitation model of social tag choices. Proceedings of the
International Conference on Computational Science and
Engineering CSE '09, pp. 66-73, IEEE Computer Society.
[7] Gardiner, J.M. 1988. Functional aspects of recollective
experince. Memory & Cognition, 16, 309-313.
[8] Halpin, H., Robu, V., & Shepherd, H. 2007. The complex
dynamics of collaborative tagging. Proceedings from
WWW’07, pp. 211-220. New York: ACM Press.
[9] Jacoby, L.L. 1991. A process dissociation framework:
Seperating automatic from intentional uses of memory.
Journal of Memory & Language, 30, 513-541.
[10] Kelley, C.M., & Jacoby, L.L. 2000. Recollection and
Familiarity: Process-Dissociation. In E. Tulving & F. Craik
Eds. The Oxford handbook of memory, pp. 215-228. Oxford:
Oxford University Press.
[11] Körner, C.; Benz, D.; Strohmaier, M.; Hotho, A. & Stumme,
G. 2010. Stop Thinking, start Tagging - Tag Semantics
emerge from Collaborative Verbosity, in Proceedings of the
19th International World Wide Web Conference WWW 2010,
ACM Press, Raleigh, NC, USA .
[12] Mandler, G. 1980. Recognizing: The judgement of previous
occurrence. Psychological Review, 87, 252-271.
[13] Narens, L., Graf, A. and Nelson, T. O. 1996. Metacognitive
Aspects of Implicit/Explicit Memory, In Implicit memory
and metacognition, Lynne Reder, Ed., Routledge, pp. 137-
[14] Nelson, D.L., Fisher, S.L., & Akirmak, U. 2007. How
implicitly activated and explicitly acquired knowledge
contribute to the effectiveness of retrieval cues. Memory &
Cognition, 35, 1892-1904.
[15] Nelson, D.L., McKinney, V.M., Gee, N.R., & Janczura, G.A.
1998. Interpreting the Influence of Implicitly Activated
Memories on Recall and Recognition. Psychological Review,
105, 299-324.
[16] Yonelinas, A.P. 2002. The Nature of Recollection and
Familiarity: A Review of 30 Years of Research. Journal of
Memory and Language, 46, 441-517.
... In line with Fu et al. (2010) we therefore conclude that ST is a phenomenon to be examined both on a word and semantic level. Particularly, social tags are not only cues supporting information search or leading to simple swarm-like imitation (Ley & Seitlinger, 2011), but seem to externalize individual thoughts that in turn evoke associated ideas and reflections of other students. Considered in this light, it becomes clearer why the simple tagging functionality of collaborative learning environments proves to support learning processes among students (e.g. ...
Social tagging is a widespread phenomenon on the Web allowing users to tag resources, such as photos, by freely cho-sen labels. Imitation of other users' tagging behavior is deemed to increase the inter-individual consistency of tag assignments. Both verbatim and semantic mechanisms have been proposed where the first case suggests reuse of the exact words and the second reuse of the concept without necessarily the word form. Here, we present a multinomial model of assigning tags that integrates these perspectives. On the basis of the fuzzy-trace theory, the model includes separate parameters for the retrieval of verbatim and gist traces of encountered tags. Results of two experiments demonstrate that both types of memory traces contribute substan-tially and independently to tag productions. We conclude that the imitation of tags may be explained within one fuzzy-trace frame-work that contributes to our understanding of emergent phenomena in social information systems.
Tagging systems represent the conceptual knowledge of a community. We experimentally tested whether people harness this collective knowledge when navigating through the Web. As a within-factor we manipulated people's prior knowledge (no knowledge vs. prior knowledge that was congruent/incongruent to the collective knowledge inherent in the tags). As between-factor we manipulated whether people had tag clouds available or not. In line with the Information Foraging Theory and with the Co-Evolution Model of individual learning and collective knowledge building, we found that people's prior knowledge and tag clouds influenced their navigation. Tags which were congruent with people's prior knowledge had a higher probability of being selected. A knowledge test showed that participants adapted their individual conceptual knowledge to the collective knowledge. This incidental learning shows that people harness collective knowledge just by navigation with tag clouds.
Conference Paper
Full-text available
The debate within the Web community over the optimal means by which to organize information often pits formalized classifications against distributed collaborative tagging systems. A number of questions remain unanswered, however, regarding the nature of collaborative tagging systems including whether coherent categorization schemes can emerge from unsupervised tagging by users. This paper uses data from the social bookmarking site delicio. us to examine the dynamics of collaborative tagging systems. In particular, we examine whether the distribution of the frequency of use of tags for "popular" sites with a long history (many tags and many users) can be described by a power law distribution, often characteristic of what are considered complex systems. We produce a generative model of collaborative tagging in order to understand the basic dynamics behind tagging, including how a power law distribution of tags could arise. We empirically examine the tagging history of sites in order to determine how this distribution arises over time and to determine the patterns prior to a stable distribution. Lastly, by focusing on the high-frequency tags of a site where the distribution of tags is a stabilized power law, we show how tag co-occurrence networks for a sample domain of tags can be used to analyze the meaning of particular tags given their relationship to other tags.
Conference Paper
Full-text available
We describe a semantic imitation model of social tagging that integrates formal representations of semantics and a stochastic tag choice process to explain and predict emergent behavioral patterns. The model adopts a probabilistic topic model to separately represent external word-topic and internal word-concept relations. These representations are coupled with a tag-based topic inference process that predicts how existing tags may influence the semantic interpretation of a document. The inferred topics influence the choice of tags assigned to a document through a random utility model of tag choices. We show that the model is successful in explaining the stability in tag proportions across time and power-law frequency-rank distributions of tag co-occurrences for semantically general and narrow tags. The model also generates novel predictions on how emergent behavioral patterns may change when users with different domain expertise interact with a social tagging system. The model demonstrates the weaknesses of single-level analyses and highlights the importance of adopting a multi-level modeling approach to explain online social behavior.
Several suggestions for a class of theories of recognition memory have been proposed during the past decade. These models address predictions about judgments of prior occurrence of an event, not the identification of what it is. The history and current status of one of these models is discussed. The model postulates the detection of familiarity and the utilization of retrieval mechanisms as additive and separate processes. The phenomenal experience of familiarity is assigned to intraevent organizational integrative processes; retrieval depends on interevent elaborative processes. Other current theoretical options are described, and relevant supportive data from the literature are reviewed. New tests of the model involving both free recall and word pair paradigms are presented. The dual process model is extended to the word frequency effect and to the recognition difficulties of amnesic patients. (68 ref) (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Recent dual-retrieval accounts of free recall postulate that a memory target can be recalled either by directly accessing its verbatim trace or by reconstructing it from semantic or other relational information. We introduce a simple paradigm, derived from the classic Estes RTT procedure, that separates direct access from reconstruction and that separates reconstruction from a metacognitive judgment process that authorizes reconstructed targets for output. Results are reported from four experiments, two that applied the paradigm to free recall and two that extended it to associative recall. The principal findings were that (a) direct access was enhanced by manipulations that made targets' surface forms easier to process or that focused recall on individual targets, (b) reconstruction was enhanced by manipulations that made targets' meaning content easier to process or that focused recall on groups of targets, and (c) such manipulations produced single dissociations, double dissociations, and reversed associations between direct access, reconstruction, and metacognitive judgment. We discuss how this paradigm might be exploited to unify dual-retrieval conceptions of recall and recognition.
To account for dissociations observed in recognition memory tests, several dual-process models have been proposed that assume that recognition judgments can be based on the recollection of details about previous events or on the assessment of stimulus familiarity. In the current article, these models are examined, along with the methods that have been developed to measure recollection and familiarity. The relevant empirical literature from behavioral, neuropsychological, and neuroimaging studies is then reviewed in order to assess model predictions. Results from a variety of measurement methods, including task-dissociation and process-estimation methods, are found to lead to remarkably consistent conclusions about the nature of recollection and familiarity, particularly when ceiling effects are avoided. For example, recollection is found to be more sensitive than familiarity to response speeding, division of attention, generation, semantic encoding, the effects of aging, and the amnestic effects of benzodiazepines, but it is less sensitive than familiarity to shifts in response criterion, fluency manipulations, forgetting over short retention intervals, and some perceptual manipulations. Moreover, neuropsychological and neuroimaging results indicate that the two processes rely on partially distinct neural substrates and provide support for models that assume that recollection relies on the hippocampus and prefrontal cortex, whereas familiarity relies on regions surrounding the hippocampus. Double dissociations produced by experimental manipulations at time of test indicate that the two processes are independent at retrieval, and single dissociations produced by study manipulations indicate that they are partially independent during encoding. Recollection is similar but not identical to free recall, whereas familiarity is similar to conceptual implicit memory, but is dissociable from perceptual implicit memory. Finally, the results indicate that recollection reflects a thresholdlike retrieval process that supports novel learning, whereas familiarity reflects a signal-detection process that can support novel learning only under certain conditions. The results verify a number of model predictions and prove useful in resolving several theoretical disagreements.
This paper begins by considering problems that have plagued investigations of automatic or unconscious influences of perception and memory. A process dissociation procedure that provides an escape from those problems is introduced. The process dissociation procedure separates the contributions of different types of processes to performance of a task, rather than equating processes with tasks. Using that procedure, I provide new evidence in favor of a two-factor theory of recognition memory; one factor relies on automatic processes and the other relies on intentional processes. Recollection (an intentional use of memory) is hampered when attention is divided, rather than full, at the time of test. In contrast, the use of familiarity as a basis for recognition memory judgments (an automatic use of memory) is shown to be invariant across full versus divided attention, manipulated at test. Process dissociation procedures provide a general framework for separating automatic from intentional forms of processing in a variety of domains; including perception, memory, and thought.
Conference Paper
Recent research provides evidence for the presence of emergent semantics in collaborative tagging systems. While several methods have been proposed, little is known about the factors that influence the evolution of semantic structures in these systems. A natural hypothesis is that the quality of the emergent semantics depends on the pragmatics of tagging: Users with certain usage patterns might contribute more to the resulting semantics than others. In this work, we propose several measures which enable a pragmatic differentiation of taggers by their degree of contribution to emerging semantic structures. We distinguish between categorizers, who typically use a small set of tags as a replacement for hierarchical classification schemes, and describers, who are annotating resources with a wealth of freely associated, descriptive keywords. To study our hypothesis, we apply semantic similarity measures to 64 different partitions of a real-world and large-scale folksonomy containing different ratios of categorizers and describers. Our results not only show that "verbose" taggers are most useful for the emergence of tag semantics, but also that a subset containing only 40% of the most 'verbose' taggers can produce results that match and even outperform the semantic precision obtained from the whole dataset. Moreover, the results suggest that there exists a causal link between the pragmatics of tagging and resulting emergent semantics. This work is relevant for designers and analysts of tagging systems interested (i) in fostering the semantic development of their platforms, (ii) in identifying users introducing "semantic noise", and (iii) in learning ontologies.
Conference Paper
This article presents a rational model developed under the distributed cognition framework that explains how social tags influence knowledge acquisition and adaptation in exploratory ill-defined information tasks. The model provides integrated predictions on the interactions among link selections, use and creation of tags, and the formation of mental categories. The model shows that the quality of tags not only influences search efficiency, but also the quality of mental categories formed during exploratory search. In addition, the model shows that aggregate regularities can be explained by microstructures of behavior that emerged from the adaptive assimilation of concepts and categories of multiple users through the social tagging system. The model has important implications on how collaborative systems could influence higher-level cognitive activities.