Challenges and Opportunities within Personal Life Archives

Abstract

Nowadays, almost everyone holds some form or other of a personal life archive. Automatically maintaining such an archive is an activity that is becoming increasingly common, however without automatic support the users will quickly be overwhelmed by the volume of data and will miss out on the potential benefits that lifelogs provide. In this paper we give an overview of the current status of lifelog research and propose a concept for exploring these archives. We motivate the need for new methodologies for indexing data, organizing content and supporting information access. Finally we will describe challenges to be addressed and give an overview of initial steps that have to be taken, to address the challenges of organising and searching personal life archives.

Full text

Challenges and Opportunities within Personal Life Archives
Duc-Tien Dang-Nguyen
Dublin City University
Dublin, Ireland
duc-tien.dang-nguyen@dcu.ie
Michael Riegler
Center for Digitalisation and Engineering
Oslo, Norway
michael@simula.no
Liting Zhou
Dublin City University
Dublin, Ireland
zhou.liting2@dcu.ie
Cathal Gurrin
Dublin City University
Dublin, Ireland
cgurrin@computing.dcu.ie
ABSTRACT
Nowadays, almost everyone holds some form or other of a per-
sonal life archive. Automatically maintaining such an archive is an
activity that is becoming increasingly common, because without
automatic support the users will quickly be overwhelmed by the
volume of data and will miss out on the potential benets that
lifelogs provide. In this paper we give an overview of the current
status of lifelog research and propose a concept for exploring these
archives. We motivate the need for new methodologies for indexing
data, organizing content and supporting information access. Finally
we will describe challenges to be addressed and give an overview
of initial steps that have to be taken, to address the challenges of
organising and searching personal life archives.
CCS CONCEPTS
•Information systems →Digital libraries and archives
;
Per-
sonalization
;Users and interactive retrieval;Evaluation of retrieval
results;
KEYWORDS
Search Engine, Lifelogging, Personal Life Archive
ACM Reference Format:
Duc-Tien Dang-Nguyen, Michael Riegler, Liting Zhou, and Cathal Gurrin.
2018. Challenges and Opportunities within Personal Life Archives. In ICMR
’18: 2018 International Conference on Multimedia Retrieval, June 11–14, 2018,
Yokohama, Japan. ACM, New York, NY, USA, 9 pages. https://doi.org/10.
1145/3206025.3206040
1 INTRODUCTION
New devices such as mobiles, tablets, wearables, in addition to
the various social media platforms such as Facebook, Instagram
or Snapchat, have became a normal part of everyday life. Conse-
quently, users end up passively collecting large volumes of data
about themselves (typically non-indexed) in their own personal life
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 prot or commercial advantage and that copies bear this notice and the full citation
on the rst page. Copyrights for components of this work owned by others than ACM
must be honored. Abstracting with credit is permitted. To copy otherwise, or republish,
to post on servers or to redistribute to lists, requires prior specic permission and/or a
fee. Request permissions from permissions@acm.org.
ICMR ’18, June 11–14, 2018, Yokohama, Japan
©2018 Association for Computing Machinery.
ACM ISBN 978-1-4503-5046-4/18/06.. .$15.00
https://doi.org/10.1145/3206025.3206040
archives, which are currently spread across multiple devices and
services. These personal life archives can contain information about
every activity an individual participates in, such as where they go,
how they get there, who they speak with, what they see and what
information they access. In eect, a multi-source, detailed, digital
life diary is being generated for every individual who chooses to
do use such technologies. The key enabling technology is the ready
availability of power-ecient sensing devices, which are embedded
in the modern cell-phones and wearables. It is our conjecture that it
is only a short step to move from sensing life to searching through
the resultant personal life archives.
Furthermore, we believe that there is a clear coverage-gap be-
tween the indexed online content and the non-indexed personal
life archive content. While conventional search engines provide
retrieval facilities over online data, the integration of personal data
in the search process has not yet occurred, apart from some lim-
ited personalisation of search results (which is more limiting than
helpful). It is our conjecture that a new generation of search en-
gine should be able to “zoom" in and out to give users information
based on their friends and families (i.e., via social network platforms,
from people having the same habits, etc.), and continue to go deeper
and must give the insights from the users themselves (from the
quantied-self information). Given the right capture technologies,
coupled with a new generation of data organisation techniques
and semantic multimedia annotation, new ways of interacting with
our own personal life archives will emerge and change the way we
view ourselves and our activities.
What is needed as a next step of development and to unlock the
potential of these data, are new ways of organizing, annotating,
indexing and interacting with personal life archives. Being able to
interact with the collected data with the same ease as one executes
a Google search nowadays, will enable a new level of insight for
the users that helps them to get a complete picture of their data
and information. This can happen on dierent granularities such
as a phone number, a location, an entire event in great multimedia
detail, or even perform an analysis of lifestyle trends over many
years.
Providing insights from personal life archives can help to im-
prove healthcare, working lives, education and social activities [
5
]
and can give new levels of self-awareness. Some obvious benets
for the users can be for example, an easy way of sharing natural
life experiences, prospective memory feedback to enhance produc-
tivity, personalised wellness feedback, better understanding the

ICMR ’18, June 11–14, 2018, Yokohama, Japan D.-T. Dang-Nguyen et al.
functioning of the human memory system, and a better understand-
ing of the associations between lifestyle, environmental context
and mortality.
Nevertheless, there are many challenges to be overcome like opti-
mised, semantically rich data capture, ecient data mining, knowl-
edge extraction and retrieval tools and the provision of appropriate
interaction methodologies. In this work we focus on identifying
the opportunity and the challenges and then suggesting the mix of
technologies that have the most potential to tackle these challenges
in the future.
Consequently, the aims of this paper are:
•
To provide an overview of past eorts to capture personal life
archives in the research community;
•
To inspire and motivate researchers in the multimedia commu-
nity to use their know-how in this new emerging and societal
important area;
•
To propose how future personal life archive data should be stored
and organised, and therefore made easy for a user to access via
an appropriate search mechanism;
•
To discuss the challenges in providing retrieval and access to
personal life archives;
In the rest of this paper we will briey describe the overall
challenges before discussing the particular requirements for data
capture, organisation, retrieval and presentation. Finally we will
motivate the benets of personal life archives and introduce our
implementations to address these challenges .
2 CURRENT STATE AND PERSONAL LIFE
ARCHIVES
In the 1940s Vannevar Bush introduced the world to the Memex,
a life knowledge organisation hypermedia system operating as a
desk-based device [
7
]. Memex introduced new concepts such as
information links or trails which are created by the individual or by
others. Memex was described as an “enlarged intimate supplement to
ones memory”. In these words, Bush had identied some of the key
issues for maintaining personal life archives, that they be enlarged
(store as much information as feasible), intimate (private to the
owner) and supplemental (working in synergy with ones memory).
The Memex provided an inspiration for the Xanadu system [
35
]
by Ted Nelson which introduced the terms hypertext and hyper-
media and was a precursor to the WWW, though with dierent
aims and goals. In 2006, Bell and Gemmel’s work on developing
the MyLifeBits [
21
] personal life experience archiving tool digital
memories [
4
,
5
], and their book Total Recall [
5
] introduced the
public to the concept of maintaining personal life archives of all
information encountered and developed the rst database focused
retrieval technologies.
Lifelogging has appeared to be an extreme activity carried out
only by a small number of pioneering enthusiasts e.g., Steve Mann [
33
],
Gordon Bell [
4
], and Cathal Gurrin [
15
,
27
]. If one explores the
reasons why only few pioneers had gathered these personal life
archives automatically are many and varied and pose a set of chal-
lenges for the community to consider. One of the most important
challenges is the privacy and ethical concerns [
14
,
36
]. It is our
conjecture that personal life archives are likely to follow the expe-
riences of cell phone cameras, social networking sites, and location
tracking services. Once the personalised experience, wellness, and
memory capture and sharing provide a wide range of benets to
end users; these concerns become of secondary importance. In
other words, society develops an acceptable usage policy and em-
braces a new technology as the benets become apparent. Another
challenge comes with the overwhelming amounts of data [
42
] that
personal life archives naturally generate. Research has indicated
that lifelog data can be summarised into useful knowledge through
i) segmenting it into a series of distinct events or activities [
16
], ii)
automatically labelling those events from both the content [
15
] and
context [
30
], iii) automatically detecting faces and event novelty
to identify those events that are more interesting to reect on [
17
],
iv) presenting segments of this personal life archive to the user as
required and v) analysing this data to provide new knowledge to
the user.
Another challenge is that personal life archives attract life ac-
tivity data at varying degrees of delity. A single sensor (e.g., an
odometer or pedometer) operates at a very low-delity whereas
the wearable video technique moves towards high-delity total
capture which digitises a much richer snapshot of life activities.
All such data has sometimes been referred to as memories, yet this
should not be considered as an actual memory capture technique,
rather a multimedia sampling of a representation of a life activity,
which, when located and presented by appropriate software to a
user would help to trigger wet-ware memory recall.
Consequently, and in agreement with Bush’s vision of the ‘en-
larged intimate supplement to ones memory’, we refer to these
lifelogs or e-memories as Personal Life Archives of sampled life
experience data that can be analysed in real-time and work with
the person’s memory to enhance life experience. Personal, in the
sense that they are intensely private archives that are gathered by
the individual for the individual, though of course they may be
stored and processed remotely and aspects of the archive may be
shared with other people or organisations. Life in the sense that
they are enlarged archives of life activity data and as such should
aim for the target of total capture of life experience. Archives in that
they will be historically stored and analysed throughout the life of
the individual and potentially even longer. While real-time access
will provide for life enriching contextual access to past activities
(in synergy with the individual’s natural memory), there is also
enormous potential for long-term analysis and understanding of
the life archive data.
3 CHALLENGES
Search and ranking is only one aspect of the challenge that infor-
mation retrieval systems have to tackle. When we are developing
retrieval systems, whether for personal life archives or simply for
WWW pages, to nd a starting point, we need to understand how
the personal life archive will be accessed. To begin our consider-
ation, contemplate how people access their own digital photo or
video archives. For suciently small archives, a browsing mecha-
nism (manually or automatic organisation into clusters based on
folders or events) is acceptable, where the selection of an axis of
browsing results in the generation of a manageable set of result
documents. Consider browsing a photo archive using date or loca-
tion. However, when the archives become larger and less organised,

Challenges and Opportunities within Personal Life Archives ICMR ’18, June 11–14, 2018, Yokohama, Japan
a search or search/browse metaphor is normally chosen to support
fast and eective access.
Contemplating personal life archives, the sheer scale of these
multi-year or multi-decade archives suggests that a browsing method-
ology is not sucient from the outset. The initial (and the only)
experiments into multi-year multimodal personal life archive search
suggests that even a basic search methodology increases the possi-
bility of a user locating desired content by a factor of three in a third
of the time [
17
]. When considering search, there are a number of
alternative search methodologies that could be considered. Firstly
keyword based search can be processed over textual narratives
generated from the sampled life experience. Another alternative
approach is to support the user in generating a new type of multi-
axes query in an ecient manner; for example, I know that my
friends Paul and Jack were there, it was a Sunday evening, and
we were in Barcelona watching a football game in a bar. A third
option is the real-time context-driven automatic querying that is
somewhat of a holy grail of this research area. The realtime sam-
pling of life experience can trigger contextual queries to support
recollection, retrieval of information and remembering intentions,
which, if presented to the user in a suitable manner, can provide
for truly novel and currently unknown applications for personal
life archives. Applications that can remind you that the person you
have just met is having a birthday today or that the last time you
bought this type of soup, you felt ill the following day.
While there has been an explosion in the amount of consumer-
sensed and -generated data now being created, stored and shared,
the ability to organize and provide useful retrieval facilities over
this data is still limited. There are many domain-specic solutions
for uses such as sensing the level of exercise or sharing a user’s
location, however, there are still few attempts to fully grasp the
full potential of sensing the person, the quantied self [
34
]. As
we progress towards more enhanced sensing of the person (total
capture), this coming world of personal life archives will pose new
challenges for the areas of multimedia contextual sensor capture,
multimedia data organisation, multimedia search and retrieval as
well as the human factors that dene how we can interact with
these personal life archives, not to mention the outstanding issues
such as privacy, security of data and supporting the important
human need to forget.
We can point out that the key challenges are to gather rich
archives in real-time in a non-intrusive manner, to organise these
archives into meaningful experiences and generate descriptive meta-
data, to provide retrieval and recommendation facilities and to
support omnipresence of access.
In order to more clearly dene the interaction challenges and
help to articulate our vision, let us again refer to the ve R’s of
memory access from Sellen & Whittaker [
42
]. The ve R’s are
recollecting, reminiscing, retrieving, reecting and remembering
intentions. Each of the ve Rs dene a dierent reason why people
access their memories, and by inference, why people would wish to
access their personal life archives. Since each R denes a dierent
way that people are able to interact with their memories, and until
such time as we have sucient numbers of people maintaining
personal life archives to get real-world usage data, they serve as a
the only source of dierent proposed interaction scenarios available.
•
Recollecting is concerned with reliving past experiences for vari-
ous reasons. For example, we may want to recall who was at an
event, or where we parked the car.
•
Reminiscing, which is a form of recollecting, is about reliving
past experiences for emotional or sentimental reasons. It is often
concerned with story-telling or sharing of life experiences with
others.
•
Retrieving (information), is a more specic form of recollecting
in which we seek to retrieve specic information from the per-
sonal life archive, such as an address, a document or a piece of
information.
•
Reecting, is a form of quantied self-analysis over the life archive
data to discover knowledge and insights that may not be imme-
diately obvious.
•
Remembering Intentions, which is more about prospective (re-
membering plans) memory than episodic memory (past experi-
ences). This is a form of planning future activities which is a life
activity that everyone engages in.
These provide valuable clues how to develop the organisation,
search and presentation elements of Personal Life archives and will
be the focus of the remainder of this paper.
3.1 Life Archive Capture & Storage
The starting point in generating personal life archives is gathering
the data in a non-intrusive manner. Prior work [
42
] has suggested
that focused capture of only the information that the individual
needs at a point in time is the best approach for supporting hu-
man memory; however we suggest that such an approach is not in
keeping with the Memex vision and would signicantly limit the
potential of Personal Life Archives. The idea of ‘total capture’ [
21
],
sampling life experience in high-delity, is that all life experience,
whether considered important or useful at the time would be cap-
tured. This will provide for a more useful, future-proof and exible
personal life archive. As an analogy, WWW search engines index
all the WWW; a search engine that selectively indexes only the
important or popular content would fail to catch trends, hot-topics
or the long-tail of user queries. Total capture of every aspect and
moment of life experience, whether it is every heart beat, our loca-
tions and motion, or everything we see and do, provides us with
new and potentially valuable information about ourselves.
Inspired by the work of Gurrin et al. in [
22
], we list dierent
categories of life archiving tools which can be applied at the time
of writing:
•
Passive visual capture. Utilising wearable devices such as Nar-
rative Clip
1
, the Microsoft SenseCam [
28
], or rst-generation
Augmented Reality glasses, will allow for the continuous and
automatic capture of life activities as a visual sequence of digital
images.
•
Passive audio capture. Audio capture could allow for the identi-
cation of events or identication people who was speaking. It is
normally can be done via any smartphone.
•
Personal biometrics. Sensing devices are becoming more common
and widely used by the quantied self community, which allow
wearer monitor their sleep duration, distance traveled, caloric
output, and other biometrics information.
1http://getnarrative.com

ICMR ’18, June 11–14, 2018, Yokohama, Japan D.-T. Dang-Nguyen et al.
•
Mobile device context. This refers to using the phone to con-
tinuously and passively capture the users context (e.g., location,
movement, or acceleration), coupled with smart watches, these
devices are able to capture much of the activities of life.
•
Communication activities. This refers to the phone or PC pas-
sively logs messages, emails, phone calls, or other contents of
communications.
•
Data creation/access activities. Logging data consumed and cre-
ated, for example, the words typed, web pages visited, videos
watched and so on.
•
Environmental context and media. According to [
22
], lifelogging
is mostly, but not exclusively about recording using wearable
technology, they could be logged (and accessed) by other sensors,
such as surveillance cameras.
•
Manual logging life activities. This refers to the indirect or direct
logging of activity that is initiated by the user, for example, video
recording, personal logs and diary.
Once sensed data has been captured, it needs to be stored, which
is one of the key challenges that needs to be addressed. In order
to support any of the ve R’s (especially reection) the personal
life archive should not be time-limited, i.e., should extend back
indenitely and life experiences (unless expressly requested by
the user) should not be deleted for reasons of storage capacity or
processing overhead. However some form of summarisation may
be necessary.
A typical lifelogging camera wearer will generate about 1TB of
data per year. For an individual, this is reasonable to assume storage
on a single computer and the assumption of Kryder’s law, that hard
drive densities will continue to increase, should see the storage
capacity keep pace with data storage requirements. However, stor-
age capacity that is sucient for periodic photo capture will not
necessarily keep pace with continuous video capture, which will
require many tens of TBs per year. Notwithstanding the relatively
low cost of digital storage, once data services scale to thousands or
millions of people, then local storage solutions would tend to be
replaced by cloud-hosting. However in the case of the data storage
requirements of personal life archives, (at current pricing models)
the data storage cost for cloud-hosting is likely to be prohibitive. An
alternative solution needs to be found that merges the convenience
and data security of a cloud-hosting service with the low cost of
individual storage solutions.
Over the coming sections we will describe our current technique
to automatically organise personal life archives.
3.2 Life Archive Organisation
The human memory system has evolved over thousands of years
to store autobiographical memories, and we believe that personal
digital archives needs to mimic how the human mind operates. Past
literature has motivated that the human mind stores information in
distinct events or episodes, that similar episodes are associated with
each other, and that more important episodes are more strongly
remembered [
17
]. However in addition to this, given the range of
data streams associated with a given important episode of interest,
a short descriptive summary narrative of that episode is required.
As a starting point, the personal digital archive information could
be arranged as follows:
•
Raw data should be hierarchically arranged and stored: Typically in
information retrieval (IR), there is a single basic unit for indexing
and retrieval, typically called the document. For many IR tasks,
this basic unit is the preferred as unit of retrieval and choosing
the basic unit is usually trivial. With lifelog data, it is not trivial
to decide what the basic unit is since the lifelogs are multimodal
with dierent modes captured at dierent frequencies (1 second
to potentially 1 day, or longer) [
14
]. In order to deal with this
problem, we followed the study in [
14
] that sort the data by
chronological order, and use the minute as the basic units.
Building up from the basic unit, we organize the data at higher
level with can be turned into more useful information. Typically,
in a full day, we know that a person encounters anything upwards
of 20 individual episodes or events, with each lasting on average
about 30 minutes, though there is a lot of variety [
18
,
31
]. Prior
work on episode segmentation analyses sensor streams from
wearable cameras to segment of life-experience into events, post-
capture [
16
], however this poses a problem. The human memory
operates in real-time, so any personal life archive, that is designed
to work in synergy with human memory should also work in
real-time. Hence, we propose that processing should occur in real-
time and that this entails data analysis on both the smartphone,
wearable devices and a cloud-based server, to operate in real-time
and upload events to the personal life archive as they happen.
We have extensively evaluated many such event segmentation
techniques on uploaded data [
16
]. However future challenges
remain in developing real-time episode segmentation on the
devices.
•
Episodes should be semantically described: To support both post-
hoc review and real-time analysis of episodes, both server-side
and device-based semantic analysis tools are needed. These act
as software sensors to enrich the raw sensor streams with se-
mantically meaningful annotations; such software sensors are
multi-layered to allow for additional derivations to be minded
from existing sensor outputs the personal life archives. For ex-
ample, raw accelerometer values on a smartphone can identify
the physical activities of a user [
3
], bluetooth and GPS sensors
or audio allow us to determine where and with whom people are
with [
9
], while using automatic detection of concepts is possible
from images [
15
]. These user activities combined with and event
ontologies help infer higher-level semantics on the lifestyle of
individuals. We currently utilise the following virtual sensors in
our personal digital archiving system; semantic date/time, mean-
ingful location, personal physical activity, social interactions (via
bluetooth), environmental context (via GPS and crowd-sourcing
of relevant tags [
20
]), semantic visual concepts automatically
identied from the photos and personal context of the user’s
life pattern. Together, these generate a detailed description of
user activities. In addition the relative importance or potential
memorability of each episode can be determined via a combina-
tion of image face detection and bluetooth people recognition,
in conjunction with the relative uniqueness of the situation the
user is in (relative to their normal lifestyle) [19].
•
A simple narrative summary of each episode should be generated:
Individuals desire quick and simple episode summaries that are
easy to understand. For example, Xu explains why such simple
summaries are desired on a cognitive functioning level [
46
],“The

Challenges and Opportunities within Personal Life Archives ICMR ’18, June 11–14, 2018, Yokohama, Japan
mental representation or situation model clearly depends upon in-
teractions between the language system and complementary, extra-
linguistic cognitive processes .. . the situation model is created by
connecting the text with knowledge derived from the reader’s long-
term memory, and involves additional demands upon attention (e.g.,
the ability to shift points of view and parse sequences of events),
working memory (the ability to retain longer term, anaphoric refer-
ences), and the contribution of emotional knowledge, visual imagery,
empathy, and abstraction” [
46
]. Concise narratives are an impor-
tant building block for personal life archives in that they are
shown to produce emotional responses to autobiographical mem-
ories and help support many of the 5 Rs of memory access. This
therefore motivates the needs to summarise the range of sensor
stream semantic annotations into a meaningful and engaging
descriptive narrative. A secondary benet of such narrative gen-
eration is that the textual narratives can also be used to support
keyword text search [39].
3.3 Life Archive Indexing and Annotation
To be able to retrieve life experiences for search or recommen-
dation from personal life archive, either later or in real-time, the
experiences and their annotations need to be indexed. An initial
assumption would be to employ state-of-the-art techniques from
articial intelligence, database search and information retrieval to
scalably index the life-experience events and provide omnipresent
access via keyword/database search, ranking, recommending and
presenting the multimedia rich life experience archive through mul-
timodal interfaces. However, we contend that to better understand
how to develop lifelogging solutions and how to support eective
access to these data, it becomes necessary not simply to view this as
a new form of multimedia retrieval challenge. Rather, it is important
to understand how people will use and access their life archives.
As a starting point, we turn again to the ve R’s of memory access
from Sellen & Whittaker [
42
]. Each of the ve Rs dene a dierent
reason why people want to access their memories, and by inference,
their personal life archives. They provide valuable clues as to how
to develop the organisation, search and presentation elements of
Personal Life archives.
•
Recollecting is concerned with reliving past experiences for vari-
ous reasons. To take an analogy from cognitive science, Recollect-
ing is concerned with accessing episodic memories. Recollecting
will require highly accurate search engines that semantically rank
content and extract just the nugget of information that is most
pertinent to the user and represent this event in as much detail as
possible to as to aid recollection. This will require conventional
information retrieval, coupled with query-specic experience
segmentation (somewhat similar to the Shot Boundary Detec-
tion [
43
] of digital video. This in itself is a motivation for ‘total
capture’.
•
Reminiscing, which is a form of recollecting, is about reliving
past experiences for emotional or sentimental reasons, some-
times alone, often with others. From information Retrieval, it will
require new techniques for narrative generation [
10
], storytelling
[
8
], topic detection and tracking [
1
] and novelty detection [
47
]
from single (and potentially from multiple individual’s archives),
all operating in conjunction with conventional multimedia docu-
ment ranking techniques, as required for Retrieving (below).
•
Retrieving (information), is a more specic form of recollecting
in which we seek to retrieve specic nuggets of information from
the personal life archive. Retrieval will require highly accurate
text, multimedia and sensor data search engines that semanti-
cally rank content and extract just the nugget of information that
is most pertinent to the user. The conventional Information Re-
trieval concept of top N ranked lists does not successfully operate
for personal life archives (unless N = 1); after all, there is marginal
benet in a system that provides a ranked list of locations for
where the car has been left. The query will dene the type of
knowledge that is required; it is unlikely that a whole document
would be expected in response to a query, as is the norm for
WWW search. Rather the personal life archive is more akin to
question answering in Wolfram Alpha than to whole document
retrieval in Google.
•
Reecting, is a form of quantied self analysis over the life archive
data to discover knowledge and insights. It includes information
summarisation from lifelog streams [
37
], event detection [
20
]
and various forms of data analysis to infer and evaluate the
importance of new semantic knowledge [
9
,
15
,
38
,
44
] from the
personal life archive. Typically such data analysis approaches
rely on articial intelligence, machine learning and various forms
of statistical analysis and should proactively recommend new
knowledge, not solely relying on a human information need as
input, due to the fact that reecting from a personal life archive
brings the potential for new knowledge discovery.
•
Remembering Intentions, is a form of planning future activities
which is a life activity that everyone engages in. This assists peo-
ple to remind or prompt them on tasks they would like to do (e.g.
post that letter), or real-time prompts on who they are talking to
(e.g. this is Paul), or giving prompts on conversation cues (e.g. last
time here together, you had just come away from seeing the new
Batman lm). Past lifelogging eorts were exclusively focused on
episodic memory as it was always a post-hoc analysis (i.e. con-
strained by technology); however now with real-time technology
available we can now consider situational awareness (and past
history of user) to provide prospective memory prompts.
Taking the ve Rs as a guide, we need to identify how to develop
ecient methods that can eectively provide insights from the life
archive, not simply in response to an explicit user query, but also
in response to real-time contextual cues.
4 POTENTIAL APPLICATIONS
Getting insights from personal live archives will very soon be a
phenomenon available to everyone and exploiting the personal life
archives will positively impact on everyone who uses the technol-
ogy. Through integration with other wearable sensors (for example
Bluetooth on cell phones) new opportunities are presented, namely:
•Easy sharing of natural life experiences:
There has been an
increased prevalence of photo sharing on social networking sites
such as Facebook, Instagram or Flickr. In particular uploads from
mobile devices are becoming ever more common, e.g., the Insta-
gram website has 700 million users at the time of writing who
produce images and videos plus additional metadata. However

ICMR ’18, June 11–14, 2018, Yokohama, Japan D.-T. Dang-Nguyen et al.
until now the user has had to make a conscious decision to cap-
ture every image or photo. This has meant that any experience a
user wants to capture had to be interrupted to take a photograph.
The lifelogging platform of passive life experience capture now
allows users to enjoy their experiences, safe in the knowledge
that the media rich experience can easily be reviewed and shared
after the experience has ended.
•Prospective memory feedback to enhance productivity:
Most
memory research exploiting visual lifelogging has focused on
retrospective episodic memory tasks [
11
]. However as real-time
upload of the visual eld-of-view becomes feasible, we can now
begin to support human prospective memory [
2
]. Given a user’s
prior set of experiences and preferences (historical lifelog data),
and their current situation (gathered via on-board mobile de-
vice and wearable sensors), prospective memory prompts can be
provided.
•Personalised wellness feedback:
Poor diet and physical activ-
ity lifestyle choices are strongly associated with the early death
of millions of people [
45
]. Passive capture lifelogging devices
oer the potential to automatically identify episodes of physical
activity [
32
] and diet [
40
]. Through learning the extent of the
user’s current physical activity and diet levels, in addition to
setting some targets, an individual can be automatically relayed
prompts to help them make the healthy lifestyle choice.
•A greater knowledge of self:
Coupled with the concept of
personal wellness, the era of personal life archives will provide
information to the individual about their own life activities and
performance; information that otherwise would go unnoticed.
One can identify trends and pattern in lifestyle and wellness over
an extended period of time.
•A memory assistant:
Neurodegerative disease aects a large
proportion of an aging population and maintaining a personal life
archive has been shown in initial small-scale studies to help oset
some of the debilitating eects of memory impairment [
28
]. Even
for individuals with fully functional memories, the ability to refer
back to the personal life archive will allow for disambiguation of
faded memories and more accurate recall of the past, when such
accuracy is needed.
•An opportunity to better understand the functioning of
the human memory system:
Given that self-reporting is no-
toriously prone to error [
32
], visual lifelogs oer the opportunity
to verify the contextual details of episodic memories recalled by
individuals and can play an important part of modern studies
into human memory.
•An opportunity to better understand the associations be-
tween lifestyle, environmental context and mortality:
Un-
derstanding the determinants and barriers to physical activity
behaviours is important in designing interventions to positively
change these behaviours [
41
]. Accurate measurement of the
type and context of physical activity episodes is therefore impor-
tant [
6
]. Examples of important context attributes of an episode of
physical activity include: whether it occurs indoors or outdoors;
the time of day it occurs; if it is alone or in companionship; and
its domain (home, occupational, etc.). Currently, some of these
attributes are subjectively measured via self-reporting, but for
interventions to be successful, accurate measurement of existing
behaviour on what people are doing and when, as well as under
what conditions, is critical. Automatic lifelogging on a cell phone
oers an opportunity to more accurately measure associations
between lifestyle, environmental context, and mortality.
The potential for personal life archives is enormous. We do ac-
knowledge that there are challenges to be overcome, such as privacy
concerns, data storage, security of data, and the development of a
new generation of search and organisation tools, but we believe
that these will be overcome and that we are on the cusp of a positive
turning point for society; the era of the quantied individual who
knows more about the self than ever before, has more knowledge
to improve the quality of their own life and can share life events
and experiences in rich detail with friends and contacts.
5 INITIAL IMPLEMENTATION
This section presents our rst implementations to tackle the chal-
lenges mentioned earlier as well as to inspire and motivate re-
searchers in the multimedia community to use their know-how in
this new emerging and societal important area.
5.1 Building Personal Archive Datasets
Researchers usually need data to evaluate their methods, and there
is no exception for researchers in personal life archive organisa-
tion and retrieval. To support such research eorts, we gathered
large volumes of lifelog data from several volunteer lifeloggers and
organized, annotated and published them for researchers as the
two lifelogging tasks in NTCIR 12 - Lifelog [
25
,
26
] and NTCIR 13
- Lifelog 2
2
. To the best of our knowledge, our collections are the
largest (in terms of number of days and the size of the collection)
and richest (in terms of types of information) datasets on personal
life archives. These datasets are summarised in Table 1.
Moreover, over the last decade, we pointed out the challenges
for building a shared personal life archive dataset [
23
], proposed
principles, built and described the whole processes from data gath-
ering to determining the roles for the people who are building,
sharing and exploiting such kind of data [
14
]. These principles can
be considered as references for systems that collect personal life
archive.
Table 1: Statistics of the Datasets.
NTCIR-12 NTCIR-13
Number of Lifeloggers 3 2
Number of days 87 90
Size of the Collection (GB) 18.18 26.6
Size of the Collection (Images) 88,124 114,547
Size of the Collection (Locations) 130 138
2http://ntcir-lifelog.computing.dcu.ie/

Challenges and Opportunities within Personal Life Archives ICMR ’18, June 11–14, 2018, Yokohama, Japan
5.2 Benchmarking Initiatives and Workshops
on Life Archive Analytics
As the rst step of building the community working on personal
life archives, we increasingly organize related workshops and pan-
els: iConf 2016
3
, Lifelogging Tools and Applications in ACM MM
2016 [24] and ACM MM 20174.
Together with these, we organize rigorous comparative bench-
marking initiatives: NTCIR 12 - Lifelog [
25
,
26
], NTCIR 13 - Lifelog 2,
and the LifeLog task [
13
] at ImageCLEF 2017 [
29
], which aim to
bring the attention of personal live archive analytics to a wide au-
dience and to motivate research into some of the key challenges of
the eld.
Table 2: Statistics of the Benchmarking Campaigns.
NTCIR-12 NTCIR-13 ImageCLEF
(2016) (2017) (2017)
Number of Tasks 2 4 2
Number of Topics 58 59 51
Number of Submissions 14 21 18
Typically, for each benchmarking initiative, together with the
dataset, we introduced several tasks which aims at advancing the
state-of-the-art research in lifelogging as an application of informa-
tion retrieval. For example, in ImageCLEFlifelog 2017 edition[
13
],
we introduce two tasks: Lifelog Retrieval Task (LRT) and Lifelog
Summarisation Task (LS). In LRT, the participants had to analyse
the lifelog data and for several specic queries, return the correct
answers, for example "In a Meeting: Find the moment(s) in which
the user was in a meeting at work with 2 or more people". In LST,
the participants had to analyse all the images and summarize them
according to specic requirements. For instance: ”Shopping: Summa-
rize the moment(s) in which user doing shopping. To be relevant, the
user must clearly be inside a supermarket or shopping stores (includes
book store, convenient store, pharmacy, etc). Passing by or otherwise
seeing a supermarket are not considered relevant if the user does not
enter the shop to go shopping. Blurred or out of focus images are not
relevant. Images that are covered (mostly by the lifelogger’s arm) are
not relevant.”
Table 2 summarises our evaluation benchmarking campaigns.
According to the results, we can conrm that the proposed dataset
is enough for the proposed topics, in which the answers can be
achieved by exploiting the provided multimodal data. For more
details of these campaigns, please see in [13, 25, 26].
5.3 A Base-line Search Engine for Life Archives
Another means of supporting the research community, besides con-
structing datasets, organizing benchmarking initiatives and work-
shops, is building a baseline search engine for personal life archives.
This baseline engine serves as a basic retrieval system where the
query is made up from basic information needs, where each of them
is asking for a single piece of information, which are already ex-
tracted and indexed. Queries were submitted that generate ranked
lists based on faceted queries, by userID, location visual concept
3http://irlld2016.computing.dcu.ie/index.html
4http://lta2017.computing.dcu.ie
Raw Data
Feature Vectors
Indexed Database
API/Interface
User
baseline search engine
collect data
interact
Figure 1: The baseline search engine architecture
and/or physical activity. A preliminary version of this search engine
can be obtained via:
http://search-lifelog.computing.dcu.ie/
,
which provides a basic retrieval model for the lifelog datasets (other
details of the data used in this baseline search engine are sum-
marised in Table 3).
Table 3: Statistics of the Data Used in the Baseline Search
Engine.
Number of Lifeloggers 1
Number of Days 90 days
Size of the Collection 12.2 GB
Number of Images 51,209 images
Number of Locations 24 locations
Figure 1 summarises how we designed the baseline search en-
gine system [
48
], as follows: from the raw lifelog data, we extract
locations, visual concepts, time, and activities and transformed this
raw data into indexable feature vectors. These feature vectors are
then indexed and hierarchically organized. Finally, a user or other
system can dene a faceted query and retrieve ranked moments
via the interface.
5.3.1 Data Organisation and Retrieval Process. To organise and
index the data, we arrange features as chronological order, and use
the minute as the basic unit of indexing and retrieval. Building up
from these minutes, we organize the data at higher level which can
be turned into more useful information, the minutes are hierarchi-
cally grouped into event nodes (typically, in a full day, a person
encounters anything upwards of 20 individual events, with each
lasting (on average) 30 minutes [
18
,
31
]), then ultimately leading to
larger units such as days and multi-day events (e.g. holidays).
In order to turn a query into specic criteria, i.e., to make the
baseline search engine able to get insights from the personal life
archive, we apply two approaches: rstly, automatic query genera-
tion by considering the terms in the query as identiers of concepts
and then searching for all images that contain those concepts, and
secondly, a ne-tunning method by manually generating a query
string i.e., the researcher (we) will read the topic and “translate" it
into the search criteria. For example, with the query:

ICMR ’18, June 11–14, 2018, Yokohama, Japan D.-T. Dang-Nguyen et al.
(a) 20160818_Taking a train (b) 20160827_Taking an airplane (c) 20160829_Taking a bus
(d) 20160830_Taking a taxi (e) 20160906_Taking a coach (e) 20160906_Taking a bus
Figure 2: Examples of the results retrieved by the proposed baseline search engine for the query “Find the moment when I
was taking public transportation or taxi at sunset."
"Find the moments when user u1 was using public
transportation or taxi at sunset."
We follow the study methodology proposed in [
49
] and “trans-
late" that query into specic required pieces of information, as
follows:
•User = {u1},
•Concepts = {sunset},
•Activity = {transport, airplane},
•time = {16:00-21:00}
•Location = {n/A}
5.3.2 Ranking. To rene the results, i.e., to increase the preci-
sion of the top retrieved images, we use a hierarchical agglomerative
clustering algorithm (see [
12
]) to group similar images into the same
cluster based on all of their features. The clusters are then sorted
based on the number of images, in decreasing rank order. Finally,
we produce the ranked list by selecting representative images from
the clusters by choosing the images closest to the center of each
cluster.
5.3.3 Results. We applied the baseline search engine to the
benchmarking campaigns mentioned in Section 5.2, serving as a
baseline for the comparison. For some topics, the scores were 1
.
00
and 0
.
92 for precision and recall, which shows the potential of
this baseline search engine. Consequently, this engine was added
as a part of the campaigns, allowing participant to obtain base-
line results and develop more complex retrieval systems on-top of
this baseline search engine. Some other examples can be seen in
Figure 2.
6 CONCLUSIONS
In this paper we presented a set of possible challenges and oppor-
tunities from personal life archives. We identied methods and
technologies that can aid users to get insights in their data, from
public resources to their social connections and closer intimacy,
and moreover, from their personal life archive, including dier-
ent granularities of information from their past experiences. We
presented the current state of the eld and provided a number
of potential benets. Furthermore, we identied and proposed so-
lutions to the challenges that arise with such data. This will be
increasingly important over the coming years as we learn more
about ourselves and have access to technologies that will help us
in many aspects of everyday life. We also showed that there are
several research challenges to address in the elds of Articial Intel-
ligence, Cognitive Science and Information Retrieval. Starting with
security and privacy, human computer interaction, and memory
science, valuable insights into how to address these challenges can
be obtained. Finally we presented a baseline search enginge that
achieves reasonable results that can be useful for other researchers
interested in the eld.

Challenges and Opportunities within Personal Life Archives ICMR ’18, June 11–14, 2018, Yokohama, Japan
REFERENCES
[1]
James Allan. 2002. Topic detection and tracking. Kluwer Academic Publishers,
Norwell, MA, USA, Chapter Introduction to topic detection and tracking, 1–16.
[2] Alan Baddeley (Ed.). 2004. Your Memory: A User’s Guide. Carlton Books.
[3]
Ling Bao and Stephen Intille. 2004. Activity Recognition from User-Annotated Ac-
celeration Data. In Pervasive Computing, Alois Ferscha and Friedemann Mattern
(Eds.). Lecture Notes in Computer Science, Vol. 3001. Springer Berlin / Heidelberg,
1–17.
[4] Gordon Bell and Jim Gemmell. 2007. A Digital Life. Scientic American (2007).
[5]
Gordon Bell and Jim Gemmell (Eds.). 2009. Total Recall: How the E-Memory
Revolution Will Change Everything. Penguin Books.
[6]
N. W. Burton, A. Khan, and W. J. Brown. 2012. How, where and with whom?
Physical activity context preferences of three adult groups at risk of inactivity.
Br J Sports Med (Jan 2012).
[7]
Vannevar Bush. 1945. As We May Think. The Atlantic Monthly 176, 1 (Jul 1945),
101–108.
[8]
Daragh Byrne, Aisling Kelliher, and Gareth J. F. Jones. 2011. Life editing: third-
party perspectives on lifelog content.. In CHI, Desney S. Tan, Saleema Amershi,
Bo Begole, Wendy A. Kellogg, and Manas Tungare (Eds.). ACM, 1501–1510.
[9]
Daragh Byrne, Barry Lavelle, Aiden Roger Doherty, Gareth J.F. Jones, and Alan F.
Smeaton. 2007. Using Bluetooth and GPS Metadata to Measure Event Similarity
in SenseCam Images. In IMAI’07 - 5th International Conference on Intelligent
Multimedia and Ambient Intelligence. 1454–1460.
[10]
Charles B. Callaway and James C. Lester. 2002. Narrative prose generation. Artif.
Intell. 139, 2 (Aug. 2002), 213–252.
[11]
Martin A. Conway and Catherine Loveday. 2011. SenseCam: The Future of
Everyday Memory Research. Memory 19, 7 (Oct 2011), 685–807.
[12]
Duc-Tien Dang-Nguyen, Luca Piras, Giorgio Giacinto, Giulia Boato, and
Francesco GB DE Natale. 2017. Multimodal retrieval with diversication and
relevance feedback for tourist attraction images. ACM Transactions on Multimedia
Computing, Communications, and Applications (TOMM) 13, 4 (2017), 49.
[13]
Duc-Tien Dang-Nguyen, Luca Piras, Michael Riegler, Giulia Boato, Liting Zhou,
and Cathal Gurrin. 2017. Overview of ImageCLEFlifelog 2017: Lifelog Retrieval
and Summarization. In CLEF2017 Working Notes (CEUR Workshop Proceedings).
CEUR-WS.org <http://ceur-ws.org>, Dublin, Ireland.
[14]
Duc-Tien Dang-Nguyen, Liting Zhou, Rashmi Gupta, Michael Riegler, and Cathal
Gurrin. 2017. Building a Disclosed Lifelog Dataset: Challenges, Principles and Pro-
cesses. In Content-Based Multimedia Indexing (CBMI).
[15]
Aiden Roger Doherty, Niamh Caprani, Ciaran O Conaire, Vaiva Kalnikaite, Cathal
Gurrin, Alan F. Smeaton, and Noel E. O Connor. 2011. Passively recognising
human activities through lifelogging. Computers in Human Behavior 27, 5 (2011),
1948–1958.
[16]
Aiden Roger Doherty, Chris J.A. Moulin, and Alan F. Smeaton. 2011. Automat-
ically assisting human memory: A SenseCam browser. Memory 7, 19 (2011),
785–795.
[17] Aiden Roger Doherty, K Pauly-Takacs, N Caprani, Cathal Gurrin, C J A Moulin,
N E OÊijConnor, and Alan F. Smeaton. 2012. Experiences of Aiding Autobio-
graphical Memory Using the SenseCam. Human-Computer Interaction 27, 1-2
(2012), 151–174.
[18]
Aiden Roger Doherty and Alan F. Smeaton. 2008. Automatically segmenting
lifelog data into events. In Ninth International Workshop on Image Analysis for
Multimedia Interactive Services. IEEE, 20–23.
[19]
Aiden Roger Doherty and Alan F. Smeaton. 2008. Combining Face Detection
and Novelty to Identify Important Events in a Visual Lifelog. In CIT: 8th In-
ternational Conference on Computer and Information Technology, Workshop on
Image- and Video-based Pattern Analysis and Applications. IEEE Computer Society,
Washington, DC, USA.
[20]
Aiden Roger Doherty and Alan F. Smeaton. 2010. Automatically augmenting
lifelog events using pervasively generated content from millions of people. Sensors
(Peterboroug 10, 3 (2010), 1423–1446.
[21]
Jim Gemmell, Gordon Bell, and Roger Lueder. 2006. MyLifeBits: a personal
database for everything. Commun. ACM 49, 1 (2006), 88–95.
[22]
Cathal Gurrin, Rami Albatal, Hideo Joho, and Kaori Ishii. 2014. A privacy by
design approach to lifelogging. Digital Enlightenment Yearbook 2014 (2014),
49–73.
[23]
Cathal Gurrin, Daragh Byrne, Noel O’Connor, Gareth JF Jones, and Alan F.
Smeaton. 2008. Architecture and challenges of maintaining a large-scale, context-
aware human digital memory. In International Conference on Visual Information
Engineering. 158–163.
[24]
Cathal Gurrin, X. Giró-i Nieto, Petia Radeva, M. Dimiccoli, H. Johansen, H. Joho,
and Vivek K Singh. 2016. LTA 2016 - The First Workshop on Lifelogging Tools
and Applications. (10/2016 2016).
[25]
Cathal Gurrin, Hideo Joho, Frank Hopfgartner, Liting Zhou, and Rami Albatal.
2016. NTCIR Lifelog: The First Test Collection for Lifelog Research. (2016),
705–708.
[26]
Cathal Gurrin, Hideo Joho, Frank Hopfgartner, Liting Zhou, and Rami Albatal.
2016. Overview of NTCIR-12 Lifelog Task. , 354–360 pages.
[27]
Cathal Gurrin, Alan F. Smeaton, and Aiden Roger Doherty. 2014. LifeLogging:
Personal Big Data. Foundations and Trends in Information Retrieval 8, 1 (2014),
1–125.
[28]
Steve Hodges, Emma Berry, and Ken Wood. 2011. SenseCam: A wearable camera
that stimulates and rehabilitates autobiographical memory. Memory 7, 19 (2011),
685–696.
[29]
Bogdan Ionescu, Henning Müller, Mauricio Villegas, Helbert Arenas, Giulia
Boato, Duc-Tien Dang-Nguyen, Yashin Dicente Cid, Carsten Eickho, Alba Gar-
cia Seco de Herrera, Cathal Gurrin, Kovalev Vassili Islam, Bayzidul and, Vitali
Liauchuk, Josiane Mothe, Luca Piras, Michael Riegler, and Immanuel Schwall.
2017. Overview of ImageCLEF 2017: Information extraction from images. In
CLEF 2017 Proceedings. Springer, Dublin, Ireland.
[30]
Ramesh Jain and Pinaki Sinha. 2010. Content without context is meaningless. In
Proceedings of the international conference on Multimedia (MM ’10). ACM, New
York, NY, USA, 1259–1268.
[31]
Daniel Kahneman, Alan B. Krueger, David A. Schkade, Norbert Schwarz, and
Arthur A. Stone. 2004. A Survey Method for Characterizing Daily Life Experience:
The Day Reconstruction Method. Science 306 (2004), 1776–1780.
[32]
Paul Kelly, Aiden Roger Doherty, Emma Berry, Steve Hodges, Alan M. Batterham,
and charlie Foster. 2011. Can we use digital life-log images to investigate active
and sedentary travel behaviour? Results from a pilot study. Intnational Journal
Behavioural Nutrition Physical Activity 8 (2011), 44.
[33]
S. Mann. 1997. Wearable computing: A rst step toward personal imaging.
Computer 30, 2 (1997), 25–32.
[34]
Jochen Meyer, Steven Simske, Katie A. Siek, Cathal G. Gurrin, and Hermie Her-
mens. 2014. Beyond Quantied Self: Data for Wellbeing. In CHI ’14 Extended
Abstracts on Human Factors in Computing Systems (CHI EA ’14). ACM, New York,
NY, USA, 95–98. https://doi.org/10.1145/2559206.2560469
[35]
Theodor H. Nelson. 1988. Managing immense storage. BYTE 13, 1 (Jan. 1988),
225–238.
[36]
David H. Nguyen, Gabriela Marcu, Gillian R. Hayes, Khai N. Truong, James
Scott, Marc Langheinrich, and Christof Roduner. 2009. Encountering SenseCam:
personal recording technologies in everyday life. In Ubicomp ’09: Proceedings of
the 11th international conference on Ubiquitous computing. 165–174.
[37]
Y. Ogras and Hakan Ferhatosmanoglu. 2006. Online summarization of dynamic
time series data. The VLDB Journal 15, 1 (Jan. 2006), 84–98.
[38]
Zhengwei Qiu, Aiden Roger Doherty, Cathal Gurrin, and Alan F. Smeaton. 2011.
Mining user activity as a context source for search and retrieval. In Semantic
Technology and Information Retrieval (STAIR), 2011 International Conference on.
162 –166.
[39]
Z.W. Qiu, Cathal Gurrin, and Alan F. Smeaton. 2016. Evaluating Access Mecha-
nisms for Multimodal Representations of Lifelogs. In International Conference on
Multimedia Modelling. I: 574–585.
[40]
Sasank Reddy, Andrew Parker, Josh Hyman, Je Burke, Deborah Estrin, and
Mark Hansen. 2007. Image browsing, processing, and clustering for participatory
sensing: lessons from a DietSense prototype. In EmNets ’07: Proceedings of the
4th workshop on Embedded networked sensors. ACM, New York, NY, USA, 13–17.
[41]
J. F. Sallis, N. Owen, and M. J. Fotheringham. 2000. Behavioral epidemiology: a
systematic framework to classify phases of research on health promotion and
disease prevention. Ann Behav Med 22 (2000), 294–298.
[42]
Abigail J. Sellen and Steve Whittaker. 2010. Beyond total capture: a constructive
critique of lifelogging. Comm. ACM 53, 5 (2010), 70–77.
[43]
Alan F. Smeaton, Paul Over, and Aiden Roger Doherty. 2010. Video shot bound-
ary detection: Seven years of TRECVid activity. Computer Vision and Image
Understanding 114, 4 (2010), 411 – 418.
[44]
D. Tancharoen, T. Yamasaki, and K. Aizawa. 2006. Practical Life Log Video Index-
ing Based on Content and Context. In Multimedia Content Analysis, Management,
and Retieval  In Proceedings of SPIE-IST Electronic Imaging.
[45]
WHO. 2007. A guide for population based approaches to increasing levels of physical
activity: Implementation of the WHO global strategy on diet, physical activity and
health. World Health Organisation.
[46]
J. Xu, S. Kemeny, G. Park, C. Frattali, and A. Braun. 2005. Language in context:
emergent features of word, sentence, and narrative comprehension. Neuroimage
25, 3 (2005), 1002–1015.
[47]
Le Zhao, Min Zhang, and Shaoping Ma. 2006. The nature of novelty detection.
Inf. Retr. 9, 5 (Nov. 2006), 521–541.
[48]
Liting Zhou, Duc-Tien Dang-Nguyen, and Cathal Gurrin. 2017. A Baseline
Search Engine for Personal Life Archives. In Proceedings of the 2nd Workshop on
Lifelogging Tools and Applications (LTA ’17). 21–24.
[49]
Liting Zhou, Luca Piras, Michael Rieger, Giulia Boato, Duc-Tien Dang-Nguyen,
and Cathal Gurrin. 2017. Organizer Team at ImageCLEFlifelog 2017: Baseline
Approaches for Lifelog Retrieval and Summarization. CLEF working notes, CEUR
(September 11-14 2017).