Nowadays a user's personal information space is fragmented into multiple repositories on their local machine as well as on remote servers. In order to enable later access to resources managed within such a cross-media information space, information has to be organised in a format that can be processed by automatic retrieval processes. We propose a general framework for personal information management based on extending a cross-media link server with supplemental metadata functionality. In addition to user generated information, our solution automatically derives metadata for classifying and associating resources based on direct interaction with the information space. Resources and metadata can be integrated by referencing external resources or information may be managed directly by the framework. The presented cross-media information management solution is not limited to a fixed set of predefined resources and can be extended based on a resource plug-in mechanism.
The pervasiveness of information that we face today leads to the “ineptitude
of getting at the record ” as already predicted in 1945 by Vannevar Bush [
The distribution of resources and metadata across various repositories, such
as sites for sharing and managing multimedia, email clients or native storage
systems, hinders an efficient information retrieval process. The duplicity of
resources and the poor semantic interconnections between information items are
consequences of information fragmentation. Another problem in the retrieval
process is the human tendency to express searches in vague natural language [
An effective solution to satisfy a user’s information needs depends on
providing the appropriate tools for information organisation, management and retrieval
based on an integrated view of different information repositories [
In an attempt to provide solutions for personal information management, we have to address at least two major issues. First, important metadata as well as information about a resource’s original context gets lost when resources are copied between repositories. For example, once an email attachment has been copied to the file system, it loses metadata such as information about the sender, the date the document was received or the email’s topic and content. A second issue deals with the capture of indirect metadata sources that can be derived from an individual’s working behaviour and interaction with the information space. For instance, an event in the news that happened the same day the user created the first version of a document is a cue which might be used as an entry point when trying to recall a specific piece of information.
We propose a solution for cross-media information management based on capturing a user’s semantic traces while interacting with different resources and storing this metadata within their personal cross-media information space based on the dimensions shown in Fig. 1. Classification results from placing resources into various semantic collections, such as named directories, the collection of resources bearing the same tag or resources used within the same website. In this way, we can gather metadata by observing explicit user-driven tagging or classifying actions within existing applications as well as classifying resources automatically based on their original contextual information. The first case is depicted in Fig. 1 as manual classification and the latter by automatic classification. Note that, in our approach, resources may be placed in several semantic collections at the same time (multiple classification). The associations category denotes the correlations resulting from resource usage patterns and scenarios as well as from recording annotations and metadata attached to the resources within their original context. While the manual associations denote explicit user annotations, automatic associations are system generated by gathering and structuring the relevant content.
We postulate that the recall of information can be improved by providing a
“selection by association” based on a flexible and extensible model for managing
both data and metadata within a personal information space. A unified view of
a user’s entire collection of resources is obtained by storing resources in the local
repository or referencing external resources based on a resource proxy approach.
This also includes the capture of information cues derived from interacting with
the information space. Motivated by the argument that it is easier to
remember information cues related to their data rather than structured schemas or
formalisms [
We start in Sect. 2 by describing our data model for cross-media information spaces. The corresponding cross-media information management framework is introduced in Sect. 3. Section 4 provides a comparison of our approach and existing solutions for personal information management. Concluding remarks are given in Sect. 5. 2
A Data Model for Cross-Media Information Spaces
Our model for personal cross-media information spaces is defined using the
semantic, object-oriented data model OM [
The schema of our extended model is shown in Fig. 2. The shaded rectangles denote collections of objects (classification) with the name of the collection in the unshaded part and the name of the associated type in the shaded part. The shapes with a solid outline are part of the original RSL model, whereas shapes with a dashed outline form part of the model extension. In the following we describe how the RSL concepts have been used to realise the ideas presented in the previous section and present the required extensions.
The central concept within the RSL model is the generic notion of an entity, the instances of which are classified by the collection Entities. An entity is an abstraction for all possible information items which might be handled in a personal information space. The abstract resource type is used to represent any kind of information resource such as different types of documents, images, videos or even physical resources, but also metadata structures and information cues that are semantically correlated with the information resources. Specific extensions of the resource type have to be implemented to represent particular types of resources. In Fig. 2, we show the extension for information cues with the type cue and its subtypes visualCue and temporalCue. annotation Annotations
Synonym
Links
temporalLink Temporal
Links
link Navigational
Links
Structural
Links (1,*) link HasSource
HasTarget
|HasChild| link (1,*) (0,*) partition Links Entities partition RefersTo link entity (1,*) (0,*) (0,*) (0,*) (0,*)
HasResolver (0,*)
resource Resources
visualCue Visual Cues contextResolver Context Resolvers
cue Cues partition
temporalCue Temporal
Cues parameter Properties (0,*) HasProperties (0,*)
selector Selectors
(1,1)
Often the possibility to address an entire resource is not flexible enough. For example, it may be just the section of a web page mentioning the contact details of a person that raised the user’s attention. The selector entity subtype enables parts of resources to be addressed. As in the case of the resources, selectors may be specialised for concrete types of information item fragments. Note that it is not our aim to store all information resources in a centralised repository, but rather to provide unified access to resources in accordance with their relevance for specific tasks. Some resources, including metadata information, may act as proxy elements to external information units. A flexible classification of resources is supported through the creation of new collections and subcollections at runtime and the multiple classification feature offered by the OM model.
The link concept offered by the RSL model is used to store associations between information resources by connecting several source and target entities. This allows links to be defined between different types of resources, parts of resources or even link entities. To illustrate how information is represented in our extended model, let us have a look at the following scenario: First thing in the morning, Adriana checks her email. An interesting talk organised by the department has been announced. She immediately searches the Web to find more details about the speaker.
We have to create three new resource objects representing the mentioned event, the speaker and the consulted web page. We then create a temporal cue object to represent the current date and system time from which we can later also derive the time of the day (e.g. “in the morning”). The semantic associations between objects are modelled through a temporal link having the temporal cue object as the source and the objects representing the event, the speaker and their personal web page as the targets.
In addition to the new temporal link type, we provide other link extensions represented by the StructuralLinks, SynonymLinks and Annotations collections shown in Fig. 2. The structural links may be used to represent specific resource structures. For example, we could represent the structure of an email as having a sender, one or several receivers, potential attachments, a subject and the content, which again may have a certain structure.
The synonym links are used to connect various representations of what is essentially the same information. For example, we can link together the address book entry, the contact details section of a paper and a web profile referring to the same person, thus obtaining the “synonym ring” of the resource. User annotations are recorded as links between the resource being annotated and the resource representing the content of the annotation. This means that a resource may be annotated not only with textual information but with any other resource, resource fragment or even with other links.
We mentioned earlier that we also collect metadata from a resource’s original context. While it is out of the scope of this paper to describe the metadata import mechanism, note that the same principle of linking entities through specific types of links can be used. Therefore, a set of new link types is created in accordance with the specification of the underlying metadata structure standards. For example, to connect resources to their Dublin Core1 descriptions we use links annotated with resources representing the Dublin Core element set.
Finally, to refine the retrieval process, the model also includes the Properties and the Context Resolvers elements. Whereas additional information such as a link’s weight might be stored as a specific link property, the context resolvers control an entity’s “visibility” that might be based on information derived from specified cues or other contextual information. 3
Cross-Media Information Management Framework The main components of the proposed cross-media information management framework are shown in Fig. 3. An extensible Resource Tracker component observes user interactions with different Data Providers such as the local file system, the mail server or the Web. Specific plug-ins help to improve the granularity by, for example, tracking visited Flickr albums and pictures or sections of PDF documents. Each resource access is detected by the corresponding resource tracker extension and reported to the Information Manager component which checks if the resource has been accessed before and is already available in the personal information space. If necessary, a new resource object is created and classified in different semantic collections based on criteria such as its type or origin. Another process creates resource objects in correspondence to any metadata associated with the resource in its original context. In this way, external metadata is either copied into the local repository or referenced by metadata resource proxies. Finally, the Annotation Service is responsible for recording
Update Metadata Personal Information
Space any user-defined annotations of visited resources. Note that if the original data provider offers some annotation functionality, the framework will rely on it, otherwise specific annotation components have to be implemented. The Information Manager is also responsible for creating specific links between a content resource and its related metadata and annotation resources. In addition, resource access patterns are tracked and stored as a special type of links between succeeding resources. Once the user shifts their work to another resource (e.g. another web page or document), the same mechanism is triggered again.
Context Providers
Data Providers Specific Plug-ins
Annotation Add-Ons Cue Tracker
Resource Tracker Information Manager
Annotation Service Store Cues
Create Links
Store Annotations
The Cue Tracker component monitors various Context Providers. While Fig. 3 shows calendar applications, the date and system time information, email clients as well as other context providers, new sources of information cues may be added based on a plug-in mechanism. In accordance with devised policies, the Cue Tracker reacts to specific events within the tracked sources. For example, temporal cues are constructed based on time intervals marking off the duration of a conversation, the morning of a certain day or a news event possibly spanning several days. Once cues are detected, the Cue Tracker contacts the Information Manager which creates specific cue objects within the personal information space. Cues are maintained in a collection of active cues as long as their extent has not yet expired. As new content resources are detected, these are connected through specific links with all active cues. Since our system supports links with multiple sources and multiple targets, each resource is added as a new target of a link having the cue as source. Access counters attached to links monitor how often a link has been visited either in relation to a search or to connect more resources. Co-occurrences of cues within the active cues collection may entail further semantic information that we plan to investigate.
Retrieving resources may be performed by simply following links between resources. A user may specify unstructured details on specific topics, such as information cues that they remember. Once the objects corresponding to a user’s query are found, trails of links may be constructed and connected resources may be returned.
Several other projects have recognised the need for a uniform information space
and the importance of semantic metadata for the retrieval of resources. Earlier
proposals have indicated the necessity of paradigms for information space
organisation. While these approaches followed just one or a few dimensions of the
information space, organising resources in terms of time [
Related approaches tend to use relational databases [
The MyLifeBits [
The Social Semantic Desktop [
We have presented a cross-media information management framework for managing a user’s personal information. An essential feature of our proposed solution is the flexibility and extensibility in terms of both resource representations as well as the capture of metadata and information cues. Metadata and information cues are constituent parts of the information space together with the information resources, supporting the information retrieval process through a “selection by association” paradigm. Last but not least, the presented approach is flexible in storing resources either locally or integrating external resources based on a resource proxy mechanism.