Everybody experiences it: our personal information from mails, documents, chats and browsing is scattered over many applications, formats, and devices, which makes it difficult to access it again some time after it has been seen for the first time. The challenge of the Memorizer project is to allow users to re-find a piece of data that they have seen before but for which they cannot remember where it is. This objective can be achieved by building a Personal Web of Data, a semantic layer that describes, aggregates, interlinks and stores metadata about information that users have accessed before. Later on, this web of data can then be exploited using smart semantic browsing. In this PhD symposium paper we consider Memorizer's grand challenge in refinding and the approach to achieve this in the period ahead.
1. Introduction
“I remember I have seen this: where is it?” This question is asked everyday by
millions of people that work in a digital information environment [
Many important related issues have been tackled in the Personal Information
Management (PIM) research area [
In spite of several projects that demonstrate the benefits of bringing together
semantic technology and PIM, such as Nepomuk [
In this PhD work, we are concerned with the use of Semantic Web in the context of re-finding digital information. In particular, we focus on the problem of data access and show that the elicitation, combination and interlinking of metadata using the semantic approach can be applied to enrich and optimize re-finding of information.
For instance, suppose that Ana is designing a web site together with Daniel. Daniel finds a reference that may be interesting to Ana: the Scriptaculous API, a Javascript API for building animations. So, he sends an e-mail recommending her to have a look on the Scriptaculous’ web site. As Ana has already decided which technology to use in the project she does not visit the web site. One year later, Ana is looking for a Javascript API for building pop-up menus on web pages. She has visited many pages but she cannot find any solutions suitable to her needs. Fortunately, she remembers that Daniel has mentioned an API (the Scriptaculous) that can be useful, however, she does not remember the name of the API, or the URI or when Daniel gave her such information. In this scenario, the system can use the user mental clues during the re-finding task by eliciting and connecting information between the user history and email messages. For instance, Ana can search for messages sent by Daniel, but as the system knows that she has just visited few pages concerning Javascript APIs, the system can filter messages sent by Daniel that also are related to Javascript concept. As result, Ana will be able to find the Scriptaculous URI by providing only her mental clues that she is able to remember, instead of wasting her time by browsing the entire mailbox or performing many queries on search engines.
Due to the nature of this PhD symposium paper, an exhaustive overview on related work cannot be given here. Instead, we focus on the identification of some specific problems that allow us to help clarify the need for a semantics-based approach.
Information Fragmentation
The process of re-finding personal data is closely related to the process of how
the personal data was collected and stored. Unfortunately, the tools that users have
today, lead them to spread their data among many applications and formats, therefore
decreasing and/or breaking the data connectivity. For instance, once a user saves an
email attachment to her file system, the connection between the file and message is
lost. Consequently, this demands a high mental load for re-finding the knowledge
about who sent that document. This lack of data connectivity is known in the
literature as information fragmentation [
To solve this problem some authors [
Information Extraction
An important part of the problem concerns the techniques to elicit relevant
metadata about the information that it is already there, e.g. file system, email, web
history, etc. Information extraction in connection to the Semantic Web can be used to
help in structuring the users’ data that is likely wrongly (less optimally) structured for
the purpose of re-finding, consequently allowing to form an integrated and universal
medium of information that (expectedly) improves the possibility to re-find data [
Little attention has being given until now to the process of re-finding information. The previous section indicates that there is still a lot of work to do regarding this challenge. In this PhD research we have started work on an approach centered around the notion of elicitation of metadata about information that user accesses, organizing it and exploiting it, in order to support re-finding tasks. In particular, we are concerned with the following research questions.
2. How do we query the integrated meta-data for supporting re-finding tasks? To outline a realistic Ph.D. proposal, we will focus on re-finding of email and web history. Our first approach for that problem is to elicit meta-data about the user’s web history and email archive. For that purpose, we are going to use available tools on the literature that can be used for extracting information from the user’s email and web history. Also, we are going to express and consolidate this meta-data in a uniform semantic layer, expressed in RDF model, which serves as a first step towards organizing the entities for re-finding. We are going to investigate the process of integrating data, and how this meta-data can be automatically interconnected. We will also investigate how to detect the user context based on her web navigation, and how to integrate such information with other elicited meta-data during the re-finding process.
All this information will be expressed in a novel ontology-independent approach. Our intention is to evaluate an approach where different terminologies can co-exist but where the system exposes to the user only a joint view of concepts, properties and classes that are similar but they were expressed in different terminologies. With this approach, the system can learn the user’s mental model and adapt its terminology to it, instead of forcing the user to adapt to the system terminology.
Since there might be information that can be a crucial mental clue for re-finding but that can not be directly obtained from the user’s personal data information, we are going to use external sources of data as WordNet1, DBPedia2 and other datasets on the Linked Data to enrich the semantic layer.
Once we have organized the semantic layer, we will expose it to users using different exploration mechanisms, such as browsing, facet navigation and keywordsearch, in order to identify the best retrieval mechanism to perform and support refinding tasks. We will evaluate the precision and recall of each retrieval mechanism.
We motivated our work with general considerations about the use of the semantic web in the process of personal information management. While there are many open problems, we focused on a specific family of interrelated problem that are centered around the notion of re-finding information using metadata about the information that they have been seen before. As a result, we expect to bring a novel architecture to represent the user’s personal data that it is closer to the mental model that the user has about her environment, therefore improving how the information can be re-found. Nevertheless, we are still in early stage of development; we expect soon the delivery of the first tools that will be part of the whole re-finding system called Memorizer.